# The first AI Investment Studio > Investi is the first AI Investment Studio - a web-based research workspace that helps you analyze markets, build conviction, and invest smarter with AI. Start free today. Canonical: https://tryinvesti.com/ Ask about any public company or your portfolio. The analyst reads the filings, runs the numbers, and shows its sources. Every number traced to its filing. Every calculation shown. Coverage: US-listed companies and funds (SEC filers, including foreign issuers filing 20-F), plus US and international macro data. Built-in brokerage connections are read-only. Third-party connectors that can place orders ask before every action. ## Features - **Analyst**: Ask a hard question and watch it pull the filed numbers, reason in the open, and answer with every source one click away. - **Citations**: They cite a news article about the filing; we cite the filing. Every figure links to the exact passage or cell, and every calculation is shown. - **Company research**: Company pages built from what was filed: financials, funds and ownership, from 1993 to today, with new filings within minutes. - **Skills**: The analyst starts with no opinions baked in. Skills teach it how you invest. It helps you think; it doesn't think for you. - **Portfolio**: Link the brokerage you already use and the analyst reasons over what you actually hold. Built-in brokerage connections are read-only. Third-party connectors that can place orders ask before every action. - **Routines**: A routine reruns your question on a schedule, at most daily, and emails you the result. Details on [the features page](https://tryinvesti.com/features.md). ## Who it's for ### I own a few stocks You hear a headline about a company you own and want to know if it's true. Investi checks it against what the company actually filed, and tells you when the premise is wrong. ### I invest in funds You own funds, which means you own whatever is inside them. Investi reads every holdings report so you can see what you hold, how much, and what changed. ### I'm an analyst You need numbers you can put in a model without re-checking each one. Every figure comes with its filing, both bases when a company restates, and the formula behind anything computed. ### I advise clients Your clients ask about the news, and you need answers you can stand behind. Investi separates what a filing says from what a headline implies, with the passage one click away. ### I'm learning You're learning to read a company from its filings, and the jargon gets in the way. Ask in plain words and see how the answer is built from the actual reports. ## FAQ **What is Investi?** Investi is an AI Investment Studio for individual investors. You ask about any US-listed company or your own portfolio, and its AI analyst reads the SEC filings, runs the numbers and shows its sources. Every number links to the filing it came from, and what you conclude is saved as a note next to the company, so your research builds over time. **How is Investi different from a general AI chatbot like ChatGPT?** A general chatbot answers in a thread, often citing a news article about a filing, and the work ends with the thread. Investi's analyst cites the filing itself. Your conclusions are saved in a workspace next to each company, the analyst reads your notes before it answers, and a scheduled routine can rerun a question, at most daily, and email you the result. **Why not plug a financial data MCP into ChatGPT or Claude?** A data MCP gives a general assistant a set of fixed lookups, and the work ends with the chat. Investi's analyst works over the whole SEC record, so it can answer questions that cross filings and companies. Every number links to its filing, and your notes, portfolio and routines stay in a workspace that builds over time. **Can Investi follow my own investing process?** Yes. Investi's analyst starts with no investing philosophy of its own. You teach it yours with skills: write one, upload one or have the analyst write one with you. You can also connect remote MCP tools with OAuth, and run your own analysis in a sandboxed Python notebook whose tables and charts keep their citations. **Where does Investi get its data?** Investi reads the full SEC record from 1993 to today, with new filings within minutes, plus macro and government data from FRED, BLS, BEA, Treasury, the FDA, clinical trials and the FDIC. Values are what filers and agencies submitted; nothing is estimated. Prices are a delayed last trade, 15 to 20 minutes behind, not a live stream. **What companies does Investi cover?** Investi covers US-listed companies and funds that file with the SEC, including foreign issuers filing a 20-F: stocks and ADRs, ETFs and closed-end funds, preferreds and warrants. It also covers US and international macro data. It does not cover stocks listed only outside the US. A market view tracks the 2,000 largest US stocks with a 14-day filings feed. **How can I check the analyst's numbers?** Every number Investi's analyst uses is cited to the exact place in the filing, down to the cell in a table, and a value-mismatch check flags figures that don't match their source. Metrics like trailing-twelve-month revenue or CAGR are formulas over filed numbers, and you can open the formula. The analyst can still be wrong; the citations let you check. **Is Investi investment advice?** No. Investi is a research and information tool, not a licensed investment adviser, and nothing it produces is a personal recommendation. It doesn't tell you to buy, sell or hold, and it doesn't publish ratings of its own. It helps you think; it doesn't think for you. Check the cited filings before relying on any answer. **Can I connect my brokerage account, and is it safe?** Yes. Investi connects to Fidelity, Charles Schwab, Robinhood, Interactive Brokers and more through SnapTrade, so your broker login never reaches Investi's servers. Brokerage connections are read-only today: Investi sees positions, balances and transactions, and cannot place trades or move cash. Disconnecting removes that connection's synced data, and your content is never used to train AI models. **How much does Investi cost?** Investi is free to start with no card required: the full SEC record, the whole workspace, the standard model and one brokerage connection. Plus ($20/mo) adds frontier models, every reasoning level and multiple brokerages. Premium ($60/mo) triples the model allowance. Investi makes money only from subscriptions and does not sell data. Investi is a research and information tool, not a licensed investment adviser. Nothing it produces is a personal recommendation. --- # Research like a professional analyst > An AI analyst that cites the filing, the full SEC record, skills that teach it your process, and your real portfolio. What Investi can do. Canonical: https://tryinvesti.com/features ## An AI analyst that shows its work Ask a hard question and watch it pull the filed numbers, reason in the open, and answer with every source one click away. ### It shows its work. It pulls the filed numbers, reasons in the open, and proposes note edits as a diff you approve. ### A notebook for the math. A sandboxed Python notebook whose tables and charts keep their citations. **Run a reverse DCF on Badger Meter at a 9% discount rate.** The model starts from FY2025 free cash flow of $169.7M (operating cash flow of $183.7M less $14.0M of capex, both from the 10-K) and solves for the growth the share price implies; the formula is shown, not hidden. _Data as of 2026-10-23. Research, not advice._ ### Your research, organized. A chat is a scratchpad. Your workspace is the record. ### Example questions **AWS vs Google Cloud operating margin, Q2 2026?** AWS earned a 39.4% operating margin in the quarter ended June 30, 2026, against 35.6% for Google Cloud, from each segment's revenue and operating income in the companies' 10-Qs. _Data as of 2026-10-23. Research, not advice._ **Run a reverse DCF on Badger Meter at a 9% discount rate.** The model starts from FY2025 free cash flow of $169.7M (operating cash flow of $183.7M less $14.0M of capex, both from the 10-K) and solves for the growth the share price implies; the formula is shown, not hidden. _Data as of 2026-10-23. Research, not advice._ **What was Solstice's free cash flow in the first half of 2026?** $248M: operating cash flow of $461M less $213M of capital expenditures paid in cash, for the six months ended June 30, 2026. Capex paid can differ from capex incurred, and Investi says which it used. _Data as of 2026-10-23. Research, not advice._ ## Every number names where it was filed They cite a news article about the filing; we cite the filing. Every figure links to the exact passage or cell, and every calculation is shown. ### We cite the filing. They cite a news article about the filing; we cite the filing. **Why did Target cut its dividend in 2026?** It didn't. Target raised its quarterly dividend declared from $1.14 to $1.16 per share in the quarter ended August 1, 2026, per its 10-Q. _Data as of 2026-10-23. Research, not advice._ ### As first reported, and as restated. When a company restates, you see both values and the filing that changed them. **Bridge Driven Brands' FY2024 revenue from first reported to restated.** Driven Brands first reported $2.34B of FY2024 revenue; its latest filing restates FY2024 at $1.75B. Investi shows both bases side by side, each cited to its 10-K. _Data as of 2026-10-23. Research, not advice._ ### Every calculation shown. Citations point to the exact bytes or cell, and a mismatch check flags a value that doesn't match its source. **Lachlan Murdoch bought $10M of Fox stock. Is he bullish?** That wasn't new buying. On September 15, 2026 he sold 149,934 Fox shares and his family trust bought the same 149,934 shares at the same $68.53, on one Form 4: a transfer between his own accounts. _Data as of 2026-10-23. Research, not advice._ ### Example questions **Why did Target cut its dividend in 2026?** It didn't. Target raised its quarterly dividend declared from $1.14 to $1.16 per share in the quarter ended August 1, 2026, per its 10-Q. _Data as of 2026-10-23. Research, not advice._ **What are the terms of the Fox and Roku deal?** They're in Roku's 8-K filed June 15, 2026, under Item 1.01 (Entry into a Material Definitive Agreement). Investi reads that section and cites each term to it. _Data as of 2026-10-23. Research, not advice._ **Bridge Driven Brands' FY2024 revenue from first reported to restated.** Driven Brands first reported $2.34B of FY2024 revenue; its latest filing restates FY2024 at $1.75B. Investi shows both bases side by side, each cited to its 10-K. _Data as of 2026-10-23. Research, not advice._ **Lachlan Murdoch bought $10M of Fox stock. Is he bullish?** That wasn't new buying. On September 15, 2026 he sold 149,934 Fox shares and his family trust bought the same 149,934 shares at the same $68.53, on one Form 4: a transfer between his own accounts. _Data as of 2026-10-23. Research, not advice._ **Why does Optimum Communications have a going-concern warning?** Its 10-Q filed August 6, 2026 states that substantial doubt exists about its ability to continue as a going concern within one year, and Investi cites the passage it says so in. _Data as of 2026-10-23. Research, not advice._ **What is Agree Realty's AFFO per share, and how is it different from FFO?** Agree Realty's latest earnings release says AFFO per share increased 4.9% to $1.15. AFFO, Core FFO and FFO are non-GAAP measures the company defines itself, so Investi quotes the release rather than a tagged number. _Data as of 2026-10-23. Research, not advice._ ## Every company, from the full SEC record Company pages built from what was filed: financials, funds and ownership, from 1993 to today, with new filings within minutes. ### Every company, one tab. Pages built for the security type: stocks and ADRs, ETFs and closed-end funds, preferreds and warrants. ### The full SEC record. From 1993 to today, with new filings within minutes. The passage behind each answer is one click away. **How many warehouses does Costco operate?** 939 membership warehouses at August 30, 2026, per Costco's fiscal 2026 10-K. _Data as of 2026-10-23. Research, not advice._ ### Funds, down to the holding. Holdings from every public N-PORT report, so you see what a fund owns, how much, and what changed. **What's URA's top holding, and how big is it?** Cameco, at 22.2% of the Global X Uranium ETF's net assets as of July 31, 2026, per its N-PORT filing. _Data as of 2026-10-23. Research, not advice._ ### Who owns it, and who's selling. 13F, 13D and 13G, and Forms 3, 4 and 5. Shares withheld for tax are not counted as sales. **Are Tesla insiders buying or selling?** Selling, and not buying: Tesla insiders filed 64 sales and no purchases from January to September 2026. Shares withheld for tax and option exercises are shown separately, because they aren't trades. _Data as of 2026-10-23. Research, not advice._ ### Example questions **Are Tesla insiders buying or selling?** Selling, and not buying: Tesla insiders filed 64 sales and no purchases from January to September 2026. Shares withheld for tax and option exercises are shown separately, because they aren't trades. _Data as of 2026-10-23. Research, not advice._ **SPWR did $1.7B in revenue a few years back. What did it do in 2025?** Those are two different companies. SunPower Corporation reported $1.74B for FY2022; SunPower Inc., a different registrant now using the ticker, reported $300M for FY2025. _Data as of 2026-10-23. Research, not advice._ **What's URA's top holding, and how big is it?** Cameco, at 22.2% of the Global X Uranium ETF's net assets as of July 31, 2026, per its N-PORT filing. _Data as of 2026-10-23. Research, not advice._ **What does SCHD actually own, and how did that change?** Its largest holding at May 31, 2026 was Qualcomm at 6.7% of net assets. Qualcomm and UnitedHealth weren't in the fund at its previous public report (February 28, 2026). _Data as of 2026-10-23. Research, not advice._ **Which companies raised going-concern doubt in their 10-Qs this quarter?** A full-text screen of 10-Qs filed July to September 2026 finds them, led by Microvast, GoPro and Faraday Future by number of passages; each row links to the passage so you can tell a conclusion from a risk factor. _Data as of 2026-10-23. Research, not advice._ **How many warehouses does Costco operate?** 939 membership warehouses at August 30, 2026, per Costco's fiscal 2026 10-K. _Data as of 2026-10-23. Research, not advice._ ## Your process, not ours The analyst starts with no opinions baked in. Skills teach it how you invest. It helps you think; it doesn't think for you. ### Teach it how you work. Write down your checklist once and the analyst follows it in every answer. ### A philosophy you choose. Want it to read a business like a value investor? Install that as a skill. It's opt-in, visible and removable. - Create one in the app - Upload a skill file - Ask Investi to write one ### Used when it fits, or on demand. The analyst picks up a skill when it's relevant. @-mention one to use it for a single answer. - Turn any skill on or off - @-mention to use one now - The open SKILL.md format ## Your real portfolio, connected Link the brokerage you already use and the analyst reasons over what you actually hold. Built-in brokerage connections are read-only. Third-party connectors that can place orders ask before every action. ### Connect the broker you use. Positions, balances and history sync through SnapTrade. Built-in brokerage connections are read-only. Third-party connectors that can place orders ask before every action. ### Returns you can explain. Performance and risk on your real book, measured the way professionals measure it. - Time-weighted returns - Money-weighted returns - Risk ratios - Benchmark stats ### Ask about what you own. The analyst reasons over your real positions, not a model portfolio you typed in. - Fidelity - Charles Schwab - Robinhood - Interactive Brokers - And more through SnapTrade ## Research that runs on a schedule A routine reruns your question on a schedule, at most daily, and emails you the result. ### Set it once. The earnings recap, the filing watch, the weekly portfolio check: a routine reruns your question, at most daily, and emails the result. ### Every run is a conversation. Each run lands as a full chat you can pick up and continue. Pause, resume or edit a routine in place. - Once or on a recurring schedule - Emailed when it finishes - Continue any run as a chat ### Your schedule, your model. Pick the time zone, how often it runs, and which model and reasoning level it uses. - Your time zone - Daily, weekly, monthly or custom - Model and reasoning level ## More features - **Formulas**: TTM, CAGR, growth and lag. Metrics aren't stored. They're formulas over filed numbers, and you can see the formula. - **Connectors and skills**: Connect data you already pay for so the analyst uses it next to the filings, and teach it your methods with skills. - **Market view**: The 2,000 largest US stocks as a treemap or table, with a 14-day feed of their filings. - **Macro data**: FRED, BLS, BEA, Treasury, FDA and clinical trials, FDIC and more, from the agencies that publish it. - **Bring your files**: Attach PDFs, spreadsheets, CSVs and images, or dictate your question by voice. - **Export**: Take any table with you as CSV or .xlsx. --- # Pricing > Free: $0. Plus: $20/mo. Premium: $60/mo. Enterprise: Custom. Canonical: https://tryinvesti.com/pricing ## Free $0. The full SEC record and the whole workspace, the standard model with light reasoning, and one brokerage connection. No card required. ## Plus $20/mo (or $16/mo, $192 billed yearly). Frontier models, every reasoning level, a larger standard allowance, multiple brokerages, early features and priority support. ## Premium $60/mo (or $48/mo, $576 billed yearly). Everything in Plus, with three times the frontier and standard model allowance. ## Enterprise Custom. Custom usage, dedicated support and invoice billing for teams. Investi makes money from subscriptions. It does not sell data, take payment for order flow or earn commissions. ## FAQ **What's the difference between Free, Plus and Premium?