Anthropic is making Claude cheaper to use, even as the bill for the compute behind it keeps climbing.
In an amended filing with the U.S. Securities and Exchange Commission, the company said revenue growth from higher usage was "partially offset by a decrease of approximately $600 million from price reductions introduced as part of new model versions and contractual discounts."
That $600 million is not revenue that disappeared from Anthropic’s books. It is the amount the company said was offset by lower pricing and discounts while usage growth was pushing revenue higher.
Compute and infrastructure costs moved sharply higher
Anthropic said compute and infrastructure expenses rose from $5.56 billion to $9.86 billion, an increase of roughly 77%.
On that basis, the company spent about $0.58 on compute and infrastructure for every $1 of revenue, up from $0.48 a quarter earlier. The filing said this measure includes computing used for research, training, and internal operations, not only the cost of serving paying Claude customers, so it should not be read as a per-prompt profit margin.
Lower-priced models are already in the market
The price cuts are not theoretical. Anthropic recently released Claude Opus 5.5, lowering token prices compared with Opus 5. The report said the model is more capable than previous versions and in some cases can perform at a level comparable to Fable.
Anthropic also introduced Haiku 5.5 in October, presenting it as a much cheaper model for high-volume workloads.
According to the company, Fable 5.1 reduced the cost of typical workloads by around 25%, while Opus 5.5 lowered customers’ cost per task by roughly 40% compared with Opus 5. Those reductions reflect both lower prices and efficiency gains, rather than a flat discount across every task. Anthropic also detailed Fable’s lower cache-read pricing in its model announcement.
Cheaper AI can drive demand, but supply is harder to line up
There is a clear business rationale behind the strategy. More capable AI at lower prices can encourage customers to use more of it. Anthropic said users have historically expanded consumption after releases like these.
The latest quarter also showed how hard it is to match demand with supply. Anthropic typically reserves compute capacity 12 to 18 months before it becomes available, leaving much of that spending relatively fixed in the near term. The report also noted that among AI enthusiasts, Anthropic is well known for quotas that tend to run out too easily.
Short-term compute commitments weighed on quarterly results
The quarter included $5 billion of compute under short-term commitments. Anthropic said that capacity cost significantly more than comparable capacity secured through multiyear contracts.
That expense helped push the company from a $974 million operating profit in the second quarter to a $567 million operating loss in the third.
IPO disclosures put longer-term obligations in focus
Anthropic has made its IPO filing public and is preparing for a Nasdaq listing under the ticker ANT, putting the company more directly in front of Wall Street.
Its updated disclosures put aggregate lease and purchasing commitments at roughly $895.4 billion, with substantially all of that tied to compute and infrastructure.
Reuters has also reported that the filing raised questions about whether infrastructure partners can secure the financing needed to deliver all of that capacity. For now, Google and Nvidia appear to be among the biggest beneficiaries of Anthropic’s growth because the company’s core business depends on their infrastructure.
Demand is growing, but pricing power remains constrained
Demand for Claude is still growing rapidly. The harder question is whether Anthropic can keep making its models cheaper faster than the cost of building and running them comes down.
The report also pointed to OpenAI, which is preparing its own IPO and offers a competing model with many usage resets and a more generous usage quota.
That remains the main issue standing in the way of higher prices at Anthropic.
