Anthropic's $9.7 Billion Research Bill Is Why the Shorts Are Lined Up for the $2 Trillion Print

Anthropic filed its Form S-1 publicly on Oct. 26, 2026, registering Class A shares for Nasdaq under the ticker ANT, with the share count and the price range left blank. The calendar is what matters for the next three weeks: marketing can begin Nov. 10, the SEC is closed Nov. 11 for Veterans Day, the earliest price-setting date is Friday, Nov. 13, and the earliest first trade is Monday, Nov. 16.

The most consequential line in the filing is not the ticker. Anthropic enters the market with $265.6 billion of non-cancelable purchase commitments against $131.7 billion of cash. Committed compute outlays already exceed cash by 2.0 times, which is why the November offering matters to lenders and hedge funds far more than it matters to model buyers.

Anthropic's cost base is not a software cost base. In the first half of 2026 it spent $9.7 billion on research and development, up 427% from a year earlier, and $8.9 billion of that sat in compute and infrastructure for model development, research and internal-use workloads. Cost of revenue — the line that carries serving paying customers — was $2.38 billion.

On the company's own adjusted measure, operating income was positive $1.62 billion in the half, against a $693 million GAAP operating loss — a gap of about $2.3 billion that the headline summary does not itemize. The 85% gross margin that flatters the business is also a function of where the training bill is booked: roughly four dollars in every five of Anthropic's compute is expensed as research, not carried in cost of revenue.

Bloomberg reported on Oct. 29 that short positioning in technology had piled up ahead of the Anthropic listing, setting up a potential squeeze if the deal lands. The measurable case sits one layer down, in AI infrastructure credit. CoreWeave had 69.8 million shares sold short at the Oct. 15 settlement — 20.1% of its public float — and on Oct. 14 its 2032 notes yielded 14.8% while the shares fell 15.5% in a session. C3.ai carried 32.7% of its float short at the same settlement. No aggregate short number for mega-cap AI has been published; the measurable evidence is the bi-monthly settlement data, and it is crowded.

The most damaging line for CoreWeave is inside Anthropic's own prospectus. Of the at least $475 billion of future cloud compute purchases Anthropic has committed to, at least $389.9 billion is pledged to Amazon. Its contracted gigawatts run to 4.7GW of NVIDIA GPUs, 4.5GW of Google TPUs, 1.2GW of Amazon Trainium and 0.8GW of AMD GPUs. No neocloud appears in that mix.

Here is the arithmetic that makes the microstructure tradeable. At $2 trillion, Anthropic would trade at roughly 61 times its annualized first-half revenue. Every other company in the chain trades between three and nineteen times sales. The market is being asked to pay a semiconductor multiple for a business whose largest cost line is contracted for a decade.

The long-horizon risk is a maturity mismatch, not a multiple. Anthropic's revenue is cancellable and its compute is not, and the gap between those two curves is where the next three years get decided.

The read-through is narrower than the headlines suggest. Anthropic's debut is not primarily a bet on Claude's model economics; at $1.5 trillion to $2 trillion it is a bet that the AI-credit chain's collateral will hold. The research bill is what forces the issue. A $9.7 billion half-year research bill against $16.29 billion of revenue makes the equity large enough to refinance the debt stack that built the compute — $265.6 billion of purchase commitments, $417 billion of payment obligations and a $63 billion loan vehicle collateralized on GPU residual values. If the print does that job, the crowded short base in the most-shorted AI names becomes the fuel. If it does not, $45.2 billion of CoreWeave net debt and $63 billion of TPU loans re-price at the same moment.

The asymmetry favours patience over aggression. The valuation argument is already owned by the market; the unsettled question is who is on the other side of the first trade. If the answer is hedge funds unwinding a crowded short into a freshly priced record listing, then the tape will look spectacular and mean very little. Allocation — not direction — is the scarce resource between Nov. 10 and Nov. 16.