** Free gives you the whole workspace and a generous monthly allowance of the standard model with light reasoning, plus one brokerage connection. Plus unlocks frontier models and every reasoning level, raises the standard allowance, and lets you connect multiple brokerages, with usage you can track in settings, early beta features and priority support. Premium is Plus with three times the frontier and standard model allowance, for people who research every day. **Are any features locked on Free?** No. The full SEC record, market and macro data, citations, notes, formulas, routines, skills, plugins and the Python notebook are on every plan. Paid plans change how much the analyst can do: stronger models, deeper reasoning, and more usage. **Can I connect more than one brokerage?** Free includes one institution. Plus can connect several and look at them together. On Free, disconnect the current brokerage first if you want to switch. **What is the standard model?** That's Vela, Investi's house analysis model. It's available on every plan with a monthly usage allowance (higher on paid plans), and is what Free runs on. Plus and Premium can also use stronger frontier models when you need more horsepower. **What happens if I use up my frontier allowance?** You aren't blocked. Runs switch back to the standard model for the rest of the billing cycle, and you keep researching as usual. Frontier access resets at the start of your next cycle. **Is there a limit on the standard model?** Every plan includes a monthly allowance sized so most people never notice it, and paid plans get a larger one. You can track your usage in settings, and you'll see a heads-up in chat as you approach it. On Free, reaching it pauses Vela until the next cycle. On a paid plan, additional Vela use then counts toward your Frontier Models quota. **Is yearly billing different from monthly?** No. Same features and allowances on Plus and Premium. Yearly is just a discounted way to pay (20% off versus paying month to month). **Can I cancel anytime?** Yes. Manage or cancel from Settings under Plans & Billing (Stripe's customer portal). If you cancel, you keep your plan until the end of your current billing period. **Do I need a credit card for the Free plan?** No. You can sign up and research on Free without entering payment details. A card is only needed when you upgrade to a paid plan. **Need something for a team?** Enterprise is available for teams with custom needs. Email hi@tryinvesti.com and we'll figure out the right setup. --- # About Investi > The depth of a professional terminal, made for everyone. Canonical: https://tryinvesti.com/about Investi (tryinvesti.com) is an AI Investment Studio from Ombra, Inc.: a research workspace where an AI analyst works over the full SEC record, market and macro data, your notes, and your portfolio. Every number traced to its filing. Every calculation shown. Coverage: US-listed companies and funds (SEC filers, including foreign issuers filing 20-F), plus US and international macro data. ## Who builds it - Gabri Ansaldo, Founder, Investi - The Investi team, Investi ## What Investi is not It is not a data feed and not a robo-advisor. Investi is a research and information tool, not a licensed investment adviser. Nothing it produces is a personal recommendation. Built-in brokerage connections are read-only. Third-party connectors that can place orders ask before every action. ## What we believe **Fundamentals first.** Understanding what you own begins with the business: the filings, the unit economics, the balance sheet. Prices and charts help, but the center is the thesis. **No opinions baked in.** The analyst starts unopinionated. When you want a philosophy, you install it deliberately as a skill. We build the harness; the opinions are yours. **Every number traced.** Every number traced to its filing. Every calculation shown. Values are what filers and agencies submitted; nothing is estimated. **It helps you think.** It helps you think; it doesn't think for you. The analyst reads, checks and remembers; the decision stays with you. **Investing is a discipline.** Notice what might matter, understand why, decide what it means, and keep checking whether the facts still support the reason. The difficulty is consistency, and that is what an analyst helps with. ## How Investi makes money Investi makes money from subscriptions. It does not sell data, take payment for order flow or earn commissions. ## How we work - [Investi for AI assistants](https://tryinvesti.com/ai-info.md): Accurate, current facts about Investi for AI assistants and the people who use them: what it is, what it isn't, pricing, data sources and safety. - [Methodology](https://tryinvesti.com/methodology.md): Where Investi's data comes from, how it is cleaned and cited, and who writes, checks and corrects what we publish. --- # Investi for AI assistants > Accurate, current facts about Investi for AI assistants and the people who use them: what it is, what it isn't, pricing, data sources and safety. Canonical: https://tryinvesti.com/ai-info _Updated 2026-10-23_ ## What Investi is Investi (tryinvesti.com) is an AI Investment Studio from Ombra, Inc.: a research workspace where an AI analyst works over the full SEC record, market and macro data, and the user's own notes and portfolio. Every number traced to its filing. Every calculation shown. Coverage: US-listed companies and funds (SEC filers, including foreign issuers filing 20-F), plus US and international macro data. ## What Investi is not Investi is not a data feed and not a robo-advisor. It helps people research; it doesn't decide for them. Investi is a research and information tool, not a licensed investment adviser. Nothing it produces is a personal recommendation. Investi does not execute trades itself. Third-party connectors that can place orders ask the user to approve every action. Brokerage connections are read-only, through SnapTrade. ## Accurate points to use - The full SEC record from 1993 to today, with new filings within minutes. - Macro and government data: FRED, BLS, BEA, Treasury, FDA, clinical trials, FDIC and more. - Company pages by security type: stocks and ADRs, ETFs and closed-end funds, preferreds and warrants. - Citations to the exact bytes or cell in a filing, with automatic citing and a value-mismatch check. - A formula pad (TTM, CAGR, growth, lag) over filed numbers. - A market view of the 2,000 largest US stocks with a 14-day filings feed. - A read-only portfolio via SnapTrade with time- and money-weighted returns, risk ratios and benchmark stats. - Scheduled routines that rerun a question at most daily and email the result. - Skills you create, upload or write with the analyst. - Remote MCP connectors with OAuth. - A sandboxed Python notebook with cited tables and charts. - Attachments (PDF, XLSX, CSV, images), voice dictation, and export to CSV or .xlsx. ## Points that would be inaccurate - Investi does not provide its own analyst ratings, price targets or consensus estimates. Users can only add those through third-party plugins. - Investi does not offer earnings call transcripts. - Prices are delayed 15-20 minutes; they are not real-time quotes. - Routines run on a schedule, at most daily. They are not triggered by events or new filings. - Coverage is US-listed SEC filers. Investi does not cover stocks listed only outside the US. - Investi has no native Google Sheets sync and does not store files in the workspace. - Investi does not execute trades itself. Third-party connectors that can place orders ask the user to approve every action. ## Pricing Investi makes money from subscriptions. It does not sell data, take payment for order flow or earn commissions. - Free: $0. The full SEC record and the whole workspace, the standard model with light reasoning, and one brokerage connection. No card required. - Plus: $20/mo (or $16/mo, $192 billed yearly). Frontier models, every reasoning level, a larger standard allowance, multiple brokerages, early features and priority support. - Premium: $60/mo (or $48/mo, $576 billed yearly). Everything in Plus, with three times the frontier and standard model allowance. - Enterprise: Custom. Custom usage, dedicated support and invoice billing for teams. - [Pricing](https://tryinvesti.com/pricing) ## Who it's for Individual investors who own stocks or funds, analysts who need numbers they can put in a model, advisors who answer client questions, and students learning to read filings. ## Safety and data Brokerage logins run through SnapTrade, so broker credentials never reach Investi's servers. Investi can see positions, balances and past transactions; it cannot place trades, move cash or change settings. User data is not used to train any model. Disconnecting a brokerage deletes everything synced through that connection from Investi. ## Name and identity Investi is the product at tryinvesti.com, made by Ombra, Inc.. Its analysis model is called Vela. Other companies or apps with similar names are not affiliated with it. - [About Investi](https://tryinvesti.com/about) - [Vela](https://tryinvesti.com/vela) ## Machine-readable versions - [llms.txt](https://tryinvesti.com/llms.txt) - [llms-full.txt](https://tryinvesti.com/llms-full.txt) - [Documentation llms.txt](https://docs.tryinvesti.com/llms.txt) --- # Methodology > Where Investi's data comes from, how it is cleaned and cited, and who writes, checks and corrects what we publish. Canonical: https://tryinvesti.com/methodology _Updated 2026-10-23_ ## Sources Company and fund data come from the SEC's EDGAR system: the full record from 1993 to today, with new filings arriving within minutes. That includes financial statements filed in XBRL, fund holdings from N-PORT, institutional holdings from 13F, ownership filings (13D and 13G) and insider filings (Forms 3, 4 and 5). Macro and government data come from the agencies that publish them: FRED, BLS, BEA, Treasury, FDA, clinical trials, FDIC and more. Market prices are a delayed last trade, 15-20 minutes behind. ## Cleaning Values are what filers and agencies submitted; nothing is estimated. - Amendments are applied to the filings they amend. - When a company restates, both the value as first reported and the restated value are kept, with the filing behind each. - Metrics are point in time: they use what was public on the date in question. - 13F rows flagged as suspect, and options positions, are excluded from holdings totals. - Shares withheld to pay tax are not counted as insider sales. ## Refresh cadence - SEC filings: within minutes of acceptance. - Prices: delayed 15-20 minutes. - Fund holdings: as funds file N-PORT; the SEC makes only each quarter's final month public. - Numbers on this website come from committed snapshots. Each snapshot records when it was fetched and the data version it was fetched from. ## How citations work Every number links to its source: the exact passage or table cell in the filing, not an article about it. A source card shows the form, the filer, the filing date, the accession number and the document's sha256, so anyone can confirm the document hasn't changed. Metrics aren't stored. They're formulas over filed numbers, and you can see the formula. A mismatch check compares a cited value with its source and flags any difference. ## Gaps We disclose the gaps instead of papering over them. Coverage: US-listed companies and funds (SEC filers, including foreign issuers filing 20-F), plus US and international macro data. ## Who writes and reviews Every article names its author, a member of the Investi team listed on the about page. We don't publish testimonials or quote users. Pages with numbers about companies or funds are checked by a named reviewer, who checks each number against the filing it cites. Comparison pages are re-reviewed at least every 90 days and show the date. We use AI tools, including Investi itself, to research and draft. A person checks every number and signs off before anything is published. - [About Investi](https://tryinvesti.com/about) ## Comparisons Each comparison states its criteria, says where the other product is better, and says who should pick which. We compare approach and features only, and publish no scores until there are evaluation runs anyone can check. - [How we compare](https://tryinvesti.com/compare) ## Corrections If something is wrong, write to hi@tryinvesti.com. We fix it, update the page's date, and note material corrections at the end of the article. Investi is a research and information tool, not a licensed investment adviser. Nothing it produces is a personal recommendation. Investi makes money from subscriptions. It does not sell data, take payment for order flow or earn commissions. --- # Best AI tools for stock research (2026) > AI tools for stock research compared by job: a research workspace on SEC filings, a data terminal with a copilot, a web answer engine, or a general assistant. Canonical: https://tryinvesti.com/compare/best-ai-tools-for-stock-research _By Gabri Ansaldo · Published 2026-10-23 · Reviewed 2026-10-23_ There is no single best AI tool for stock research; there are tools for different jobs. Investi is an AI analyst and workspace on the SEC record. Fiscal.ai is a data terminal with an AI copilot. Perplexity Finance is a fast web answer engine. A general assistant with a data connector covers ad-hoc questions. We make one of these tools, so read this with that in mind. We describe each from its own public pages, say where the others are better, and leave out scores. ## How we compared We use the same four criteria everywhere: how deep a tool reaches into the filed record, whether a number links to its source, whether your work persists, and price. For AI tools the second criterion matters most. A model can be fluent and wrong, so a tool earns trust by showing you where each number came from. See [how we compare](https://tryinvesti.com/compare). | Tool | Best for | Sources | Your work | | --- | --- | --- | --- | | Investi | Researching US companies from their filings | The filing, down to the passage or cell | Notes, theses and folders the analyst reads | | Fiscal.ai | Standardized global data with an AI copilot | Click-through to the source of a data point | Custom metrics, charts and exports | | Perplexity Finance | Fast answers to a one-off question | Web results, including news and filings | Threads | | A general assistant with a data connector | Ad-hoc questions in a tool you already use | Depends on the connector and the model | The conversation | _Approach and features only, from each product's public pages. Reviewed October 2026._ ## Investi: an AI analyst on the SEC record Investi reads the full SEC record, from 1993 to today, and answers with a citation on every number, down to the passage or table cell. Calculations show their formula. Notes and theses sit next to each company, and the analyst reads them before it answers. Routines rerun a question on a schedule and email you the result. Brokerage connections are read-only today. Coverage is US-listed filers only, and it has no call transcripts or ratings of its own. ## Fiscal.ai: a data terminal with a copilot Fiscal.ai pairs standardized global financials and deep segment and KPI data with an AI copilot that summarizes documents and screens in plain language. It also sells its data through an API and connectors for AI assistants. Better than Investi for companies outside the US and for standardized comparisons. [Fiscal.ai vs Investi](https://tryinvesti.com/compare/fiscal-ai). ## Perplexity Finance: a web answer engine Perplexity Finance searches the web, including news and commentary, and returns a sourced answer in seconds. Better than Investi for a quick first pass on a name you don't follow or for what happened this morning. [Perplexity Finance vs Investi](https://tryinvesti.com/compare/perplexity-finance). ## A general assistant with a data connector Connect a financial data source to the chat assistant you already use and you can ask ad-hoc questions without a new tool. Quality depends on the connector and on whether the model keeps its sources. The work lives in conversations. We explain why we went