Beyond the AI Bubble. Go Vertical. Go Industrial.

Capital Intensity Defines the Platform Race; Defensibility Defines the Vertical One.

The hot rounds; the 2 a.m. ‘Are we in?’ texts.

You can hear the market breathing hard again, elbows out, repricing the future by the hour. Artificial intelligence is, at once, the most capital-hungry technology of the last generation and the most prosaic story in enterprise software on the factory floor.

One side of the market looks like sovereign project finance; the other looks like a classic, sweaty block-and-tackle: tickets closed faster, denials overturned, scrap reduced, days sales outstanding pulled forward.

We have been here before. Platform shifts inflate expectations faster than cash flows can catch up, public markets get skittish, and venture cyclicality begins to feel like a law of nature.

Every so often, however, the market rediscovers a wedge—an operating model that turns exuberance into earnings.

In mobile, it was distribution moats and payments; in cloud, it was usage-based pricing and land-and-expand. In AI, the wedge increasingly looks vertical, not horizontal. The portfolio is a barbell: at one end, eye-watering frontier-model rounds that concentrate risk; at the other, industry-specific AI that converts domain constraints into defensible software margins.

Gravity well: the skyscraper phase

Start with the gravity well. Anthropic raised $13 billion at a $183 billion post-money, tripling its March mark, with an annualized revenue run rate crossing $5 billion in August and Claude Code alone pacing at a $500 million run rate.

Demand reportedly topped $25 billion, with a syndicate that blended crossover giants and Middle Eastern capital. The subtext is capital intensity: building frontier models now resembles hyperscale infrastructure, with cash burn measured in data centers rather than headcount.

For venture funds without compute contracts in their back pocket, these rounds are both thrilling and binary. Either the model providers keep compounding and compress downstream margins, or they reset and everything from late-stage marks to secondaries reprices. It is a skyscraper—rare, expensive, impossible to ignore. But skyscrapers do not make a city livable for everyone; they define the skyline for a handful of owners.

Unless you can secure model supply chains, bargaining power in the cloud, and global distribution leverage, the platform layer is a spectator sport.

Plumbing vs. Promises: a Darwinian sort

The valuation regime is telling on itself. Private tags at the high end have outpaced what history suggests the U.S. IPO machine can routinely absorb, even on an inflation-adjusted basis.

That is not an indictment of fundamentals so much as an admission that our old exit plumbing was designed for smaller animals. If and when these towers list, the syndicate will have to turn up the water pressure for a building taller than any of its pipes were built to serve.

You can believe in AI’s secular power and still be honest about the plumbing. In Benjamin Graham’s long-familiar framing, the voting machine is working overtime; the weighing machine will have its say later.

Call it a bubble if you like. I would call it a Darwinian sorting mechanism.

And here is the question: