The Prospectus Test

The decision

Both leading AI labs have now filed to go public. Anthropic submitted a confidential S-1 on June 1. OpenAI followed on June 8. Neither filing has appeared on the SEC’s public database yet.

Anthropic’s leaked anyway. On September 28, Reuters reported the contents of a 261-page draft prospectus. About 80 of those pages are risk factors. 48 cover the business. Anthropic is reportedly aiming for a valuation above $2 trillion, with a listing expected after the November midterms.

OpenAI’s filing hasn’t leaked. What we have is reporting: a run rate nearing $70 billion, enterprise sales that more than doubled since July, and talks about raising money at around $1.4 trillion as a bridge to a listing that may slip into 2027.

On September 14, we scored both companies on the MOATE index. Anthropic came in at 2.88, Solid. OpenAI came in at 3.13, inside Deep. A prospectus is the best stress test a scorecard can get, so we ran ours against the numbers.

The strategic why

Start with Anthropic’s revenue, because it’s the reason the $2 trillion number exists.

The company made $4.6 billion in 2025. By July 2026, it was running at $65 billion a year. Quarterly revenue went from $4.7 billion in Q1 to $11.5 billion in Q2. One analysis puts Claude Code at close to $8 billion of that Q2 figure. Anthropic passed OpenAI’s run rate in April and stayed ahead through the summer.

No software company has grown like that. That’s the bull case, and it’s real.

Now the other side of the ledger. The operating loss in 2025 was about $8 billion, nearly 3x the year before. The headline net loss was $42 billion, but about $34 billion of that is a non-cash accounting charge on convertible financing. Ignore it. The real story is the commitments.

Anthropic has signed up for $518 billion in compute and infrastructure, and 80% of it can’t be cancelled. Broadcom, which builds the custom chips behind Google’s TPUs, accounts for about $161 billion. Google and Amazon account for about $110 billion each. Microsoft accounts for about $31 billion. That’s roughly $113 of commitments for every dollar of 2025 revenue.

Then there’s the line that matters most for a moat: 2 customers made up nearly 25% of 2025 revenue, and neither has a long-term contract. The filing doesn’t name them.

OpenAI’s picture is thinner but leans the other way. ChatGPT passed 1 billion weekly users in July. Enterprise revenue overtook consumer revenue the same month. Ads reached a $1 billion run rate by late August. Its losses are bigger in absolute terms: leaked internal projections show about $278 billion of negative free cash flow through 2030. But in April it renegotiated its Microsoft deal, ending Azure exclusivity, and Amazon committed up to $50 billion. OpenAI now runs on more than one cloud.

One company has the better year. The other has the better structure.

What the filings change

We moved one dimension for each company. Everything else held.

Anthropic, independence: 3 to 2. This was already its weakest dimension. The prospectus makes it worse in 3 ways. Every watt still comes from a company that also competes with it. A quarter of revenue sits with 2 customers who can leave whenever they want. And on September 25, the D.C. Circuit ruled 2 to 1 that the Pentagon can keep its supply-chain-risk designation on Anthropic. An earlier lower-court win had pointed the other way. Now the appeals court has spoken. Few companies on our index have this many outside parties who can change their economics on their own.

We considered raising Anthropic’s cornered resource score on the strength of Claude Code. We didn’t. Two weeks ago, AWS released Strands Harness and showed it matching Claude Code’s accuracy on the same Claude model at roughly 40% lower cost. A product that earns 2/3 of a company’s revenue and can be matched by a free download isn’t a cornered resource yet.

That puts Anthropic at 2.83. It’s still Solid, now a rung below Palantir and barely above Qualcomm’s 2.81.

OpenAI, independence: already a 4. This is the one we had right in September, and the filings explain why. We scored OpenAI as a company whose leash had got longer: it can ship on any cloud, Microsoft no longer takes a revenue share, and its payments to Microsoft are capped through 2030. Nothing in the last month changed that, so nothing moves. Microsoft still owns about 27%, and $278 billion of projected cash burn keeps OpenAI dependent on investors. That’s why it was a 4 and not a 5, and why it stays one.

That leaves OpenAI at 3.13, comfortably inside Deep.

We left OpenAI’s brand at 8, even though ChatGPT’s market share fell below 50% for the first time in May. The name is still the generic word for the category. When that stops being true, the score moves.

The gap between the 2 labs widened from 0.25 to 0.29. The company with the faster-growing revenue fell further behind on structure.

The risk

For Anthropic, the risk is in the commitments. $518 billion of mostly non-cancelable spending is a bet that the revenue curve keeps going. If the July run rate turns out to be a peak, perhaps because 2 big customers shift spend or a free harness pulls coding work to cheaper models, those commitments don’t shrink with it. Revenue is flexible. Contracts aren’t.

For OpenAI, the risk is that it hasn’t shown its numbers yet. Every OpenAI figure here comes from reporting and leaks. Its S-1 could move the score either way, and a company that projects losses through 2030 is asking for a lot of faith. OpenAI itself has said a listing “may be a while.”

For our index, the risk is the obvious objection: a scorecard that marks down the company with the best revenue in software history might simply be wrong. That’s fair. The index doesn’t measure how fast a kingdom is growing. It measures how hard it would be to take. Today, Anthropic is easier to take than its revenue suggests.

Revenue is a scoreboard. A moat is a wall.