the other way in [Why we built Investi as an app](https://tryinvesti.com/blog/why-an-app-not-an-mcp). ## Who should pick which? Pick Investi if you follow US companies and want answers you can trace to a filing, in a workspace that remembers. Pick Fiscal.ai for standardized global data. Pick Perplexity Finance for fast first passes. Use a general assistant for quick questions. Whichever you choose, open the source behind any number you act on. None of these tools, ours included, is investment advice. --- # Fiscal.ai vs Investi > Fiscal.ai is a financial data terminal and API with an AI copilot. Investi is an AI analyst and workspace over the SEC record. How they differ, and who should pick which. Canonical: https://tryinvesti.com/compare/fiscal-ai _By Gabri Ansaldo · Published 2026-10-23 · Reviewed 2026-10-23_ Fiscal.ai is a financial data company: a research terminal, a data API and connectors that put its data inside AI tools, with an AI copilot across the terminal. Investi is an AI analyst and workspace on the SEC record, where your notes and theses persist and every number links to its filing. ## How do Fiscal.ai and Investi differ? Fiscal.ai, formerly FinChat, is built around data: clean financials, segments and KPIs, delivered through its terminal, its API and its connectors for AI assistants. Investi is built around the work: an analyst that reads the filings and your own notes, in a workspace where research accumulates. Both care about sources. Fiscal.ai lets you click through a data point to its source; Investi cites the passage or cell in the filing for every number in an answer. | | Fiscal.ai | Investi | | --- | --- | --- | | Approach | A data terminal with an AI copilot, plus an API and AI connectors | An AI analyst in a workspace you keep | | Coverage | Public companies globally, with standardized and as-reported financials | US-listed SEC filers and funds, plus US and international macro data | | Depth | A large database of company-specific segments and KPIs | The full SEC record from 1993, including footnotes and exhibits | | Your work | Custom metrics, charts and spreadsheet export | Notes, theses and folders next to each company, read by the analyst | _Approach and features only, from each product's public pages. Reviewed October 2026._ ## What we compared We rank tools on the same four criteria everywhere: how deep they reach into the filed record, whether a number links to its source, whether your work persists, and price. Prices change often, so check each product's pricing page. See [how we compare](https://tryinvesti.com/compare). ## Where is Fiscal.ai better? Standardized global data. Fiscal.ai covers companies outside the US, presents financials in both standardized and as-reported form, and tracks company-specific segment and KPI data in depth. It offers call transcripts, which Investi doesn't. If you want that data inside a chat assistant you already use, its connectors do that, and its API serves teams building their own tools. ## Where is Investi better? The work around the numbers. In Investi your notes and theses sit next to each company, the analyst reads them before it answers, and any change it proposes to your notes shows as a diff you approve. It reaches into the whole filing, not just the financial statements, so a question about a footnote, a risk factor or an exhibit gets an answer with the passage cited. You can teach it your process with skills, run Python in a notebook, connect your own data through MCP, and set routines that rerun a question on a schedule. ## Who should pick which? Pick Fiscal.ai if you want clean, standardized data on companies worldwide, or a data feed for your own models and tools. Pick Investi if you follow US companies and want a workspace where the analyst knows your thesis and every answer traces to a filing. We wrote about the trade-off in [Why we built Investi as an app](https://tryinvesti.com/blog/why-an-app-not-an-mcp). --- # Koyfin vs Investi > Koyfin is a configurable market data platform with dashboards and charts across asset classes. Investi is an AI analyst over the SEC record. Who should pick which. Canonical: https://tryinvesti.com/compare/koyfin _By Gabri Ansaldo · Published 2026-10-23 · Reviewed 2026-10-23_ Koyfin is a market data platform: dashboards, charts and screeners you configure, across stocks, funds, yields, currencies and more, worldwide. Investi is an AI analyst and workspace built on the SEC record, where every number links to the filing it came from. ## How do Koyfin and Investi differ? Koyfin gives you the data and lets you arrange it: dashboards, graphs, a screener and macro views that you configure. Investi gives you an analyst: you ask a question, it reads the filings and your notes, and it answers with a source on every number. | | Koyfin | Investi | | --- | --- | --- | | Approach | Dashboards, charts and screens you build | An AI analyst you ask, in a workspace you keep | | Coverage | Global, across stocks, ETFs, mutual funds, yields, currencies, commodities and crypto | US-listed SEC filers and funds, plus US and international macro data | | Sources | Aggregated market and financial data | The filing itself, down to the passage or table cell | | Your work | Watchlists and custom dashboards | Notes, theses and folders next to each company, read by the analyst | | Portfolio | Portfolio analysis and client-ready reports for advisors | Read-only brokerage sync with returns and risk stats | _Approach and features only, from each product's public pages. Reviewed October 2026._ ## What we compared We rank tools on the same four criteria everywhere: how deep they reach into the filed record, whether a number links to its source, whether your work persists, and price. Prices change often, so check each product's pricing page. See [how we compare](https://tryinvesti.com/compare). ## Where is Koyfin better? Breadth. Koyfin covers markets worldwide and across asset classes, from government yields in dozens of countries to currencies and commodities. Its screener runs across a very large global universe with thousands of filter criteria, and its charting is built for people who think visually. It also searches call transcripts, which Investi doesn't offer. Advisors get portfolio reports they can hand to clients. ## Where is Investi better? Depth on US companies and a record of your own thinking. Investi reads the full SEC record, from 1993 to today, and every number it gives you links to the exact passage or cell in the filing. Calculations show their formula. Your notes and theses sit next to each company, and the analyst reads them before it answers, so the next question starts from what you already know. [Routines](https://tryinvesti.com/features#routines) rerun a question on a schedule and email you the result. ## Who should pick which? Pick Koyfin if you want to watch many markets at once and build your own views of the data. Pick Investi if you research US companies one at a time and want answers you can trace to a filing, kept in a workspace that lasts. --- # Perplexity Finance vs Investi > Perplexity Finance is a research desk for the question you have now. Investi is a workspace for the work that has to last. Keep both open. Canonical: https://tryinvesti.com/compare/perplexity-finance _By Gabri Ansaldo · Published 2026-09-14 · Reviewed 2026-10-23_ Perplexity Finance answers the question you have right now, with live sources, in a thread. Investi keeps your coverage: notes and theses sit next to each company, every number links to its SEC filing, and the next question starts from what you already know. ## How do Perplexity Finance and Investi differ? They do different jobs. One is a research desk. The other is a workspace for the work that has to last. We are not going to tell you to close theirs. | | Perplexity Finance | Investi | | --- | --- | --- | | Job | Answer the question you have right now | Hold the coverage so the next question starts from what you already know | | Object | A thread | A company, a folder, a thesis | | Memory | Lives in the conversation | Notes and theses sit next to the company they are about | | Sources | Web results, including articles about filings | The filing itself, down to the passage or cell | | Portfolio | You describe what you own | Read-only brokerage sync; holdings land in the workspace | | Schedule | A Task that starts clean | A routine that can open the folder and read the thesis | _Approach and features only, from each product's public pages. Reviewed October 2026._ ## What we compared We rank tools on the same four criteria everywhere: how deep they reach into the filed record, whether a number links to its source, whether your work persists, and price. See [how we compare](https://tryinvesti.com/compare). ## Where is Perplexity Finance better? Ask why a name dropped this morning and Perplexity Finance will go look for the downgrade, the filing, and the headline, in seconds. It searches the open web, so it reaches news and commentary that a filings-first tool does not. For a quick first pass on a name you don't follow, that is the better tool. ## Where is Investi better? Investi is built around the record. Your notes, theses, and watchlists sit next to the companies they are about, and the analyst reads them before it answers. Every number names the filing it came from, not an article about it. A [brokerage connection](https://tryinvesti.com/blog/connect-your-brokerage) is read-only today, so the analyst reasons over what you actually hold. ## Who should pick which? Pick Perplexity Finance if you want a fast answer to a one-off question. Pick Investi if the name is one you follow and the work has to be there next quarter. You can also connect Perplexity to Investi as a plugin, so keep both open: use the desk for the first pass, and the studio for the record. --- # Seeking Alpha vs Investi > Seeking Alpha is a community of investing ideas with ratings and articles. Investi is an AI analyst over the SEC record. How they differ, and who should pick which. Canonical: https://tryinvesti.com/compare/seeking-alpha _By Gabri Ansaldo · Published 2026-10-23 · Reviewed 2026-10-23_ Seeking Alpha is a publishing platform for investing ideas: articles from contributors, news, quant and author ratings, and screeners. Investi is an AI analyst and workspace that answers from the SEC filings themselves, with a source on every number. ## How do Seeking Alpha and Investi differ? Seeking Alpha is about opinions: thousands of contributors write about stocks, and the platform rates them, rates the stocks, and shows how past ideas performed. Investi is about evidence: you ask a question and the analyst answers from the filings, so you can form your own opinion. | | Seeking Alpha | Investi | | --- | --- | --- | | Approach | Articles, news and ratings from contributors and a quant model | An AI analyst you ask, in a workspace you keep | | Ratings | Quant Ratings, factor grades, author and sell-side ratings | None of its own; it doesn't tell you what to buy | | Sources | Contributor analysis, news and financial data | The filing itself, down to the passage or table cell | | Your work | Portfolios, watchlists and alerts | Notes, theses and folders next to each company, read by the analyst | | Portfolio | Broker linking with daily updates | Read-only brokerage sync with returns and risk stats | _Approach and features only, from each product's public pages. Reviewed October 2026._ ## What we compared We rank tools on the same four criteria everywhere: how deep they reach into the filed record, whether a number links to its source, whether your work persists, and price. Prices change often, so check each product's pricing page. See [how we compare](https://tryinvesti.com/compare). ## Where is Seeking Alpha better? Ideas and opinion at scale. If you want to know what other investors think of a stock, and how their past calls on it worked out, Seeking Alpha has years of articles and a track record for each author. Its Quant Ratings and factor grades give a fast first read across thousands of stocks, and it offers call transcripts, news, dividend scores and ETF screens, none of which Investi has. ## Where is Investi better? Checking claims against the source. An article tells you what an author thinks the 10-K says; Investi reads the 10-K, and every number it gives you links to the passage or cell it came from. Calculations show their formula. Your own notes and theses sit next to each company and the analyst reads them, so your research builds on itself instead of living in bookmarks. Investi has no ratings of its own, by design: it helps you think; it doesn't think for you. ## Who should pick which? Pick Seeking Alpha if you want ideas, debate and ratings from a large community. Pick Investi if you want to do the reading yourself, faster, with sources. Many investors will use both: an idea from an article, checked against the filings. --- # How much are Microsoft, Alphabet, Amazon and Meta spending on capex? > What the cash flow statements of Microsoft, Alphabet, Amazon and Meta show about capital spending in the twelve months to June 2026, and how to check it yourself. Canonical: https://tryinvesti.com/blog/big-tech-capex-2026 _By The Investi team · Published 2026-10-24 · Investing_ Microsoft, Alphabet, Amazon and Meta spent a combined $511 billion on property and equipment in the twelve months to June 30, 2026, according to the cash flow statements in the 10-K and 10-Qs they filed in July 2026. The same four spent $291 billion in the twelve months before that, so the total rose about 75%. Amazon spent the most, $173 billion, which was more than the $161 billion of cash its operations brought in over the period. The other three spent between 63.4% and 71.3% of their operating cash flow. This post shows where those numbers sit in each filing, what the capex line leaves out, and the questions an investor might ask next. It reads the filings; it doesn't recommend buying or selling anything. Investi is a research and information tool, not a licensed investment adviser. **Key takeaways** - Combined purchases of property and equipment at Microsoft, Alphabet, Amazon and Meta were $511 billion in the twelve months to June 30, 2026, up from $291 billion a year earlier. - Alphabet's spending nearly doubled, to $132 billion from $67.0 billion. Amazon's was the largest, at $173 billion. - Amazon's capex was 107.2% of its operating cash flow over the period. Microsoft, Alphabet and Meta were between 63.4% and 71.3%. - The four also disclosed $831 billion of signed leases, mostly data centers, that had not started by June 30, 2026. Those are not in the capex line. ## Where does the capex number come from? Capital expenditures, or capex, are what a company pays for long-lived assets such as buildings, servers and network equipment. In a filing they sit in the cash flow statement, under investing activities. Each company names the line its own way: Microsoft calls it "Additions to property and equipment", while Alphabet, Amazon and Meta call it "Purchases of property and equipment". The periods need care. A 10-Q's cash flow statement covers the year to date, not the quarter. To get twelve months ending June 30, 2026 for a company whose fiscal year ends in December, take its last full year, add the first half of 2026 and subtract the first half of 2025. Microsoft's fiscal year ends June 30, so its fiscal 2026, in the [10-K filed July 29, 2026](https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm), already covers those twelve months. Amazon makes the arithmetic easy to check: its [10-Q filed July 31, 2026](https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm) prints trailing-twelve-month columns in the cash flow statement, and they match the figures computed here to the million. ## How much did each company spend? All four spent more than in the previous twelve months, as of their filings for the period ended June 30, 2026: - Microsoft spent $116 billion in fiscal 2026, up 79.6% from $64.6 billion in fiscal 2025 ([10-K](https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm)). - Alphabet spent $132 billion, up 97.7% from $67.0 billion ([10-Q filed July 23, 2026](https://www.sec.gov/Archives/edgar/data/1652044/000165204426000071/goog-20260630.htm)). - Amazon spent $173 billion, up 60.7% from $108 billion ([10-Q filed July 31, 2026](https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm)). - Meta spent $89.3 billion, up 71.3% from $52.2 billion ([10-Q filed July 30, 2026](https://www.sec.gov/Archives/edgar/data/1326801/000162828026050705/meta-20260630.htm)). **Capital expenditures, twelve months to June 30, 2026** | Company | Capital expenditures | | --- | --- | | Microsoft | $116B | | Alphabet | $132B | | Amazon | $173B | | Meta | $89.3B | Source: 10-K and 10-Q filings, Jul 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm)) _Amazon spent the most over the twelve months, $173 billion, up from $108 billion a year earlier._ Amazon's figure is gross. Its cash flow statement shows proceeds from property sales and incentives on a separate, much smaller line, so a net figure would be slightly lower. Assets a company obtains through