The new book from Rob May
Investing in AI
Six Mental Models to Tell Moats from Traps
The same AI product can be a moat, a trap, or a lottery ticket. The difference is not the technology. It is the economics.
The SaaS playbook assumes smooth adoption curves, tidy unit economics, and margins that improve as you scale. AI companies break those assumptions. Every prompt can carry a cost, usage swings wildly, and a product that dazzles in a demo can still be a bad business.
Investing in AI replaces that playbook with six mental models for spotting businesses built to last, even when the technology shifts beneath them. It draws on ten years and more than 135 investments in AI at HalfCourt Capital.
Published September 15, 2026 by Bat House Press. Foreword by Karim R. Lakhani of Harvard Business School.
Also at Bookshop.org, Apple Books, Kobo, Books-A-Million.
What readers are saying
“Rob is the rare investor, founder, and operator that saw the rise of AI, knew the right players to invest in, and saw where the future was headed. As an avid reader of his newsletter, I read it like the bible of the future of AI.”
“Understanding AI isn't just about following the latest breakthroughs. It's about recognizing the larger forces reshaping business, markets, and society. Rob brings that perspective to everything he does.”
“If you want to know what is really going on now and what will happen in AI, you should read this book. Rob May has been writing about AI for longer than most people have been aware of the concept. I have found his writing and insights to be accurate, incisive, and at times, prescient.”
Everyone agrees AI is real. Almost no one knows how to invest in it.
In 2015, Rob May walked into investor meetings asking for money to spend the next thirty years investing in artificial intelligence. Most of the people in the room still thought of AI as a punchline from the 1980s. Seven years later ChatGPT reached a hundred million users in two months, and the conversation flipped: investors no longer needed convincing that AI mattered. They just had no idea how to do it.
Most responded the way they did to the dot-com and SaaS booms. They made wide, scattered bets. They looked for companies that resembled the big players and assumed similar metrics. They anchored on early growth and hype instead of asking whether costs, data, and competitive position could sustain it. Those habits worked for decades. In AI they fail, because AI companies see nonlinear adoption and abandonment unlike anything in traditional software. Jasper raised at a $1.5 billion valuation as a friendly wrapper on GPT-3, then watched customers go straight to ChatGPT. DeepSeek repriced a whole layer of "AI infrastructure" in a matter of days.
This book is written for a general investing audience: angels, LPs, public-market investors, and operators who allocate capital. You do not need a technical background. You need a better set of rails for your judgment.
The six mental models
Each model has its own chapter, grounded in deals Rob has seen, and each ends with concrete investor actions you can use on your next decision.
Mental model 1
AI is a collection of technologies
Neural nets, Bayesian methods, symbolic systems, and evolutionary algorithms have different cost structures, moats, and risks. Treating AI as one bucket hides the differences that matter. Invest in defensibility, not in a single paradigm.
AskWhat kind of AI is this, and how hard is it to replicate?
Mental model 2
AI challenges traditional SaaS assumptions
Per-seat pricing, zero marginal cost, smooth ARR curves, and automatic land-and-expand break down when software does work on your behalf. The mistake is not using SaaS instincts. It is using them automatically.
AskWhat work is this product actually doing, and how does pricing map to it?
Mental model 3
Expect major market dislocations
Most markets move slowly. AI does not. A single technical result can force the market to reprice an entire layer overnight. Assume fast repricings and sudden business-model breaks are normal, and position to benefit from shocks.
AskWhich three to five assumptions have to hold for this investment to work?
Mental model 4
AI changes markets and corporate strategy
When creation gets cheap, choke points move. Like the internet shifting newspaper profits to discovery, AI pushes moats toward coordination and control points. The winners sit at the new choke point, not the best AI story.
AskWhat becomes scarce in this value chain as AI makes the work abundant?
Mental model 5
AI companies can scale faster than before
Speed is real, but it is a weak signal unless you know what causes it, what it costs, and what will persist. Revenue quality beats revenue quantity: production usage is not a proof of concept.
AskHow much of this revenue is production, and how much is pilots?
Mental model 6
Be careful evaluating AI COGS
AI margins can improve dramatically as human labor comes out, or deteriorate as inference bites. Some companies reach software-like margins; others stay closer to services. Underwrite the trajectory, not today's margin.
AskWhy will gross margin be higher in three years, and what could stop it?
18 questions · about 5 minutes
Which of the six models matters most for you?
Take the 18-question AI Investor Assessment. It places you in one of 30 profiles by how far along you are, how you behave when markets move, and which part of the AI stack suits you, then points you to the mental models to focus on first.
