Anthropic closed a $65 billion Series H yesterday at a $965 billion post-money valuation. Altimeter, Dragoneer, Greenoaks, Sequoia led. Capital Group, Coatue, D1, Baillie Gifford, Blackstone, Brookfield, D.E. Shaw, DST, Fidelity participated. The same week, OpenAI filed confidentially for an IPO. Both companies are now pointed at Q4 2026 listings near or above a trillion dollars.

A round this size buys a lot of things. One of the things it doesn’t buy is exemption from arithmetic.

The multiple

Anthropic’s last published annualised revenue was around $14 billion. At $965 billion post-money, that’s a roughly 69x revenue multiple on a company that isn’t profitable yet. Nvidia at peak price-to-sales hit about 30. Broadcom peaked near 33. Palantir touched 112 before correcting. The historical pattern for megacap tech sustaining price-to-sales above 30 is a subsequent 75–90% drawdown.

This isn’t an indictment of Anthropic specifically. The product works. The revenue growth is real. The team is good. None of that is in dispute. What’s in dispute is whether 69x is a price that can survive contact with public markets and a normal interest-rate environment.

The Series H investors aren’t doing diligence on revenue multiples. They’re buying IPO allocation. That’s a different calculation, and it’s the same one that ran in 1999.

What the round actually changes

The structural arguments for an AI bubble — capex against revenue, model commoditisation, the gap between announced spend and consumer monetisation — none of them are weakened by this round. Bain & Company’s estimate that the AI industry needs to generate around $2 trillion in annual revenue to justify current infrastructure investment is unchanged. US AI capex projections of roughly $500 billion against US consumer AI spending of roughly $12 billion are unchanged. DeepSeek’s demonstration that frontier-class models can be trained for a fraction of what the incumbents spend is unchanged.

What the round changes is timing. The bubble thesis has been arguing late-2026 to mid-2027 as the likely correction window. The Series H now points sharper. Both Anthropic and OpenAI have signalled Q4 2026 IPOs. That converts a vague range into a binary event with a date.

A dual mega-IPO has two outcomes. The IPOs price at or above target, in which case the bubble extends another twelve to eighteen months at higher altitude and a wider gap. Or one of them prices below target, in which case sentiment cascades immediately and the correction starts that quarter. There isn’t a soft-landing path through it.

The macro backdrop doesn’t help. Jerome Powell stepped down as Fed Chair this month. The most cited correction trigger in the bubble literature — rate moves forcing capital discipline on overheated AI capex — is now layered on top of a leadership transition at exactly the wrong moment.

What it doesn’t change for builders

The model is still subsidised. Right now, today, the foundation-model companies are pricing consumer access well below the cost of producing it. That subsidy is what makes solo development at frontier capability possible — for me, for everyone reading this, for the small Irish teams who’ve started shipping AI-native products in the last eighteen months.

The Series H extends Anthropic’s runway through the IPO event regardless of which way the IPO breaks. That’s good. What it does not extend is the subsidy. After IPO, public markets require cash flow justification, not narrative. Consumer-tier pricing pressure intensifies the day after listing, whether the listing succeeds or fails.

The build window for solo and small-team developers using subsidised frontier AI just shrank from “twelve to eighteen months” to “until the IPO repriced or rejected.” That’s roughly five to seven months from now.

The steelman

There’s a counter-argument worth naming. On the same day the round closed, Anthropic shipped Claude Opus 4.8 — forty-one days after Opus 4.7, the fastest minor-version cadence the company has run. That’s the underwriter’s case for $965 billion: cash flow doesn’t justify the price, but build velocity might. No other foundation-model lab is shipping at this rate, and the cadence has been accelerating, not slowing.

It’s the strongest non-financial argument in the bullish file. It doesn’t refute the 69x revenue multiple. It says the multiple is buying the option on a company that’s compounding capability faster than it’s compounding revenue — and that eventually those curves converge. The skeptical reading: shipping cadence is a function of capital, not the other way around. The reason Anthropic can release every six weeks is that it just raised $65 billion. When the funding cycle ends, so does the cadence. We’ll know which reading is right after the IPO.

The 1999 signature

In the late stages of the dot-com bubble, the most reliable signal wasn’t valuation alone. It was who was buying. When the late-stage cohort piled in at prices the early-stage cohort wouldn’t touch, you were in the part of the cycle where pricing had decoupled from analysis.

The Sequoia-Altimeter-Dragoneer-Greenoaks cohort at $965 billion is that cohort. They’re not wrong to be there — IPO allocation is genuinely valuable, and the firms that get it tend to do well on the listing pop regardless of the underlying. But “we’ll do well on the pop” is not the same argument as “this is worth $965 billion.” It’s the argument that says the next person paying $1.5 trillion is the one who has to worry.

The technology survives. The internet did. The dot-com bust didn’t end e-commerce; it ended the business models that priced themselves like the dot-com bust wasn’t coming. The AI equivalent is that the underlying capability keeps compounding while the business architecture that priced it reorganises violently.

Build now. Own your infrastructure. Don’t take on fixed costs that depend on subsidised access continuing.

The date is on the calendar.

Todd McCaffrey is a New York Times bestselling author, founder of FoxxeLabs Limited, and an MSc candidate in Cyberpsychology at ATU Letterkenny. This piece extends the framework set out in The Economics of AI Disruption and updates the bubble timing analysis for the May 2026 Anthropic Series H close.