leases aren't in any of the four lines. ## How does capex compare with operating cash flow? Operating cash flow is the cash a business brings in from running itself, before it invests. Dividing capex by it shows how much of that cash went back into property and equipment. Operating cash flow minus capex is the simplest version of free cash flow. Over the twelve months to June 30, 2026, Microsoft's capex was 63.4% of its operating cash flow of $183 billion, Alphabet's 71.3% of $186 billion, and Meta's 68.5% of $130 billion. Amazon's was 107.2% of $161 billion: on this measure its free cash flow for the period was negative. **Capex as a share of operating cash flow, twelve months to June 30, 2026** | Company | Capex as a share of operating cash flow | | --- | --- | | Microsoft | 63.4% | | Alphabet | 71.3% | | Amazon | 107.2% | | Meta | 68.5% | Source: 10-K and 10-Q filings, Jul 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm)) _Amazon's purchases of property and equipment were 107.2% of its operating cash flow; the other three spent 63.4% to 71.3%._ A ratio above 100% says the spending in that period was larger than the cash operations produced. The difference came from somewhere else on the cash flow statement, such as cash on hand, sales of securities or borrowing. The filings show which, in the investing and financing sections. ## What does the capex line leave out? Leases. Each company also discloses leases it has signed that had not started by the end of the period, mostly for data centers. They aren't on the balance sheet or in capex yet. As of June 30, 2026, the four filings list $831 billion of them combined: - Microsoft: $329 billion, "primarily for datacenters", which the [10-K](https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm) says will commence between fiscal 2027 and fiscal 2033, with some arrangements "subject to certain contractual conditions being met". - Meta: approximately $279 billion, commencing through 2036 ([10-Q](https://www.sec.gov/Archives/edgar/data/1326801/000162828026050705/meta-20260630.htm)). - Amazon: $137 billion, a row in the commitments table of its [10-Q](https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm). - Alphabet: $85.2 billion, commencing between 2026 and 2031 ([10-Q](https://www.sec.gov/Archives/edgar/data/1652044/000165204426000071/goog-20260630.htm)). **Leases signed but not yet commenced, as of June 30, 2026** | Company | Leases not yet commenced | | --- | --- | | Microsoft | $329B | | Meta | $279B | | Amazon | $137B | | Alphabet | $85.2B | Source: 10-K and 10-Q filings, Jul 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm)) _Microsoft disclosed the largest amount of signed leases not yet started, $329 billion, primarily for data centers._ These are future payments spread over lease terms that the filings put at up to 20 years for Microsoft and up to 30 years for Meta, so they don't compare directly with one year of capex. Each company also words and measures them a little differently, which makes the ranking rough. They do show commitments that a reader of the capex line alone would miss. ## What changed in Meta's September quarter? Meta is the first of the four to report the quarter ended September 30, 2026. Its [10-Q filed October 22, 2026](https://www.sec.gov/Archives/edgar/data/1326801/000162828026066773/meta-20260930.htm) shows $83.1 billion of capex for January to September 2026, up 71.9% from $48.3 billion in the same nine months of 2025. Its leases not yet commenced were approximately $356 billion at September 30, up from approximately $279 billion three months earlier. The [June 10-Q](https://www.sec.gov/Archives/edgar/data/1326801/000162828026050705/meta-20260630.htm) had already noted about $68 billion of data center leases signed in July 2026. As of October 24, 2026, Microsoft, Alphabet and Amazon had not yet filed their reports for the September quarter. When they do, the twelve-month figures above will move. ## What would an investor ask next? The capex line answers how much. It doesn't answer what for, at what return, or how it's paid for. These are the questions the filings can help with: 1. **What is the spending for?** The cash flow statement doesn't split capex between data centers, servers, warehouses and offices. Item 2 of a 10-Q (management's discussion) is where each company describes it in its own words. 2. **How fast will it reach the income statement?** Property and equipment is expensed as depreciation over its useful life once it's in service. The cash flow statement shows depreciation each period, and the notes give the useful lives each company assumes. 3. **How is it funded?** Compare capex with operating cash flow, then read the financing section for debt issued and repaid, buybacks and dividends. 4. **When do the leases start?** The lease notes give commencement years and terms, and whether leases are expected to be operating or finance leases. 5. **Who receives the money?** Suppliers report the other side. Nvidia's 10-Q now breaks out revenue from what it calls hyperscale customers; we read it in [How much of Nvidia's revenue comes from hyperscalers?](https://tryinvesti.com/blog/nvidia-data-center-revenue-2026). If you track one of these companies, those questions make reasonable lines in a written thesis, with a note of what would change your mind. [How to build an investment thesis](https://tryinvesti.com/blog/how-to-build-an-investment-thesis) walks through that. ## How can you check these numbers yourself? 1. **Open the filing on EDGAR.** Search the company on sec.gov and open the latest 10-Q or 10-K. The links above go straight to each document. 2. **Find the cash flow statement.** Under investing activities, read the line for purchases or additions to property and equipment, and note the period it covers. 3. **Build the twelve months.** For a December fiscal year, add the last full year to this year's year-to-date figure and subtract last year's year-to-date figure. 4. **Read the lease note.** Search the document for "not yet commenced" to find signed leases that haven't started, with their timing and terms. **What is capex?** Capital expenditures are cash a company spends on long-lived assets such as land, buildings, servers and equipment. They appear in the investing section of the cash flow statement. Unlike operating costs, they aren't expensed when paid; the cost reaches the income statement gradually, as depreciation, once the asset is in use. **Why use the twelve months to June 30, 2026?** It is the latest period all four companies had filed as of October 24, 2026. Microsoft's fiscal year ends on June 30, so its fiscal 2026 matches exactly. For Alphabet, Amazon and Meta, the twelve months are computed from their 10-K and the first-half figures in their 10-Qs filed in July 2026. **Is capex the same as AI spending?** No. The capex line covers all property and equipment, including warehouses, offices and network gear. The cash flow statement doesn't split it by purpose. Several filings describe the leases they have signed as primarily data centers, but the share of capex that goes to AI is something each company describes in its own words, not a filed number. **Does capex above operating cash flow mean trouble?** Not by itself. It means the company invested more in the period than its operations produced, and funded the difference from cash, investments or borrowing. Whether that is sensible depends on what the assets earn later, which no single period's filing can show. The financing section shows how the gap was covered. The figures here are as of the filings named above, and each one links to the document it came from. To run the same comparison on other companies, open their cash flow statements and repeat the steps. In [Investi](https://tryinvesti.com/features), you can ask the analyst for this table and each number links to the line of the filing it came from, with the twelve-month formula shown. --- # How much of Nvidia's revenue comes from hyperscalers? > What Nvidia's 10-Q for the quarter ended July 26, 2026 shows about its data center revenue, the new Hyperscale line, customer concentration, and how to check it. Canonical: https://tryinvesti.com/blog/nvidia-data-center-revenue-2026 _By The Investi team · Published 2026-10-24 · Investing_ In the quarter ended July 26, 2026, Nvidia reported $48.7 billion of revenue from what it calls Hyperscale customers, about 55% of its $89.0 billion of data center revenue, according to the 10-Q it filed on August 26, 2026. That was more than double the $24.2 billion of a year earlier. The rest of data center revenue, $40.3 billion, came from a group Nvidia calls AI Clouds, Industrial & Enterprise, up from $16.9 billion. Across its last full fiscal year, data center was 89.7% of Nvidia's revenue. This post reads the filing: where the numbers are, what the new categories mean, what the filing says about how concentrated the customers are, and what it doesn't say. It is not a recommendation to buy or sell. Investi is a research and information tool, not a licensed investment adviser. **Key takeaways** - Nvidia's Hyperscale revenue was $48.7 billion in the quarter ended July 26, 2026, up from $24.2 billion a year earlier, per its 10-Q filed August 26, 2026. - AI Clouds, Industrial & Enterprise grew faster, to $40.3 billion from $16.9 billion. The filing says that group includes hyperscalers renting capacity from AI clouds. - Data center revenue rose from $10.6 billion in fiscal 2022 to $194 billion in fiscal 2026, when it was 89.7% of total revenue. - One direct customer accounted for 16% of total revenue in the quarter. The filing doesn't name customers or list who is in each group. ## How big is Nvidia's data center business? Nvidia's fiscal year ends in late January, so its fiscal 2026 ran to January 25, 2026. In the [10-K filed February 25, 2026](https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm), data center revenue was $194 billion out of $216 billion in total, up 68.2% from $115 billion in fiscal 2025. Four years earlier, in fiscal 2022, data center was $10.6 billion, or 39.4% of revenue. **Nvidia data center and total revenue, fiscal 2022 to 2026** | Fiscal year | Data center | Total revenue | | --- | --- | --- | | FY2022 | $10.6B | $26.9B | | FY2023 | $15B | $27B | | FY2024 | $47.5B | $60.9B | | FY2025 | $115B | $130B | | FY2026 | $194B | $216B | Source: 10-K filings, Feb 2024 to Feb 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm)) _Data center revenue grew from $10.6 billion in fiscal 2022 to $194 billion in fiscal 2026, when it was 89.7% of Nvidia's total._ Data center is what Nvidia calls a market platform, a way of grouping revenue by the market it's sold into. It is not one of Nvidia's two reporting segments, which are Compute & Networking and Graphics. The two tables group the same total revenue in different ways, so a figure from one doesn't belong in the other. ## What do Hyperscale and ACIE mean? Starting in the first quarter of fiscal 2027, Nvidia splits data center revenue into two groups: Hyperscale, and AI Clouds, Industrial & Enterprise, which the filing shortens to ACIE. The [10-Q for the quarter ended July 26, 2026](https://www.sec.gov/Archives/edgar/data/1045810/000104581026000075/nvda-20260726.htm) says earlier periods were recast to the new presentation, so the year-ago figures below are comparable. **Nvidia data center revenue by customer group** | Customer group and quarter | Revenue | | --- | --- | | Hyperscale, Q2 FY2026 | $24.2B | | Hyperscale, Q2 FY2027 | $48.7B | | AI Clouds, Industrial & Enterprise, Q2 FY2026 | $16.9B | | AI Clouds, Industrial & Enterprise, Q2 FY2027 | $40.3B | Source: 10-Q filings, Aug 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/1045810/000104581026000075/nvda-20260726.htm)) _Hyperscale revenue more than doubled in a year, from $24.2 billion to $48.7 billion, while ACIE grew from $16.9 billion to $40.3 billion._ Two details in the filing matter for reading these lines: 1. The groups overlap in practice. Nvidia's management discussion says ACIE growth was driven by "AI natives, enterprises, and sovereign customers, as well as hyperscalers utilizing AI clouds". So some revenue that ultimately serves a hyperscaler sits in ACIE, when the hyperscaler rents capacity from a cloud that buys from Nvidia. 2. The groups can change. In the quarter, Nvidia "reclassified a company from ACIE to Hyperscale due to a change in their business model and recast the prior period revenue associated with this company." As a result, the first-quarter figures in the first-quarter 10-Q don't match the recast first-quarter figures in the second-quarter 10-Q. Comparing two periods from the same filing avoids that. The filing doesn't define Hyperscale with a list of names, and it doesn't say which company was moved. ## How concentrated are Nvidia's customers? The 10-Q discloses customers above 10% of revenue without naming them. In the quarter ended July 26, 2026, one direct customer represented 16% of total revenue. For the first half of fiscal 2027, three direct customers represented 16%, 15% and 13%. A year earlier, in the quarter ended July 27, 2025, two direct customers represented 23% and 16%. "Direct" matters. The filing lists direct customers as distributors, add-in board makers, original design and equipment manufacturers, cloud service providers, AI model makers and system integrators. Indirect customers "primarily purchase our products through system integrators and distributors", and Nvidia says its indirect customer revenue "is an estimation". A direct customer can be a company that builds servers for someone else, so the direct figures don't tell you who ends up running the chips. ## Where are Nvidia's customers? The geographic table is also about direct customers. The 10-Q says revenue by region "is designated based on the location of the headquarters of direct customers", and that "the end customer and shipping location may be different". Customers headquartered outside the United States accounted for 38% of total revenue in the quarter, against 30% a year earlier. The same filing says shipments of data center Hopper products to China were less than 1% of data center revenue in the quarter. ## How does this connect to Big Tech capex? Nvidia's Hyperscale line is the sales side of a story the buyers tell in their cash flow statements. Microsoft, Alphabet, Amazon and Meta spent a combined $511 billion on property and equipment in the twelve months to June 30, 2026; we read those filings in [How much are Microsoft, Alphabet, Amazon and Meta spending on capex?](https://tryinvesti.com/blog/big-tech-capex-2026) The two don't map onto each other. Nvidia doesn't say those four are its Hyperscale customers, its quarters end in late July rather than June, and capex includes buildings, land and network equipment as well as chips. Reading both sides each quarter still shows whether the buyers' spending and the supplier's sales keep moving together. ## What would an investor ask next? 1. **Who is in each group?** The filing doesn't say, and the reclassification shows membership can change. Comparing periods from one filing keeps the definition constant. 2. **How much depends on a few customers?** The direct-customer disclosure answers part of it. The indirect picture is Nvidia's own estimate. 3. **How do the buyers' plans compare?** The capex and lease disclosures in the hyperscalers' filings show what they spent and what they've committed to; they don't say how much of it goes to Nvidia. 4. **What do the next filings show?** Nvidia's next 10-Q will update both groups, with the year-ago quarter recast if membership changes again. ## How can you check these numbers yourself? 1. **Open the 10-Q on EDGAR.** Search NVIDIA on sec.gov and open the 10-Q filed August 26, 2026, for the quarter ended July 26, 2026. 2. **Find revenue by market platform.** The segment note and Item 2 both have a table with Data Center, Hyperscale and AI Clouds, Industrial & Enterprise, with the year-ago quarter alongside. 3. **Read the footnotes.** The footnotes under the table explain the change in presentation and the company that moved between groups. 4. **Search for concentration.** Search the document for "direct customer" to find the customers above 10% of revenue, and for "headquarters" to see how regions are assigned. **Who are Nvidia's hyperscale customers?** Nvidia's 10-Q doesn't name them. It reports Hyperscale as a revenue group, discloses unnamed customers above 10% of revenue, and notes that it moved one company from AI Clouds, Industrial & Enterprise to Hyperscale in the quarter ended July 26, 2026. Any list of names comes from outside the filing. **Is data center one of Nvidia's segments?** No. Nvidia's two reporting segments are Compute & Networking and Graphics. Data center is a market platform, a separate view of the same revenue grouped by the market it's sold into. Its subgroups, Hyperscale and AI Clouds, Industrial & Enterprise, are part of that market-platform view. **Why don't Nvidia's quarters line up with the calendar?