About
Media
Startups & Experiments
Writing
Topics
Resources
Contact
Posts
InvestingOctober 30, 2026 / 10 min

Why Rob May and I Built Investing in AI, and Why It Reads Only SEC Filings

Rob May and I built Backupify together. Now we've co-founded Investing in AI: the home of Rob's book and newsletter, plus research on the AI build-out that uses nothing but what companies file with the SEC. Here is why we built it and how it works.

VCVik ChadhaFounder • Operator • Investor
Illustration of a magnifying glass over a stack of filings, with layered blocks representing the AI supply chain
Part of the Venture Studio series
Explore topic
First-hand Experience
Co-founder of Investing in AI with Rob May(Previously co-founded Backupify with Rob (acquired by Datto); founder and CEO of Scalable Ventures since 2010)
Investing in AI is the newest company I've co-founded, and the first one I've built with Rob May since Backupify. It is the home of Rob's book, Investing in AI: Six Mental Models to Tell Moats from Traps, and of his weekly newsletter. It also publishes research on the AI build-out that follows one rule: every figure comes from what a company files with the SEC, and every claim can be traced back to the sentence in the filing it came from. This post explains why we built it, what is on the site today, and the working rules that shape everything we publish. If you want the site's own account, the about page has it. Here I want to give the view from my side of the partnership.

Rob and I Have Done This Before

Rob and I co-founded Backupify, the cloud-backup company that Datto acquired. After that we went in different directions. Rob spent the next decade investing only in AI. He co-founded HalfCourt Capital, which has backed AI companies exclusively since 2015, founded Talla, and now runs Neurometric, which works on inference optimization for multi-model AI systems. His book draws on more than 135 AI investments. I spent the same decade building software companies and investing in others, mostly through Scalable Ventures, the venture studio I've run since 2010 (I wrote about why I chose the studio model separately). Over the last few years nearly everything we build at the studio has AI at its core, and I've written about the AI tools that now run parts of my companies. Put those together and you get two people who have seen AI from three sides: the founder's, the operator's and the investor's. When we compared notes, we kept landing on the same problem.

The Gap We Kept Seeing

Everyone now agrees AI matters. What hasn't caught up is how people judge AI companies. Most investors still carry habits from the dot-com and SaaS eras: spread lots of small bets, compare a company to whoever looks similar, and take early growth at face value. Those habits break down in AI for reasons Rob lays out in the book:
  • Adoption and abandonment both move faster. A product can win a market in a year and lose it the next, because switching from one model or tool to another is often cheap.
  • Inference costs can eat margins. In classic SaaS, serving one more customer cost almost nothing. In AI, every query has a compute bill, so gross margins look different and can change quickly.
  • A single technical result can reprice a whole layer. A new model, a cheaper chip or a better training method can change who has pricing power from one quarter to the next.
The second point matters to anyone building AI products: the cost of running them has to be part of pricing from day one, in a way it never was for traditional software. And on the investing side, companies routinely describe AI with more confidence than their numbers support. The story a company tells about AI and the economics it reports are not always the same thing. That gap is what Investing in AI is built to measure.

What the Book Gives You: Six Mental Models

Rob's book is the framework. It doesn't try to teach the technology. It gives a general investing audience six mental models for telling a durable AI advantage from a trap:
  1. AI is a collection of technologies, not one thing.
  2. AI challenges traditional SaaS assumptions.
  3. Expect major market dislocations.
  4. AI changes markets and corporate strategy.
  5. AI companies can scale faster than before.
  6. Be careful evaluating AI cost of goods sold.
The same AI product can be a moat for one company, a trap for another and a lottery ticket for a third. The models are a way of asking the right questions before deciding which. The newsletter keeps them current as the market moves; much of what ended up in the book appeared there first. If you want to see which models matter most for how you already invest, the site has an eighteen-question AI Investor Assessment that places you in one of 30 investor profiles and points you to the models to start with.

Why Filings, and Only Filings

The book gives you the questions. The research side of the site tries to answer them for public companies, and we made an early decision that turned out to shape everything else: we read only what companies file with the SEC. That means 10-Ks, 10-Qs, earnings releases filed as 8-Ks and, for foreign companies such as TSMC and ASML, their annual 20-F reports. We made that choice for three reasons. Filings are the one place a company has legal exposure for what it says. Press releases, conference talks and earnings-call commentary can be as optimistic as management likes. A 10-K is signed, audited where it counts, and subject to securities law. When a filing says a customer accounted for a certain share of revenue, or that a lease hasn't started yet, that statement carries weight a keynote doesn't. Filings can be checked by anyone. Every reading on the site links to the filing it came from and quotes the sentence that supports it. If a reader thinks we got something wrong, they can open the source and see for themselves. That is a higher bar than "according to our analysis," and it is a bar I'd want from anyone whose research I rely on. Filings are structured enough to refresh. Much of the financial data in SEC filings is tagged in XBRL, which means the numbers can be pulled, recomputed and updated after every quarter without retyping anything. A page written once and left alone goes stale. A page computed from the filings stays true after the next quarter, as long as the copy around it is checked too. The cost of this rule is that we can't say things filings don't say. The big AI labs are private, so our pages on who funds OpenAI and who funds Anthropic show only what listed companies' filings state about them, and they say so plainly. Some stakes reported in the press don't appear in any filing, so they don't appear on the site. I think that restraint is the point.

What's on the Site Today

The AI stack map

The centerpiece of the research is the AI stack map: who gets paid when AI gets built. It splits the companies supplying the AI build-out into eight layers, ordered by the way the money flows:
  1. AI clouds: the buyers
  2. Servers and systems
  3. Chips: accelerators and custom silicon
  4. Memory and storage
  5. Networking and optics
  6. Power, cooling and data centers
  7. Foundry, equipment and design tools
  8. Software and applications: the return
Each layer has its own page with the companies in it, their revenue and growth, and how much of the money flowing in they keep as gross profit. Deep dives follow the capex money from the cloud buyers down the chain and map the circular financing between suppliers and customers, where a vendor invests in or lends to the company that buys from it. There is also an explainer on whether AI is a bubble that tracks a set of indicators against a year earlier, with no forecast attached. The software layer is the one I care most about as a builder. It is where the build-out is supposed to pay off, and it is where the gap between AI claims and reported economics is easiest to see.

Company pages

There are now company pages for more than fifty companies: the most-watched AI names, Nvidia, Microsoft and TSMC among them, and every company in the software layer. Each one combines a written profile with sections generated from its own filings: quarterly trends, segment results, how it compares with its layer, customer concentration, financing links, backlog and commitments, and when the next filing is due. A company joins the coverage only if it files with the SEC and AI is material to its revenue.

Tools, reports and data

The site has a capex tracker and two calculators, one for how long AI capex takes to pay back and one for how depreciation choices change reported profit. Where the calculations need an assumption the filings don't provide, the assumption is a slider and it is labelled as one. After the five largest cloud buyers have all reported a quarter, we publish a quarterly AI Stack Report: who spent, who got paid, who kept it and what changed. Each report is frozen the day it goes out. Live pages update with the data; reports don't, so what we said stays visible next to what happened. For journalists and researchers there is a press and data room with citable figures, charts that can be embedded, and CSV downloads, free for editorial use with credit. There is also a glossary that explains the terms that show up in AI companies' filings, from hyperscalers and neoclouds to high-bandwidth memory and remaining performance obligations.

The AI Signal Gauge

The longest-running project is the AI Signal Gauge, a prototype that asks of each company the questions the book asks and answers them from filings: is its AI earning its cost, and does what it says about AI match what it reports? The public gauge pages walk through a demo of six fictional companies, with tickers that belong to no real listing, so a demonstration rating can never be mistaken for a view on a real stock. Separate evidence pages show what the gauge reads in real companies' filings, quoted and linked, without any rating attached.

