Anthropic is making Claude cheaper to use. Its computing bill, however, is moving in the opposite direction.

Price reductions and customer discounts offset roughly $600 million of the company's third-quarter revenue growth, while compute and infrastructure expenses surged about 77%—far outpacing 47% revenue growth. Anthropic swung to an operating loss for the quarter, though short-term compute costs and a noncash charge complicate the picture.

In its amended SEC filing, Anthropic said revenue growth from increased usage was "partially offset by a decrease of approximately $600 million from price reductions introduced as part of new model versions and contractual discounts."

Meanwhile, compute and infrastructure expenses rose from $5.56 billion to $9.86 billion, an increase of roughly 77%.

That means Anthropic spent roughly 58 cents on compute and infrastructure for every dollar of revenue, up from 48 cents a quarter earlier. The measure includes computing used for research, training, and internal operations—not just serving paying Claude customers—so it is not a per-prompt profit margin.

The $600 million figure is not revenue that vanished from Anthropic's books. Rather, it is the amount by which price reductions and discounts offset revenue growth driven by higher usage. Demand, in other words, is not the constraint: usage is climbing, and the pressure comes from discounts and infrastructure costs growing faster than sales.

The price of cheaper AI

The cuts are concrete. Claude Opus 5.5, which lowered token prices compared with Opus 5, was recently released and is more capable than any previous Anthropic model, at times even on par with Fable. Anthropic also introduced Haiku 5.5 in October, pitching a much cheaper model for high-volume work.

Anthropic says Fable 5.1 reduced the cost of typical workloads by around 25%, while Opus 5.5 brought customers' cost per task down roughly 40% compared with Opus 5. The reductions reflect both lower prices and efficiency gains rather than straight discounts across every task, and Anthropic detailed Fable's lower cache-read pricing in its model announcement.

There is a business case for that strategy: cheaper, more capable AI can encourage customers to use more of it, and Anthropic says users have historically expanded consumption after such releases. But the latest quarter also shows how difficult it is to synchronize 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—capacity arriving now was, in many cases, booked before the current pricing rounds were set. The company has also drawn criticism among AI enthusiasts because its usage quotas tend to deplete too easily.

The quarter included $5 billion of compute under short-term commitments that Anthropic says 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.

Bigger long-term stakes

The longer-term stakes are larger. Anthropic's IPO filing means the company is preparing to go public, putting it squarely in Wall Street's sights. Its updated disclosures put aggregate lease and purchasing commitments at roughly $895.4 billion, substantially all related 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 that capacity. For now, Google and Nvidia appear to be the biggest beneficiaries of Anthropic's growth, as its core business depends on their infrastructure. How those financing questions are resolved, and whether revenue growth can keep closing the gap with discounts and compute costs, are the threads to follow as Anthropic moves through the IPO process.

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. OpenAI, which is also preparing its own IPO, offers a competitive model with many usage resets and a more generous usage quota—a key factor keeping Anthropic from raising prices. Upcoming disclosures will show whether the gap between cheaper models and costlier compute is narrowing.