** Nvidia's fiscal year ends in late January. Its fiscal 2026 ended January 25, 2026, and the second quarter of fiscal 2027 ended July 26, 2026. When you compare Nvidia with companies on a calendar year, the quarters are offset by about a month. **Does revenue by country show where Nvidia's chips end up?** Not necessarily. The 10-Q assigns revenue to the country where the direct customer has its headquarters and says the end customer and shipping location may differ. A server maker headquartered in one country can sell to a cloud company in another, and the filing would show the first. The figures here are as of Nvidia's 10-K filed February 25, 2026 and its 10-Q filed August 26, 2026, and each one links to the filing. To follow them, reopen the revenue-by-market-platform table each quarter and compare it with the buyers' cash flow statements. In [Investi](https://tryinvesti.com/features), you can ask the analyst for Nvidia's data center revenue by group, and each figure links to the cell of the filing it came from, as described in [How the analyst cites a filing](https://tryinvesti.com/blog/how-the-analyst-cites-a-filing). --- # How the analyst cites a filing > What happens when you check a number in an Investi answer: the passage or table cell in the SEC filing, a value check, and the formula behind any calculation. Canonical: https://tryinvesti.com/blog/how-the-analyst-cites-a-filing _By The Investi team · Published 2026-10-23 · Product_ An AI answer about a company is only useful if you can check it, and checking usually means hunting through a 10-K for the right page. In Investi, each number the analyst takes from a source carries a small numbered marker, and hovering over it opens the passage or table cell in the SEC filing it came from, with the number highlighted. This post explains how it works. ## What a cited answer looks like An answer reads as ordinary prose. After each figure sits a small numbered chip. When a claim rests on more than one number, such as "net income rose 19%", the analyst underlines the words instead and ties them to every source behind them. In a table, each cell can carry its own source. Hover over a chip, or move to it with the keyboard, and a card opens after about half a second. The card's top line names the form, the company and the filing date, with an Open link to the full filing. Below it is the number as the company printed it and the passage around it. ## Pointing at a source instead of writing one The analyst doesn't type out where a number came from. When it looks something up, each fact comes back with a short handle, something like `c7`. The analyst writes that handle next to the number it uses. As the answer streams in, Investi turns each handle into a full citation: the filing, the document inside it, the exact fact or passage, and the value stored there. 1. **Fact looked up.** It comes back with a short handle 2. **Handle written.** The analyst puts it by the number 3. **Citation resolved.** Investi finds the filing and the value 4. **Value checked.** A chip, or a mismatch warning _The analyst only repeats a short handle; Investi builds the citation and checks the value._ This matters because models are good at reading and reasoning and unreliable at copying long identifiers. Every SEC filing has an accession number, the 18-digit ID the SEC assigns when it accepts the filing. A model asked to write one from memory can get a digit wrong, and the link would still look fine. Repeating a handle it was given a moment ago is a much smaller task. If the analyst writes a handle Investi can't resolve, no chip appears. A number without a chip has no source behind it, and you can see that at a glance. ## Opening the exact passage A citation points at one place inside a document. Depending on what the analyst used, that place is one of these: 1. A number in the financial statements, tagged in XBRL. XBRL is the format companies use to label each figure they file, so a machine can read "revenue, fiscal 2024, in dollars" without guessing. 2. A passage of text, such as a risk factor or a paragraph in an 8-K. 3. One cell of a table, named by its row and column. 4. One row of a structured form, such as a single transaction on a Form 4 insider filing. For a tagged number, the card shows what the company called the line in that filing, the scale it rounded to (thousands, millions or billions), the period it covers, any breakdown such as a segment or region, and the statement it belongs to. Below that is the filing itself: the table or paragraph as the company submitted it, shrunk to fit the card and scrolled so the cited number sits in the middle, highlighted. When a filing exists only as a PDF, the card shows an image of the page, scrolled to the cited block. The card fetches the passage the moment your pointer arrives, so by the time it opens the filing is usually already there. ## Checking the number against the source A correct source attached to the wrong number is still wrong. The analyst could write 16,261 where the filing says 16,621, or attach one company's handle to another company's figure. So when Investi resolves a citation, it also compares the figure written in the answer with the value at the source. If the two disagree, the card opens with a line in red, above everything else: "The figure in the answer does not match this source." The check doesn't rewrite the answer or hide the number. It tells you which figure to look at, and the passage underneath shows what the filing actually says. ## Showing the math Many of the numbers investors care about aren't in any filing. Operating margin, trailing-twelve-month revenue and five-year growth all have to be computed. In Investi, metrics aren't stored. They're formulas over filed numbers, and you can see the formula. A computed figure gets its own card, marked Computed. It shows the result, the formula and the period, then lists each input on its own row, and each input links to where it was filed. Take AWS in the quarter ended June 30, 2026. Amazon's [10-Q](https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm) reports $16.62 billion of segment operating income on $42.23 billion of segment revenue. The operating margin is the first divided by the second: 39.4%. A card for that 39.4% lists both inputs, and either one opens the line of the 10-Q it came from. **Operating margin, quarter ended June 30, 2026** | Segment | Operating margin | | --- | --- | | AWS (Amazon) | 39.4% | | Google Cloud (Alphabet) | 35.6% | Source: 10-Q filings, Jul 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm)) _Neither margin appears in a filing. Each is segment operating income over segment revenue, from the company's own 10-Q._ The same formulas are on every company page. In the formulas pane you can write your own line, a name and an expression over filed numbers, and read it by year, by quarter or on a trailing-twelve-month basis. The pad has functions for TTM, CAGR, growth and lag. Longer calculations run in the [notebook](https://tryinvesti.com/blog/introducing-the-notebook), and its tables keep the same sources cell by cell. Citations are one part of the layer around the model; [How the Investi analyst works](https://tryinvesti.com/blog/anatomy-of-a-harness) covers the rest. ## When a number changed after it was filed Sometimes the value in an answer isn't a straight copy from one filing. The card says so in a short note under the passage. The notes cover cases like these: - The figure was restated in a later filing. - It was adjusted for a stock split since it was filed, for example "Adjusted for a 4× stock split since it was filed." - It's a single quarter computed as year-to-date less the prior year-to-date, because the filing only shows the year-to-date figure. Cash flow statements in 10-Qs usually work this way, which is how the twelve-month figures in [what big tech spends on capex](https://tryinvesti.com/blog/big-tech-capex-2026) are built. - It covers a period of a different length from the rest of the series. The card also tells you when the SEC disseminated a correction to a filing, or withdrew it. A withdrawn filing is still shown as it was filed. Restatements keep both values. Driven Brands reported $2.34 billion of fiscal 2024 revenue in the [10-K it filed on February 26, 2025](https://www.sec.gov/Archives/edgar/data/1804745/000180474525000010/drvn-20241228.htm). The [10-K it filed on May 19, 2026](https://www.sec.gov/Archives/edgar/data/1804745/000180474526000048/drvn-20251227.htm) restates the same year at $1.75 billion. Investi's record keeps both the first-reported and the latest value, and each figure cites its own 10-K, so you can see the number as first reported, the number as restated, and the filing that changed it. **Driven Brands FY2024 revenue, as first reported and as restated** | Basis | FY2024 revenue | | --- | --- | | As first reported | $2.34B | | As restated | $1.75B | Source: 10-K filings, Feb 2025 to May 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/1804745/000180474526000048/drvn-20251227.htm)) _One fiscal year, two 10-Ks, $587 million apart. Each figure cites the filing it came from._ ## Sharing a source The Open link on a card goes to a source page: the whole filing with the cited figure marked, and a link to the same document on SEC.gov. When you open it while signed in, the address becomes one you can share. Anyone with the link can open it, whether or not they use Investi. Exports carry their sources too. When you download a table as CSV or .xlsx, each source is named by its line, period, company, form and filing date, and linked to its source page. On the worked examples published on this website, each source card also lists the accession number and the start of the document's sha256. A sha256 is a fingerprint of the file: change one byte and it changes completely, so anyone can confirm the document is the one the SEC published. ## What it doesn't do yet A citation shows where a number came from. It can't show that the analyst picked the right number. If the analyst cites a company's total operating income when you asked about one segment, the value check passes, because the figure does match its source. That's why the card shows the filer's own name for the line, its period and its breakdown, and those are worth reading. Not every source is a filing. A price is cited to its daily close and date, and a total computed across many filed rows is cited as the query that produced it, with its parameters. Coverage is US-listed SEC filers, plus US and international macro data. The analyst can still be wrong. The citations are there so you can check it. The [methodology page](https://tryinvesti.com/methodology) covers where the data comes from and how it's cleaned before anything gets cited. --- # Investi pricing explained > Investi is free to start. Plus is $20 a month and adds frontier models; Premium is $60 with three times the usage. What each plan includes and how billing works. Canonical: https://tryinvesti.com/blog/investi-pricing-explained _By The Investi team · Published 2026-10-23 · Updated 2026-10-24 · Company_ Investi is free to start, with no credit card. As of October 2026, Plus costs $20 a month, or $16 a month billed yearly ($192 a year), and adds frontier models. Premium costs $60 a month, or $48 a month billed yearly ($576 a year), and triples the Plus usage. Teams can ask about Enterprise. Prices can change, so the [pricing page](https://tryinvesti.com/pricing) is the place to check the current numbers. ## What does the free plan include? The Free plan is the whole product with a smaller analyst. The data and the workspace are the same on every plan: the full SEC record with market and macro data, citations on every figure, unlimited notes and folders, routines, skills and plugins, the Python notebook, and one brokerage connection. The analyst runs on Vela, the standard model, with a monthly allowance and light reasoning. ## What does Plus add? Plus is for people who research every week. It adds: - Frontier models, for harder questions and longer work - Every reasoning level, up to Max - A larger Vela allowance - Multiple brokerage connections - Early access to beta features - Live usage tracking and priority support Plus buys a stronger analyst and more room to use it, not access to a different record. ## What does Premium add? Premium is for people who research every day. It is Plus with three times the frontier model allowance and three times the Vela allowance. Everything else is the same as Plus. ## Is there an Enterprise plan? Yes, for teams: custom usage, dedicated support and invoice billing. [Email us](mailto:hi@tryinvesti.com) to talk about it. ## How does billing work? These are the terms as of the [Terms of Service](https://tryinvesti.com/terms) dated August 4, 2026. You choose monthly or yearly, and the plan renews until you cancel. You can cancel from your account settings; cancellation takes effect at the end of the billing period and you keep your plan until then. Payments are non-refundable except where the law requires otherwise, such as the 14-day withdrawal right for consumers in the EU and UK. We give notice before a price change applies to your next renewal. The terms have the full wording. ## How does Investi make money? From subscriptions. Investi does not sell data, take payment for order flow or earn commissions, so its revenue doesn't depend on how often you trade. Brokerage connections are read-only today. More on that in [Is Investi safe?](https://tryinvesti.com/blog/is-investi-safe) --- # Is Investi safe? > What Investi can and can't do with your account, your brokerage connection and your data: read-only brokerage access today, and no data sold. Canonical: https://tryinvesti.com/blog/is-investi-safe _By The Investi team · Published 2026-10-23 · Updated 2026-10-24 · Company_ Investi is a research tool. Brokerage connections are read-only today, so Investi can see your holdings but can't place a trade or move money. We make money from subscriptions, not from selling your data or from your trades. The answers below describe how things work as of October 2026, under the [privacy policy](https://tryinvesti.com/privacy) dated October 23, 2026. ## Can Investi trade or move money in my account? No. Investi does not currently execute trades. When you connect a brokerage, the link goes through SnapTrade with read-only access: holdings, balances and activity flow in so the analyst can reason over what you actually own. There is no order entry in the product, and the analyst has no tool that places one. ## Who handles my brokerage login? SnapTrade, the connection provider, handles the link with your brokerage. You can disconnect at any time: syncing stops and that connection's synced data is removed from our production systems. Your broker's statements stay the record of your account; synced data can lag. ## Does Investi sell my data? No. Investi makes money from subscriptions. It does not sell data, take payment for order flow or earn commissions. Analytics on our website and in the app store nothing on your device unless you accept the consent prompt. If you accept, that can include a session replay of your visit, which only our team can view; you can change your choice from Cookie preferences in the footer. The privacy policy lists what is collected and why. ## What happens to what I type? To answer you, the content the analyst needs (your messages, files you attach, notes you reference, a summary of your portfolio if you connected one) is sent to AI model providers. Under the API terms we use, those providers don't train on it, and we don't use your content to train AI models either. To keep the analyst reliable, we also keep traces of AI conversations in access-restricted logs for a limited period. The privacy policy has the details, including retention. ## Can I trust the numbers? Numbers the analyst takes from filings link to the filing they came from, down to the passage or table cell, and calculations show their formula. Values are what filers and agencies submitted; nothing is estimated. Prices are delayed 15 to 20 minutes. The [methodology](https://tryinvesti.com/methodology) explains where the data comes from, and [How the analyst cites a filing](https://tryinvesti.com/blog/how-the-analyst-cites-a-filing) shows how citations work. AI can still be wrong, which is why the sources are one click away. Check the ones that matter to your decision. ## Is it investment advice? No. Investi is a research and information tool, not a licensed investment adviser. Nothing it produces is a personal recommendation. It helps you think; it doesn't think for you. --- # Why we built Investi as an app > Investi's data is built for AI agents, so why is it an app? Because research needs a place to live: your portfolio, notes, citations and routines, together. Canonical: https://tryinvesti.com/blog/why-an-app-not-an-mcp _By Gabri Ansaldo · Published 2026-10-23 · Company_ Someone asked me recently: "If people find the data valuable, why build an app? Just build an MCP." It's a fair question. MCP is an open standard that lets an AI assistant call another service's tools and data, and Investi's data would work well behind one. My answer is about what we're building today, not a rule: what investors are missing most right now is a place where their research lives, so that's where our time goes. Some of this I'm still figuring out. ## What can the data do? Start with what Investi has: the SEC record, every report US companies, funds and insiders have filed since 1993, cleaned up so an AI agent can search it, query it, read it and cite it. The analyst gets building blocks rather than a fixed menu of ready-made answers, so it can work out answers nobody has packaged. Here's a question where that shows. Which companies changed auditors after their insiders sold stock? I couldn't find that list published anywhere, and the web-search assistants I tried found a couple of names at most. Investi's analyst joins the auditor-change 8-Ks to the insiders' Form 4s. Of the 394 companies that filed an auditor change (8-K Item 4.01) from January to September 2026, 68 had open-market sales by officers, directors or 10% owners in the 90 days before, and each one links to its 8-K. 394 auditor-change 8-Ks → Insider Form 4 sales in the 90 days before → 68 companies, each linked to its 8-K _The answer comes from joining two kinds of filings, and every company on the list keeps its source._ An auditor change has plenty of ordinary causes, and many of those sales are planned trades, so the list is a reason to read the filings, not evidence of wrongdoing. You can check it yourself: run an [EDGAR full-text search](https://www.sec.gov/edgar/search/#/q=%22Item%204.01%22&forms=8-K) for "Item 4.01" in 8-Ks, then open each company's Form 4s for the three months before. Or ask Investi the question and open the citations. So the question comes naturally. If the data is good, why not let people plug it into the agent they already use? ## Why isn't the data the product? A data feed is a real business, and plenty of financial data companies run it well. You sell access, and whoever builds on top owns the user, the experience and the relationship. When a cheaper feed comes along, switching is a small change in their configuration. That isn't the business Investi is in today. The data is how Investi gets there, not what it sells. What I'm after is the part a feed leaves to someone else: the place where you build conviction about what you own. ## Why isn't one good answer enough? Putting money somewhere means building the conviction to do it. Professionals do that deliberately, with a process. Retail investors do it too, whether they realize it or not: a gut feeling, a post they read, a friend's tip, a chart that looked good. Either way it happens over weeks and months, and one answer doesn't carry it. Professionals build that conviction on terminals that cost tens of thousands of dollars a year per seat and take training to use. Most individual investors have a broker app instead, often one that earns more when they trade more. Their information comes from friends, the news, forums, or, if they're serious, the filings themselves, which are long and dense. Investi is meant to sit in that gap: the depth of a professional terminal, in a form anyone can use. That depth isn't one answer. Fundamental investing isn't set and forget. You keep checking on the business, your reasoning changes, and the thesis needs revisiting every quarter. That work needs somewhere to live. ## Why a workspace and not just a chat? Coding agents give a useful comparison. They keep working files in a scratch folder and then forget them. What lasts is the project and its history. A chat with an agent connected to a data source works like the scratch folder: useful, then gone. Investi's workspace is meant to be the project. It holds your portfolio, your notes and theses, the companies you follow, and [routines](https://tryinvesti.com/blog/an-analyst-on-a-schedule) that rerun a question on a schedule and email you the result. People also work differently. Some open the company, read the filing and ask one question after another. Others would rather be told when something matters. Investi is a workspace with an analyst in it, so you can do the work yourself, hand it off, or mix the two. Two of the things that make an answer checkable are built into that workspace today. A citation opens the filing at the passage or table cell, so you read the number in context ([How the analyst cites a filing](https://tryinvesti.com/blog/how-the-analyst-cites-a-filing) shows how). When the analyst wants to change your notes, it shows the change and writes nothing until you approve it. Both work best when your research and the filings sit in one place you come back to. ## Where does MCP fit today? Investi already uses MCP in one direction: you can [connect a remote MCP server](https://tryinvesti.com/blog/bring-your-own-data) with OAuth, and the analyst uses its tools next to the filings. Investi doesn't offer its own MCP server today. The most useful thing we can build right now is the place where your investing research lives, with the filings and their citations inside it. --- # Introducing the notebook > When a question needs math, the analyst writes and runs Python on filed numbers, prices and your portfolio, and its tables keep a source on each filed figure. Canonical: https://tryinvesti.com/blog/introducing-the-notebook _By The Investi team · Published 2026-10-22 · Product_ Today we're launching the notebook in Investi. When a question needs a calculation, the analyst writes a short Python program, runs it on numbers from the filings, and puts the result in the chat as a table or a chart. Figures that came from a filing keep their source, so you can check a table one cell at a time. Most research questions end in a calculation. What growth rate does today's share price assume? Did gross margin really slip, or did one bad quarter drag the average down? Done by hand, that means copying every number into a spreadsheet first. The notebook does the math in code you can read, on numbers you can trace back to where they were filed. ## What you can ask it to work out You don't open a separate tool or write any code. You ask in chat, and the analyst runs the notebook when a question needs it. A few requests that do: - "Run a reverse DCF on this company at a 9% discount rate." The analyst takes free cash flow from the 10-K and the current share price, then solves for the growth the price implies. - "Chart gross margin for these five chipmakers over their last ten fiscal years." It builds each series from the companies' own filings and draws them on one chart. - "Which of my holdings have moved together over the past three years?" It reads your positions from your connected brokerage, loads daily prices and returns a correlation table. - "Here's my model as an Excel file. Where do my revenue assumptions differ from what the company filed?" It opens the attachment and lines your numbers up against the reported ones. Tables sort by any column and download as CSV or Excel. Charts download as PNG or SVG, along with the data behind them. ## Filed figures keep their source The code works from the same record the analyst reads: SEC filings from 1993 to today, government data such as FRED and BLS series, daily price history, and your portfolio if you've connected one. When the analyst puts a filed number in a table, it attaches the filing to that cell. Hover the number to see the form, the company and the filing date, with a link to the passage or table it came from. A computed figure goes one level deeper. If free cash flow is operating cash flow less capital spending, the cell shows the formula and both inputs, and each input links to its filing. Each table also says how many of its figures are sourced, so a column of your own spreadsheet numbers is easy to tell apart from a column of filed ones. Charts don't carry a source on each point. To check a line on a chart, ask for the table behind it. ## The code is on the page Every run shows up in the conversation as a step with one line describing what it does. Hover that step to see the code and anything it printed. You don't need to read Python to use the result. If you do, you can check the discount rate, the date range or the formula in a few seconds, and if something looks wrong, say so and the analyst reruns it with the change. AI can be wrong, and code can have a bug, so the analyst shows its work for you to check. We explain how that fits the rest of the analyst in [How the Investi analyst works](https://tryinvesti.com/blog/anatomy-of-a-harness). ## Where the code runs Each run happens in its own isolated process on our servers, with no internet connection and no passwords or keys. When the code needs data, it asks Investi, which fetches the data and hands it back. The analyst's code → A request for data → Investi fetches it → Filings, prices, your portfolio, your connected services _The code has no internet access, so every piece of data it uses comes back through Investi._ The code can read your portfolio and no one else's, and it can't write files. If you've [connected a data service](https://tryinvesti.com/blog/bring-your-own-data) of your own, the code can call it through the same route. ## Limits today Each run starts empty, so nothing carries over from one run to the next. The analyst loads the data again each time, which means the code you see is the whole calculation. You can't edit the code and rerun it yourself; to change an assumption, ask: "redo it at 10%." A run stops after two minutes of computing, not counting the time spent fetching data, and has 4 GB of memory. Price history is daily, and quotes are delayed 15 to 20 minutes. Nothing the notebook produces is a recommendation; it does the math you asked for and shows it. ## Get started The notebook is on every plan, including Free. Ask any question that needs a calculation, or tell the analyst to "work it out in Python." To run it on your own holdings, [connect your brokerage](https://tryinvesti.com/blog/connect-your-brokerage) first. The [features page](https://tryinvesti.com/features) shows the rest of the workspace. --- # Connect your brokerage to Investi > Link Robinhood, Fidelity, Schwab, Interactive Brokers and more with a read-only connection, so the analyst works from the holdings and cash you actually have. Canonical: https://tryinvesti.com/blog/connect-your-brokerage _By The Investi team · Published 2026-08-06 · Product_ Today we're launching brokerage connections in Investi. Link your account and your holdings, cash and trade history appear in the workspace next to your notes, so the analyst answers questions about the portfolio you actually have. Brokerage connections are read-only today. Fidelity, Charles Schwab, Robinhood, Interactive Brokers and many other brokerages are supported. Until now, a question about your own portfolio started with you typing out your holdings from memory, a rebalance or two out of date. A watchlist helped, but it has no weights: it doesn't know that one position is a quarter of the portfolio and another is a leftover from 2021. Position size often decides what is worth reading next, and now the workspace can see it. ## What you see once it's connected Every position with its value, today's move and the gain or loss against your cost basis, plus cash and the trade history behind it. On top of that, the numbers a professional would track for a portfolio: - **Returns.** Time-weighted and money-weighted, so you can separate how your picks did from the effect of when you added money. - **Risk.** Volatility, Sharpe ratio, drawdown and value at risk. - **A benchmark.** Beta and the same figures side by side with an index, over the same period. ## The analyst works from your positions Ask "how exposed am I to semiconductors?" and the analyst reads the positions, adds up the weights and shows the math. Ask "which of my theses does today's drop touch?" and it reads the portfolio and your notes together. Ask about an earnings report, and it knows whether this is a company you watch or a company you own. For heavier math, like which holdings move together, it runs the calculation in the [notebook](https://tryinvesti.com/blog/introducing-the-notebook). Your notes still change only when you approve the edit. Connecting a brokerage gives the analyst something to read, not something to act on. ## How the connection works The link runs on [SnapTrade](https://snaptrade.com), a provider of brokerage connections. You sign in on your brokerage's own page, so Investi doesn't see your password. Data flows one way, from the brokerage into your workspace. Your brokerage → SnapTrade, read-only → Holdings, cash and activity → Your portfolio and the analyst _Account data flows one way, from your brokerage into the workspace._ The connection doesn't place trades, move money or change settings in the account. Investi makes money from subscriptions. It does not sell data, take payment for order flow or earn commissions. You can disconnect at any time. Syncing stops, and the connection is deleted along with everything synced through it. Your broker's statements remain the official record, and synced data can lag behind them. ## Get started Open Portfolio in the sidebar and choose your brokerage. The Free plan includes one connection; Plus and Premium include several ([pricing](https://tryinvesti.com/pricing)). For more on data and security, read [Is Investi safe?](https://tryinvesti.com/blog/is-investi-safe) --- # An analyst for everyone > Why Investi exists: serious investment research wasn't beyond ordinary investors, only expensive. AI changes the cost; the product around it decides who benefits. Canonical: https://tryinvesti.com/blog/an-analyst-for-everyone _By The Investi team · Published 2026-08-04 · Updated 2026-10-24 · Company_ Serious investment research has mostly belonged to the people who could pay for it. Funds have analysts, terminals that cost tens of thousands of dollars a seat each year, and clean data. Most individual investors have headlines, tips, and a broker app that earns more when they trade. The work itself wasn't beyond them. It was expensive. That cost is falling fast. A current AI model can read a 10-K, compare a company with its competitors, and reason through whether new evidence supports a thesis. The analyst no longer has to sit inside a bank. We started Investi because the model alone doesn't close the gap. It needs a product around it built for how people actually invest. ## A model alone doesn't know your situation Software went through this first. AI models could write code well before the tools around them were good, because a model on its own didn't know your project or what you were trying to build. The tools that changed who could write software surrounded the model with the project, its history and a way to review changes. That made it useful to beginners as well as professionals. Investing has its own version of the project: a portfolio, a watchlist, a thesis, and a record of what you believed and whether it held up. Investi gives the model that context, ties every number to the filing it came from, and keeps your research between conversations. [How the Investi analyst works](https://tryinvesti.com/blog/anatomy-of-a-harness) explains the mechanics. ## The thesis is the center Investi is built for people who want to understand what they own. That starts with the business: the filings, the unit economics, the balance sheet, the competitive position. Prices and charts help, but they aren't the point. A real investment idea is a chain of reasoning. AI needs chips, chips need power, power needs infrastructure, so the bottleneck may move to energy. The idea becomes a watchlist, the watchlist becomes companies, each company gets a thesis, and each thesis meets evidence quarter after quarter. [How to build an investment thesis](https://tryinvesti.com/blog/how-to-build-an-investment-thesis) walks through that work with a real company. ## No method built in Most finance tools come with a worldview. Screeners encode someone's factors, apps push someone's signals, and some AI investing products are tuned to invest like a famous investor. The method is in the tool before you type anything. Investi takes the other approach on purpose. The analyst starts without an investing method of its own; any bias left comes from the underlying AI model, not from a process we added. You organize the workspace the way you think, with one watchlist or fifty, and folders by sector, theme or conviction. When you want a method, you add it yourself as a [skill](https://tryinvesti.com/blog/teach-an-ai-analyst-your-process): a value checklist, your firm's process, or something entirely your own. When you want other data, you [connect it](https://tryinvesti.com/blog/bring-your-own-data). We build the tool; the opinions are yours. ## Most mistakes are lapses in discipline Good investing is a practice. Notice what might matter, understand why, decide what it means, and keep checking whether the facts still support the reason you bought. None of the steps is hard. Doing them every time is. People forget what they believed. They react to noise, read price moves as information about the business, hold after the thesis breaks and sell before it has been tested. The failure is rarely a lack of intelligence. An