The Rules We Work By

Building a research product with a co-founder means agreeing on rules before the first disagreement. Ours are on the site, and these are the ones that matter most to me. Economics over excitement. We care about what an AI product costs to run, who pays for it and what becomes scarce, not how good the demo is. Primary sources, quoted. If we can't point to the sentence in a filing, we don't make the claim. Test before we trust. A signal earns a place in the gauge only after it has been written down in advance and tested against history. Most ideas fail that test. We say so rather than quietly dropping them. It is the same discipline a venture studio needs, where most ideas should die before they become companies. "Not disclosed" is an answer. When a company doesn't report enough to judge, we say so instead of guessing. Missing evidence is information too. Corrections in public. When we get something wrong, we fix it and log the correction on our editorial standards page. Not investment advice. Nothing on the site tells anyone to buy or sell anything. The book and the research are tools for sharper judgment. The decisions stay with the reader.

How It Fits With the Rest of My Work

People who follow my work sometimes ask how a research site fits with a venture studio. For me it is the same question asked at a different scale. At Scalable Ventures I'm deciding whether an AI product will earn more than it costs before we build it. At Investing in AI we ask whether public companies' AI is earning more than it costs after they've built it. The two feed each other: what the filings show about AI margins is useful when pricing our own products, and the operating view is useful when reading filings skeptically. It also connects to the other new project on my startups page: I'm a co-founder of AI in NYC, the podcast, newsletter and closed-door dinner series for people building and backing AI companies in New York, which Rob co-hosts. The conversations there are a good test of whether what we read in filings matches what founders and investors see on the ground.

Where It Goes Next

Investing in AI is new. The stack data refreshes after every quarter's filings, new company pages are added when they clear the coverage rule, and each quarterly report adds to a public record we can't revise later. Rob's newsletter on Substack is the best way to follow along, and the site's work with us page covers briefings for boards, leadership teams and investors. If you've read my posts on build vs. buy for AI or AI due diligence for startups, Investing in AI is the public-markets version of the same habit: look past the AI story to the AI economics, and show your sources.

Frequently asked questions

What is Investing in AI?

Investing in AI (investinginai.pro) is the home of Rob May's book, Investing in AI: Six Mental Models to Tell Moats from Traps, and his weekly newsletter. It also publishes research on the AI build-out drawn only from SEC filings: an AI stack map in eight layers, company pages, quarterly AI Stack Reports, a press and data room, and the AI Signal Gauge prototype. Rob May and Vik Chadha are co-founders.

Why does Investing in AI use only SEC filings?

Filings are where a company has legal exposure for what it says, anyone can check them, and their XBRL-tagged figures can be recomputed after every quarter. Every reading on the site links to its filing and quotes the supporting sentence. The trade-off is that the site can't state what filings don't, so its pages on private labs such as OpenAI and Anthropic show only what listed companies disclose.

What are the six mental models in Rob May's book?

AI is a collection of technologies; AI challenges traditional SaaS assumptions; expect major market dislocations; AI changes markets and corporate strategy; AI companies can scale faster than before; and be careful evaluating AI cost of goods sold. Together they help a general investing audience tell a durable AI advantage from a trap.

Is Investing in AI investment advice?

No. Nothing on the site tells anyone to buy or sell a security. The book and the research are tools for sharper judgment, and the gauge's demo uses fictional companies so a demonstration rating is never mistaken for a view on a real stock.
Scalable Ventures logo
Flagship VentureFounder & CEO of Scalable Ventures — The AI-focused studio where I build, fund, and scale capital-efficient B2B SaaS.
Visit Scalable Ventures

Related Articles

Explore more insights on entrepreneurship, AI, and leadership:

Explore More

Dive deeper into related topics and resources:
On this page