AI Stocks June 07, 2026
Three years into the generative AI boom, the market is finally starting to ask the question that matters: who is actually making money from this? For investors, the answer separates durable compounders from melting ice cubes wearing AI badges.
The Idea
Every technology revolution produces two kinds of winners on the way up — companies building real economic moats, and companies riding narrative momentum. Both can deliver outsized returns. But only one survives the cleanup phase. The strategy here is simple: own AI businesses whose profitability is structurally improving, and treat the rest as trades, not investments.
How It Works
Moat-building AI companies share a handful of identifiable traits. They turn AI capability into a recurring revenue stream with expanding gross margins, not just rising top-line growth. They benefit from data flywheels — every customer interaction makes the product harder to replicate. And critically, their return on invested capital (ROIC) is rising, not falling, even as capex balloons.
Momentum plays, by contrast, tend to show the opposite pattern: revenue growth funded by stock-based compensation, gross margins compressing under cloud compute costs, and a narrative that depends on the next product demo rather than the last earnings report.
A working checklist:
- Unit economics: Are gross margins on AI products above 60%, and trending up?
- Customer lock-in: Net revenue retention above 115%? Switching costs measurable?
- Capital discipline: Is incremental capex generating incremental free cash flow within a reasonable horizon?
- Pricing power: Can the company raise prices without losing customers, the way Buffett described as the single most important business test?
- Proprietary data or distribution: Something competitors cannot replicate by writing a check to a hyperscaler.
Why This Has Worked Historically
This framework isn’t new — it’s the same lens Warren Buffett applied to newspapers in the 1970s, Peter Lynch applied to retail in the 1980s, and Aswath Damodaran has applied to every tech cycle since. In The Little Book That Builds Wealth, Pat Dorsey laid out the four classic moats: intangible assets, switching costs, network effects, and cost advantages. AI doesn’t change those categories — it just creates new ways to build them.
The dot-com era is the cleanest historical analogue. Between 1998 and 2000, hundreds of companies with “.com” in their names rallied alongside Cisco, Amazon, and Microsoft. By 2003, Pets.com and Webvan were gone. Amazon, which had a real moat in logistics and customer data, dropped 95% from peak to trough — then compounded at roughly 30% annually for the next two decades. The infrastructure layer (Cisco) made real money on the buildout but became a value trap for fifteen years because the hardware lacked recurring economics. The moat mattered. The narrative didn’t.
Howard Marks has made the same point repeatedly in his memos: the most dangerous words in investing are “this time it’s different.” AI is genuinely transformative — and it will still be governed by the same laws of competition and capital returns.
How to Apply It Today
Start by separating your AI exposure into three buckets:
- Picks and shovels: Semiconductor designers, foundries, networking, and power infrastructure. Evaluate based on cyclicality and customer concentration.
- Platform owners: Hyperscalers and foundation model providers. Evaluate based on unit economics per query and capex-to-FCF conversion.
- Application layer: Software companies embedding AI into existing workflows. Evaluate based on net retention and gross margin trajectory.
For each holding, write a one-page thesis answering: What does this company own that competitors can’t buy? If you can’t answer it in two sentences, it’s probably a momentum play. Position size accordingly.
Also useful: track the ratio of AI-related capex to AI-related free cash flow across the megacaps quarterly. When the gap stops widening, the market’s tolerance for “spend now, monetize later” stories will narrow fast.
Risks to Keep in Mind
Three failure modes are worth flagging. First, moats can be misidentified. Many businesses look like they have switching costs until a 10x better product arrives — Nokia and BlackBerry are reminders. Second, valuation still matters. A great business bought at the wrong price is a mediocre investment, a point Seth Klarman has made repeatedly in Margin of Safety. Third, regulatory and energy constraints could reshape AI economics in ways that punish today’s apparent winners. The companies with the largest capex commitments have the most to lose if power costs, chip supply, or model licensing rules shift.
Nassim Taleb’s reminder applies: in a world of fat tails, survival is the precondition for compounding. Own the businesses built to survive the cleanup phase — and treat the rest as what they are.