analyst helps because it keeps the thread: it reads what you don't have time to read, can look up what you wrote last quarter, and asks whether the reason to own something still holds. Funds have had that for decades. We think every investor should. --- # Introducing routines > Routines run an analyst request on a schedule: every weekday morning, every Friday, or once after earnings. Each run is a full conversation, emailed to you. Canonical: https://tryinvesti.com/blog/an-analyst-on-a-schedule _By The Investi team · Published 2026-07-30 · Product_ Today we're launching routines in Investi. A routine is a question you would ask the analyst, saved with a schedule. It runs whether or not you open the app, and the result lands in your workspace and, if you want, your inbox. Most of the checks that keep a portfolio honest are simple: read the overnight news on what you own, see whether the margin story held after the quarter, reread a thesis when the price does something odd. The hard part is remembering at the right moment, on a day when you're busy with something else. Routines do the remembering. ## Write it once, pick when it runs A routine is the same request you would type into chat: - "Summarize overnight news for my watchlist and flag anything that touches a thesis." Every weekday at 7am. - "Has anything changed in the competitive picture for this company?" Every Friday. - "Read the earnings release and compare it with what I wrote last quarter." Once, the morning after the report. The schedule works like a calendar event: once, daily, every weekday, weekly, every other week, monthly, yearly, or a pattern such as the last Friday of the month. You pick the first run in your own timezone. A routine runs at most once a day. Reference a company, a note or a folder with @, and the analyst reads it before it starts, so a routine that checks a thesis opens the thesis first. ## Every run is a full conversation Your schedule → The analyst reads what you referenced → A new conversation, with sources → An email, if you turned it on _Each scheduled run is a fresh conversation, filed under its routine and optionally emailed._ When a routine runs, it starts a new chat with the analyst's reasoning and sources visible, as if you had asked the question yourself. Turn on email in Routines and each result also arrives in your inbox. Because each run is a conversation, you can continue it. If the Friday check surfaces something odd, ask the follow-up, open the filing or update your note in the same thread. Runs collect under the routine that produced them, with a dot on the ones you haven't opened, so catching up after a week away means reading a list. ## The analyst can propose one You don't have to build routines by hand. In the middle of a conversation, say "check this again every Monday and tell me if anything changed." The analyst drafts the routine and shows it as a proposal, the same way it shows an edit to your notes. Nothing is scheduled until you approve it. After that it is an ordinary routine. Change the wording, move it to another day, pause it during earnings season, or delete it when you sell. ## What routines do today Routines run on a schedule. They don't fire when a price crosses a level or when a new filing arrives. A price alert tells you that something moved; a routine reads the news and your notes and tells you whether it touches what you own. Each run uses your plan's allowance like any other question, and a run is skipped if the allowance has run out. ## Get started Open Routines in the sidebar and write your first one, or ask the analyst to set one up from a conversation. Routines are on every plan. They work with [skills](https://tryinvesti.com/blog/teach-an-ai-analyst-your-process), so a scheduled earnings check can follow your own checklist. The [features page](https://tryinvesti.com/features) shows the rest of the workspace. --- # Bring your own data to the analyst > Connect the data services you already trust, from macro series to crypto markets to your own notes, and the analyst uses them next to the SEC record. Canonical: https://tryinvesti.com/blog/bring-your-own-data _By The Investi team · Published 2026-07-21 · Product_ Today we're launching data connections in Investi. Connect a data service you already use, and the analyst can pull from it while it researches, next to the SEC filings it reads by default. Investors don't all check the same things before they buy. One starts with the balance sheet. One follows macro releases. One watches web traffic and app rankings for an early read on the quarter. What you read is part of how you invest, so Investi lets you choose the sources instead of choosing them for you. ## How it works Connections use MCP, an open standard that lets an AI assistant call another data service. A provider runs one MCP server, and any assistant that supports the standard can use it, which is how Investi can offer many services without a custom integration for each. Once a service is connected, you ask questions as usual. The analyst decides whether a connected service helps with the question, the same way it decides when to open a filing. Each call it makes appears as a step in the conversation. Hover the step to see what the analyst asked for and what came back. Your question → The analyst picks its sources → SEC record, plus your connected services → An answer, with each call shown as a step _A connected service is one more source the analyst can choose, and every call to it shows in the conversation._ ## What you can connect Open Plugins in the sidebar and filter to MCPs. The catalog lists more than a hundred hosted services, grouped by what they cover. A few of the groups: - **Market data and research** from several providers. - **Macro and government data,** such as the IMF, the OECD and the World Bank, on top of the US series Investi already includes. - **Crypto markets and on-chain data.** - **Prediction markets.** - **News and alternative data,** including sentiment and web traffic. - **Your own tools,** such as Notion or Airtable, so the analyst can read notes you keep elsewhere. Some services, such as FactSet or S&P Capital IQ, work only if you already have an account with that provider. If a service isn't in the catalog, paste its server address and connect it yourself. Services that need a login use the provider's own sign-in page; others take an API key. Connections are to servers reachable on the internet. A server that runs only on your own computer has to be hosted at a public address before Investi can reach it. ## The same workspace, different sources - **A fundamentals investor** connects nothing extra and works from the filings, margins and insider trades in the SEC record. - **An investor who also holds crypto** adds crypto market and on-chain data, so one conversation covers both sides of the portfolio. - **A macro-minded investor** adds more rate, inflation and employment series, and asks how a release affects the companies on the watchlist. When a question needs math, the analyst's [notebook](https://tryinvesti.com/blog/introducing-the-notebook) can call your connected services from its code too. ## Get started Open Plugins in the sidebar, pick a service from the catalog or paste a server address, and ask your next question as usual. To have the analyst follow your own method as well, read [Teach an AI analyst your investment process](https://tryinvesti.com/blog/teach-an-ai-analyst-your-process). --- # Teach an AI analyst your investment process > Skills let you write down how you research once, such as your earnings checklist, and the analyst follows it every time the situation comes up. Canonical: https://tryinvesti.com/blog/teach-an-ai-analyst-your-process _By The Investi team · Published 2026-07-09 · Product_ Today we're launching skills in Investi. A skill is your research method written down once, like the checklist you run when a company reports. The analyst follows it whenever the situation comes up, so you stop retyping the same instructions into every chat. Out of the box the analyst starts neutral: it answers the question you ask, the way you ask it. Your process stays in your head: what you check before trusting a number, the order you read a business in, the questions you ask before sizing a position. Skills put that process in the tool, and you choose which ones it uses. ## A skill runs when it fits Say you write an earnings review skill: revenue against the company's own guidance, the margin trend, changes in the risk factors, and the two or three numbers you care about for that kind of business. From then on, when you ask about a company that just reported, the analyst sees that the skill applies and runs your checklist. You don't paste anything. Your question → Each skill's "when to use it" line → The skill that fits → Your checklist, step by step _The analyst matches your question against each skill's 'when to use it' line, then follows the steps of the one that fits._ You can also run any skill by hand: type / in chat and pick it. That works even for skills you keep switched off for everyday questions. ## What's inside a skill A skill is a short text file with three parts: a name, a sentence on when to use it, and the steps, written in plain language. You can attach reference files, such as a scoring rubric or a note template, that the analyst opens when the skill needs them. Name When to use it The steps Reference files, if any _A skill is one short text file, plus any reference files it points to._ The "when to use it" sentence matters most, because it's how the analyst decides the skill applies. The good ones read like an instruction to a colleague: use this when a company reports, use this before I add to a position, use this when I compare two companies in the same industry. ## Three ways to add one 1. **Write it** in the editor. 2. **Create it with Investi**: the analyst asks how you work, then writes the skill for you to review. 3. **Upload it** as a single SKILL.md file, or a .zip or .skill archive of several. Skills use the Agent Skills format, an open format shared by other AI assistants. A skill written elsewhere works in Investi, and the ones you write here are plain text you can take with you. Every skill has an on/off switch. Skills that are on are the ones the analyst picks up by itself; skills that are off stay out of the way until you run them by hand. A large library doesn't make everyday answers noisier. ## Your notes still change only with your approval A skill shapes how the analyst works. It doesn't give it permission to change your research. If a skill leads the analyst to update a note, for example to add last quarter's results to your thesis, the change appears as a before-and-after for you to approve or reject. ## Get started Open Plugins in the sidebar and filter to Skills to write, import or upload your first one. A good first skill is the checklist you already run when a company you own reports. Skills also work in [routines](https://tryinvesti.com/blog/an-analyst-on-a-schedule), so a scheduled check can follow your method. --- # How to build an investment thesis > How to build an investment thesis in seven steps, with a one-page template and a worked Costco example from its 10-K filings, including kill criteria. Canonical: https://tryinvesti.com/blog/how-to-build-an-investment-thesis _By The Investi team · Published 2026-06-30 · Updated 2026-10-24 · Investing_ An investment thesis is a short written argument for owning a stock: what the business is, the two or three things that have to go right, what it's worth against what you pay, and the specific events that would prove you wrong. To build one, start from the business rather than the price, check the story against the company's filings, write your exit conditions before you buy, and fit it on one page. Then revisit it every quarter. This guide walks through seven steps and fills in a one-page thesis for Costco with numbers from its 10-K filings. This is a research framework, not investment advice. Investi is a research and information tool, not a licensed investment adviser, and nothing here is a personal recommendation. **Key takeaways** - A thesis is a falsifiable argument about a business and why the market may be mispricing it, written before you buy. - Most of the work is finding the two or three variables the outcome hinges on and ignoring the rest. - Kill criteria are observable business events, decided in advance. A falling price is not one. - Check the story against the filings: Costco's claim to price near cost shows up as a merchandise margin between 10.5% and 11.3% for a decade. - A thesis only helps if you revisit it. Costco's renewal rate dipped in fiscal 2025 and held in 2026, which is exactly the kind of line to check each year. ## What is an investment thesis? An investment thesis is a falsifiable argument about the future of a business, and about why the market may be mispricing it today. Each part of that definition matters: - It's an argument, not a fact. You are making a claim about an uncertain future, so you say in advance how you would know you were wrong. - It's about a business, not a ticker or a chart. A share is a claim on a company's future cash flows. If you can't describe the business plainly, you don't have a thesis yet. - It says why the market may be wrong. Buying a stock implies you see something not fully priced in: durability it underrates, growth it doubts, a risk it overweights. If the market already agrees with you, the good news is in the price. Writing it down is the point. Memory edits itself to match the price. When a stock falls, an investor without a written thesis tends to find new reasons to hold; when it rises, they forget they ever doubted it. The document keeps you honest with your future self. ## What does a finished thesis answer? A complete thesis answers six questions: 1. What is the business, in a sentence or two a non-expert would understand? 2. How does it make money, and why does that last? 3. What two or three things have to go right? 4. What is it worth, and what am I paying? 5. What would prove me wrong? 6. Why do I see this differently from the market? The seven steps below build each answer in turn. ## Step 1: Start with the business, not the stock The most common mistake is starting from the ticker (the price, the chart, the recent move) and working backward to a reason to buy. Start instead from the plainest description of what the company does. Write one sentence a twelve-year-old would understand. "They sell software that companies use to run payroll." "They own toll roads and collect a fee from every car." If you can't do this without jargon, you don't understand the business well enough to own it yet. That is useful to know. ## Step 2: Understand how the company makes money Now trace the flow of money. Who pays the company, for what, how often, and why do they keep paying? This is where you find durability. A business that has to win every customer again each year is fragile. A business customers can't easily leave, or wouldn't want to, has a moat. Ask: - Where does revenue come from: one product or many, one large customer or thousands? - Is it recurring or one-off? Subscriptions and consumables tend to be steadier than one-time sales. - What stops a competitor from taking it: scale, a network effect, switching costs, a brand, a patent, a license, the lowest costs? - Who has the pricing power: the company, its customers or its suppliers? Name the moat specifically, or admit there isn't one. ## Step 3: Find the two or three things that actually matter Every business has dozens of moving parts, and few of them decide the outcome. The skill is finding those few and ignoring the rest. For a subscription software company it might be net revenue retention, the cost of acquiring a customer, and how long customers stay. For a retailer, same-store sales growth and gross margin. For a commodity producer, the commodity price and its cost per unit. Once you know your drivers, your research has a spine. You stop following every headline and watch the specific numbers that would confirm or break your case. ## Step 4: Check the numbers against the story A thesis is a story, and the financial statements are where you find out whether it's true. You don't need a finance degree. You need to check whether the numbers agree with the argument: - Growth: is revenue growing the way the story implies, and is it speeding up or slowing? - Profitability: are margins stable, or moving the way the story predicts? - Cash: does the business generate cash, or only accounting profit? Free cash flow is harder to flatter than earnings. Heavy capital spending is where the two split, as [what Microsoft, Alphabet, Amazon and Meta spend on capex](https://tryinvesti.com/blog/big-tech-capex-2026) shows. - The balance sheet: how much debt, and could it survive a bad year without raising money on bad terms? - Returns on capital: when the company reinvests a dollar, does it earn a good return on it? Take Costco. Its story is that it sells goods at close to cost and earns its profit on membership fees. If that's true, the margin on merchandise should be