Take the assessmentPrototype
The AI Signal Gauge
Is this company’s AI earning its cost, and does what it says match what it files? A filings-only rating on a five-step scale, from Talking to Compounding, with “insufficient disclosure” as an honest state of its own.
It reads only what companies file with the SEC: the 10-K, the 10-Q and the quarterly earnings release. Software companies are compared with software companies on whether AI is eroding their gross margin; chip makers, cloud providers and data-centre operators are compared with each other on whether revenue follows the build and the accounting holds up. The demo walks through six fictional companies, and the real-evidence pages show what it reads in Nvidia’s, Microsoft’s and four other companies’ filings, quoted and linked.
See the demoFrom the foreword
Rob is one of the few people in this field who thinks from first principles rather than from the prevailing narrative.
When the change is one of kind, the comfortable move, reaching for the comparison that worked last time, is precisely the move that leads you astray. You need a new mental model. That is exactly what Rob provides.
Beyond the six models: moats, traps, and lottery tickets
The second half of the book turns the models into decisions. The triad of AI value sorts any AI story into three buckets. Commoditized AI is table stakes that everyone deploys to stay in the game, and a trap if you overpay for it. Core AI makes a company better at what it already does best, a compounding engine if you buy it at the right price. Expansion AI is hidden optionality, new business lines born from data exhaust, and a lottery ticket the market often misprices.
From there Rob covers ways to invest around AI without picking individual winners (sell what AI consumes, own the new scarcity, price the layer), how to evaluate traditional companies adopting AI, how to pair a thesis with your mandate, and where value actually accrued in the history of large language models.
What's inside
- 1Why Investing in AI Requires New Mental Models
- 2AI Isn't Solved Yet (We're Only in the Second Inning)
- 3Mental Model #1: AI Is a Collection of Technologies
- 4Mental Model #2: AI Challenges Traditional SaaS Assumptions
- 5Mental Model #3: Expect Major Market Dislocations
- 6Mental Model #4: AI Changes Markets and Corporate Strategy
- 7Mental Model #5: AI Companies Can Scale Faster Than Before
- 8Mental Model #6: Be Careful Evaluating AI COGS
- 9The Triad of AI Value
- 10Ways to Invest Around AI
- 11Evaluating Traditional Companies Adopting AI
- 12Pairing Your Investment Thesis with Markets
- 13The History of Large Language Models: Where Value Actually Accrues
Foreword by Karim R. Lakhani, introduction, conclusion, and acknowledgments. 243 pages.
About the author
Rob May
Rob May has started five technology companies over the past twenty years, including Neurometric, the AI inference infrastructure company he runs today, and Backupify. He is also a cofounder and general partner of HalfCourt Capital, which has invested exclusively in AI since 2015 and has made more than 135 investments.
Rob cohosts the AI in NYC podcast and writes the Investing in AI newsletter on Substack. He holds a BS in electrical engineering and an MBA from the University of Kentucky, and lives in New York City.
Read Rob every week
The book grew out of years of the Investing in AI newsletter. Subscribe for Rob's weekly take on where value is moving in AI, including chapter excerpts and the ideas that did not fit in the book.
Subscribe on SubstackQuestions about the book
Who is Investing in AI written for?
A general investing audience: angel investors, LPs, fund managers, public-market investors, and operators who allocate capital. It assumes you know how to evaluate a business, not how to build a model. Founders can also read it in reverse, as a guide to how a sharp investor will diligence their company.
What are the six mental models?
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 COGS. Each has its own chapter with investor actions at the end.
What is the difference between a moat and a trap?
The book's triad of AI value sorts AI into three buckets. Commoditized AI is table stakes and becomes a trap if you overpay for it. Core AI strengthens what a company already does best and compounds if bought at the right price. Expansion AI creates new business lines and behaves like a lottery ticket the market often misprices.
Can I read part of the book before buying?
Yes. Rob has published several chapters on his Investing in AI Substack, including chapter one and the chapter on market dislocations.
Was AI used to write the book?
Yes, deliberately. Rob and his editor first had Claude draft a chapter, found it competent but strategically empty, and changed approach: AI helped outline, reorder arguments, and turn notes into drafts, and Rob rewrote everything in his own voice. The conclusion argues investors should use AI the same way, to sharpen judgment rather than replace it.
What formats is it available in?
Kindle, hardcover, and paperback on Amazon; hardcover, paperback, and NOOK ebook at Barnes & Noble; and through Bookshop.org, Apple Books, Kobo, and Books-A-Million. The book is 243 pages and was published on September 15, 2026 by Bat House Press.