Business of GTM

Anthropic's Sequential Growth Fell to 47% as Compute Commitments Hit $895B

Anthropic's amended S-1 shows Q3 revenue near $17 billion, up 47% on Q2 after 144% the quarter before. Price cuts and discounts took about $600 million off the quarter's growth.

Anthropic's black wordmark is vertically centered against a warm off-white background, beside monochrome server racks and a bold red geometric arc.

TL;DR

  • Anthropic's amended S-1 puts Q3 revenue at $16.96 billion, up 47% on Q2 after 144% growth the quarter before.
  • Growth is slowing while compute commitments climb to $895.4 billion. That makes year-over-year growth the wrong number to plan on.
  • Forecast AI-priced products on quarter-over-quarter growth. At your next renewal, expect AI vendors to offer lower prices in exchange for committed volume.

Anthropic's amended S-1, filed November 2, reports $16.96 billion in Q3 revenue, up 1,127% from $1.38 billion a year earlier. Revenue leaders should plan around the sequential figure instead. Revenue grew about 47% from Q2 to Q3, after 144% in Q2 and 132% in Q1, and nine-month revenue reached $33.24 billion. Working from the filing's quarterly table, we calculate that Anthropic added about $6.8 billion of quarterly revenue in Q2 and about $5.4 billion in Q3. That's the first shrinking dollar gain in the seven quarters shown. The same filing puts purchase commitments and leases at about $895.4 billion, "substantially all of which is related to compute and infrastructure."

The year-over-year number mostly measures how small the base was, and a four-digit growth rate says little about next year. The sequential line shows a revenue curve that bent sharply in one quarter while the obligations behind it kept rising. An AI buyer negotiates against that pairing. The cost side is also what our earlier read of the first-half filing and its cloud marketplace sales left open.

Price cuts offset $600 million of growth as compute costs climbed

The filing puts a number on pricing. Between Q2 and Q3, higher usage was partly offset by "a decrease of approximately $600 million from price reductions introduced as part of new model versions and contractual discounts." On our numbers, adding that back would lift Q3's sequential growth only from about 47% to about 52%, so price cuts explain little of the slowdown. On the cost side, the company swung from a $974 million operating profit in Q2 to a $567 million operating loss in Q3. The filing puts that mainly down to two items. One is $5.0 billion of compute bought under short-term commitments, which costs much more than its multi-year capacity. The other is $2.0 billion of non-cash charges from its donation-matching program. Compute and infrastructure came to $9.86 billion, or 58% of revenue, up from 48% in Q2. Adjusted operating income stayed positive, though adjusted operating margin fell from 19% to 15%.

The $23.29 billion Q3 net loss looks worse than it is. About $21.9 billion of it came from fair-value changes on convertible notes, derivative liabilities and SAFEs. Those non-cash changes also account for most of the $104 billion nine-month loss, CNBC noted. With $146.6 billion in liquidity, the company faces a margin squeeze, and nothing in the filing points to an operating collapse.

The commitments are the likely source of pricing pressure. The filing says Anthropic typically buys compute 12 to 18 months ahead, so those costs are "generally fixed in the near term." Its revenue is far less fixed. Consumption made up 79% of nine-month revenue, and many top customers are "not subject to long-term contracts" and can cut or stop spending at any time. Remaining performance obligations, the contracted revenue not yet recognized, stood at $13.8 billion on September 30, below a single quarter's revenue. Tomasz Tunguz of Theory Ventures named the mismatch in a read of the first filing, which counted $637 billion of commitments. The commitments are largely fixed, Tunguz wrote, but the "revenue is consumption-based with no lock-in." Of the amended filing's $895.4 billion, about $69.8 billion falls due in the next 12 months and $447.5 billion before the end of 2030.

Anthropic already offers the trade. The filing says a growing number of enterprise customers sign non-cancelable annual or multiyear contracts with minimum spend "in exchange for a committed price list." The filing doesn't say what comes next. We expect a vendor carrying fixed compute costs against revenue that can walk to push that trade harder, through committed-use deals and pooled budgets. Marketplace billing gives it another route. We made the same argument about 2027 AI renewals asking buyers to commit.

Growing 47% in a quarter on an $11.6 billion base is still extraordinary, and this company has slowed and reaccelerated before. Using the same table, we count sequential growth falling from 109% to 76% to 48% through 2025. It then jumped to 132% and 144% after Opus 4.5 launched with a lower cost per task, a pattern Tunguz also flagged. The slowdown may come from supply rather than demand. The filing estimates that 5% to 7% of nine-month revenue came from customers breaking its usage policy or commercial terms, and enforcement stepped up in August and September. The reliance on short-term compute also suggests capacity was tight. The filing itself gives no cause for the slowdown. Some commitments can also move. About $213.8 billion sits in arrangements with exercisable termination rights, or with conditions that must be met before they start, such as a counterparty securing financing. The SpaceXAI agreements can be canceled on 90 days' notice after early 2027.

What this means for revenue leaders forecasting AI-priced products

If you built a plan on Anthropic's year-over-year growth, or on the usage curve boards now compare you to, rebuild it on sequential growth. Model each AI-priced line quarter over quarter. Assume the rate can fall by about two-thirds in a single quarter, as it did from 144% to 47%, because the filing shows it happening at the top of the market.

If you sell a product priced on AI usage, plan for price cuts at each model release. Anthropic absorbed about $600 million of them in one quarter. When your model vendor cuts per-task prices, your customers will expect some of that passed through. Build a price-reduction assumption into next year's revenue per unit instead of holding it flat.

At your next AI renewal, ask whether price reductions at new model versions flow through to committed rates. If they don't, a rate you locked before a release can end up above what new buyers pay. Our piece on OpenAI's discounts covers the rest of that negotiation.

The year-over-year number mostly measures how small the base was.

Business of GTM

Before your next AI renewal, ask what discount a commitment buys and whether price cuts at new model versions flow through to your committed rates.

By Business of GTM