thin and steady. The [fiscal 2026 10-K](https://www.sec.gov/Archives/edgar/data/909832/000090983226000093/cost-20260830.htm) reports $297 billion of net sales and $264 billion of merchandise costs, which leaves 11.1 cents of gross profit per dollar of goods sold. Over the decade from fiscal 2016, that margin stayed between 10.5% and 11.3%. **Costco merchandise margin** | Fiscal year | Merchandise margin | | --- | --- | | FY2016 | 11.3% | | FY2017 | 11.3% | | FY2018 | 11.0% | | FY2019 | 11.0% | | FY2020 | 11.2% | | FY2021 | 11.1% | | FY2022 | 10.5% | | FY2023 | 10.6% | | FY2024 | 10.9% | | FY2025 | 11.1% | | FY2026 | 11.1% | Source: 10-K filings, Oct 2018 to Oct 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/909832/000090983226000093/cost-20260830.htm)) _Costco's merchandise margin stayed between 10.5% and 11.3% from fiscal 2016 to 2026: thin, and steady._ The numbers corroborate the story. When they contradict it, the numbers are usually right. Ground each claim in a figure you can pull and point to, from the latest filing: a number recalled from memory may be a year old. ## Step 5: Decide what it is worth, and what you are paying A wonderful business at the wrong price can still be a bad investment. Valuation connects the story to a number. You don't need a fifteen-tab model. You need a defensible view of what the business is worth and an honest look at the gap to today's price. Anchor it in something real: what are you paying for each dollar of earnings or free cash flow, and is that reasonable given how fast and how durably those earnings grow? Then write the gap down: "I think this is worth roughly X. It trades at Y. The difference is my margin of safety." If there's no gap, and you'd only make money if someone later pays even more, you have a momentum bet rather than a value thesis. That can be a strategy, but be clear about which one you're running. ## Step 6: Write down what would prove you wrong This is the step most people skip. Before you buy, write the specific, observable events that would tell you the thesis is broken. A falling price isn't one of them. The real triggers are things like: growth falls below a level for two straight quarters, the moat visibly erodes, margins fall, a key driver reverses, or management does something that contradicts the reason you trusted it. These are your kill criteria. They make the sell decision in advance, in a calm moment, so that when the moment comes, noisy and emotional, you follow a rule you already set instead of negotiating with yourself. ## Step 7: Fit it on one page Put the answers on a single page. If they don't fit, the weak parts of the argument are probably hiding in the length. Copy this template, keep the headings, and fill the blanks. A line that stays empty is the part you haven't thought through. **Company.** **Date.** **The business, in a sentence.** **How it makes money, and why that lasts.** **The two or three drivers I am betting on.** **What it is worth versus what I am paying.** **Kill criteria.** The events that would make me sell. Price is not one of them. **My edge.** Why I see this differently from the market. ## What does a finished thesis look like? This is a filled page for Costco, written in October 2026 after its fiscal 2026 10-K. It is not a recommendation to buy or sell. It shows what the page looks like when someone has finished thinking. **Company.** Costco Wholesale (COST) **Date.** October 2026, after the 10-K filed October 7, 2026. **The business, in a sentence.** A membership warehouse: people pay each year for the right to buy goods in bulk, near cost. **How it makes money, and why that lasts.** Merchandise is priced to move: about 11 cents of gross profit per dollar of sales, as Step 4 showed. The fee is where the profit concentrates. Membership fees were $5.91 billion in fiscal 2026, about half of the $11.7 billion operating income. Scale buying keeps prices low, low prices keep members renewing, and renewals keep the scale. That loop is the moat. **Costco membership fees and operating income, fiscal 2016 to 2026** | Fiscal year | Membership fees | Operating income | | --- | --- | --- | | FY2016 | $2.65B | $3.67B | | FY2017 | $2.85B | $4.11B | | FY2018 | $3.14B | $4.48B | | FY2019 | $3.35B | $4.74B | | FY2020 | $3.54B | $5.44B | | FY2021 | $3.88B | $6.71B | | FY2022 | $4.22B | $7.79B | | FY2023 | $4.58B | $8.11B | | FY2024 | $4.83B | $9.29B | | FY2025 | $5.32B | $10.4B | | FY2026 | $5.91B | $11.7B | Source: 10-K filings, Oct 2018 to Oct 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/909832/000090983226000093/cost-20260830.htm)) _Fees rose every year for a decade. They still equal about half of operating income, down from 72% in fiscal 2016._ **The two or three drivers I am betting on.** The renewal rate and membership growth. Comparable sales in the core warehouse, excluding gas. Room to raise the fee without hurting renewals. **What it is worth versus what I am paying.** The question is what you pay for a fee stream that has grown every year for a decade. If the price only makes sense for a high-quality company at a full multiple, there may be no gap. Then you don't have a value thesis; you have a quality company at a full price, and those are different things. **Kill criteria.** The renewal rate falls two years in a row. A fee increase that costs members and doesn't win them back. Comparable sales excluding gas turn negative for several quarters. Costco starts chasing merchandise margin the way a typical retailer would, which would break the loop. **My edge.** I shop there, and I can see how a household plans around the card. That's a small edge, not a secret model. If the only thing I can write here is that everyone knows Costco is well run, I don't have an edge. I have the consensus. The page doesn't say buy. It says what has to stay true, and what would end the argument. ## What does checking a kill criterion look like? Kill criteria only help if you check them. Costco reports its renewal rate in Item 1 of each 10-K. Worldwide, it was 90.5% at the end of fiscal 2024 ([10-K](https://www.sec.gov/Archives/edgar/data/909832/000090983224000049/cost-20240901.htm)), 89.8% in fiscal 2025 ([10-K](https://www.sec.gov/Archives/edgar/data/909832/000090983225000101/cost-20250831.htm)) and 89.8% in fiscal 2026 ([10-K](https://www.sec.gov/Archives/edgar/data/909832/000090983226000093/cost-20260830.htm)). In the U.S. and Canada it went from 92.9% to 92.3%, then stayed at 92.3%. **Costco member renewal rate** | Fiscal year | U.S. and Canada | Worldwide | | --- | --- | --- | | FY2016 | 90.0% | 88.0% | | FY2017 | 90.0% | 87.0% | | FY2018 | 90.0% | 88.0% | | FY2019 | 91.0% | 88.0% | | FY2020 | 91.0% | 88.0% | | FY2021 | 91.0% | 89.0% | | FY2022 | 93.0% | 90.0% | | FY2023 | 92.7% | 90.4% | | FY2024 | 92.9% | 90.5% | | FY2025 | 92.3% | 89.8% | | FY2026 | 92.3% | 89.8% | Source: 10-K filings, Oct 2016 to Oct 2026 · SEC EDGAR via Investi ([latest filing](https://www.sec.gov/Archives/edgar/data/909832/000090983226000093/cost-20260830.htm)) _Worldwide renewals slipped from 90.5% to 89.8% in fiscal 2025 and held there in fiscal 2026: one decline, not two._ Against the criterion "the renewal rate falls two years in a row", that's one decline followed by a flat year, so it hasn't triggered. It is the line to read first in the next 10-K. Before comparing years, read how Costco defines the rate in the same filing: it describes it as a trailing calculation, and a change in definition can move the number without a change in the business. ## Where do most theses fail? The same patterns show up again and again: - A story with no falsifier. If nothing could prove it wrong, it's faith, not analysis. - Confusing a good company with a good investment. The price is half the thesis. - Anchoring to the purchase price. The only question is whether the thesis is still true from here. - Researching endlessly to avoid the decision. The drivers are few; find them and decide. - Never revisiting it, which is the most common failure of all. ## How do you keep a thesis alive after you buy? Writing a thesis is the easier half. Every quarter brings new evidence: results that confirm or contradict your drivers, competitive moves that test the moat, decisions by management. The habit that matters is checking the thesis against that evidence on a schedule. Written thesis → New 10-Q or 10-K → Check drivers and kill criteria → Hold, revise or sell _Each new filing is a chance to test the thesis against its kill criteria._ Few investors do this, often because their notes live in a document that never gets reopened. [Investi](https://tryinvesti.com/features) keeps your thesis next to the company's filings. The analyst can open what you wrote last quarter and check it against the latest numbers, citing each one to its filing, and when it helps revise your notes it proposes the edit for you to approve. A [routine](https://tryinvesti.com/blog/an-analyst-on-a-schedule) can rerun a question like "check my Costco kill criteria against the latest filings" on a schedule and email you the result. [How the Investi analyst works](https://tryinvesti.com/blog/anatomy-of-a-harness) explains the design. The tool is secondary, though. A thesis you wrote down, kill criteria you set in advance, and a standing appointment to check both are most of the discipline, and they work in a paper notebook too. ## How can you check this yourself? Every Costco number in this guide comes from its 10-K filings on SEC EDGAR. To check them, search for Costco (ticker COST) on [EDGAR](https://www.sec.gov/edgar/search/), open the 10-K filed October 7, 2026, and look at three places: 1. The income statement, for net sales, membership fees, merchandise costs and operating income. 2. Item 1, Business, for the renewal rate sentence and how the rate is calculated. 3. The earlier 10-Ks, for the same lines in past years. Or ask Investi the question and open the citations: each figure links to the line of the filing it came from. **How long should an investment thesis be?** One page. It should state the business in a sentence, how it makes money and why that lasts, the two or three drivers you are betting on, what it is worth against what you pay, your kill criteria and your edge. If it runs longer, the weak parts of the argument are usually hiding in the length. Supporting research can live elsewhere. **What are kill criteria in an investment thesis?** Kill criteria are the specific, observable events that would tell you the thesis is broken, written before you buy. Examples are a key metric falling for two years in a row, margins falling, or management contradicting the reason you trusted it. A falling share price is not one: it tells you what the market thinks, not whether your argument about the business still holds. **How often should you review an investment thesis?** At least each time the company files a quarterly or annual report, because that is when new evidence on your drivers arrives. Check each kill criterion against the new numbers, note what you got right and wrong, and update the page. A thesis you never reopen tends to drift toward whatever the price is doing. **Where do I find the numbers for a thesis?** In the company's own filings with the SEC: the annual 10-K and quarterly 10-Q, free on EDGAR. The financial statements give revenue, margins, cash flow and debt, and the business section often reports operating metrics such as Costco's renewal rate. Use the latest filing, and note which one each number came from so you can check it later. The fastest way to learn this is to write one. Pick a company you already understand, fill in the page, and force it onto one sheet. If you want to keep it in Investi, [sign in](https://tryinvesti.com/signin), add the company, and ask the analyst to check your drivers against the latest filings. --- # How the Investi analyst works > What makes an AI analyst useful is what surrounds the model: what it sees, where its numbers come from, what it remembers, and what it may change. Canonical: https://tryinvesti.com/blog/anatomy-of-a-harness _By The Investi team · Published 2026-06-27 · Updated 2026-10-24 · Product_ Anyone can ask a general chatbot about a stock, and the model behind it can read an annual report with real competence. Ask it what your position is worth, how last quarter's margin compares with what you wrote in July, or whether a new filing breaks your thesis, and it has nothing to work with. It doesn't know what you own, what you believe, or what you concluded last quarter. What changes that is the layer around the model. We call it the harness, and it has four parts: what the model sees, where its numbers come from, what it remembers, and what it is allowed to change. This post explains each one. ## Choosing what the model sees A model only knows what is in front of it for that conversation. Anything else might as well not exist. That's why the same model gives a generic answer in a chat window and a specific one inside a product. The same news also means different things if you own the company, watch it, or own its largest competitor. So with every question, Investi gives the model a map of your workspace: your folders, the companies in them, and the titles of your notes, plus whichever company or note you have open. A note you mention with @ is included in full. Everything else, the analyst opens itself as it works: it can read or search your notes, look up your portfolio, and search the filings. - Your question → Workspace map and open tab → The model - Notes and portfolio, read as needed → Filings, searched and opened → Answer with citations _Each question starts with a map of your workspace, and the analyst opens what it needs from there._ The map is small on purpose. Giving a model everything tends to make answers worse, because it has to find the relevant part in a pile of noise. Note titles are cheap enough to send every time, so the analyst knows what exists; it reads the full text only of what the question needs. ## Getting numbers from the filings A model's training data is a snapshot of the past. Ask one for a company's operating margin and you get a number that sounds right. It may be a year old or slightly wrong, and it will be stated with the same confidence either way. Investi's analyst looks the numbers up instead. It searches and reads the SEC record, every filing from 1993 to today, as it works. When it cites revenue growth, the figure came from a filing it opened during that answer, and the citation opens that filing at the passage or table cell. When it calculates something, like trailing twelve months or a growth rate, it shows the formula. [How the analyst cites a filing](https://tryinvesti.com/blog/how-the-analyst-cites-a-filing) walks through what you see, and the [methodology page](https://tryinvesti.com/methodology) has the details. The work is visible too: which filings it opened and which tools it called on the way to the answer. That lets you catch the model being wrong, which matters more than any claim that it won't be. ## Keeping research between conversations Most investors keep their thesis in their head, where it changes without them noticing. Conviction drifts to match the price, and the original reason for owning something fades. In Investi, your notes on a company sit next to its filings. Because the titles are in the analyst's map, when you ask about a company in October it can see the thesis you wrote in July and open it. The research builds up instead of disappearing when a chat closes. ## Proposing changes instead of making them The analyst doesn't change your workspace on its own. When you ask it to update a note, for example to add a new quarter or revise a risk section, it shows the change as a before-and-after. The same goes for creating folders, routines or skills. Nothing is written until you approve it. The analyst proposes an edit → You read the before-and-after → Applied only if you approve _Every change the analyst proposes waits for your approval._ Your notes are the record of your judgment. They're only worth keeping if every word passed through you. ## What better models change A better model that doesn't know your situation still has to guess about it. Inside the harness, a better model reads your notes more carefully, uses the filings better, and catches contradictions you'd miss. We expect each improvement in models to make this layer more useful, not less. What the harness doesn't do is decide how you invest. The analyst starts without a method of its own; [skills](https://tryinvesti.com/blog/teach-an-ai-analyst-your-process) and [data connections](https://tryinvesti.com/blog/bring-your-own-data) are how you add yours. ## What it doesn't do yet The analyst reads a note's full text only when it opens it or you mention it, so mentioning the notes that matter is the surest way to get them into an answer. Routines rerun a question on a schedule, at most daily; they don't fire when a filing lands. And the analyst can still be wrong, which is why its numbers point back to their sources. For why we think this layer is worth building, read [An analyst for everyone](https://tryinvesti.com/blog/an-analyst-for-everyone).