Anthropic’s Cheaper AI Is Driving Growth—and Putting Pressure on Its Business

Preface

Anthropic is making its Claude models less expensive to use, with the aim of encouraging customers to adopt them for more tasks. The strategy has helped bring down the cost of common workloads and accompanied rapid growth in usage. But the company’s latest financial disclosures reveal a difficult trade-off: price reductions and customer discounts reduced third-quarter revenue growth by approximately $600 million, while compute and infrastructure expenses rose sharply.

The figures do not mean that Anthropic lost $600 million in revenue already earned. Rather, the company said that lower prices and contractual discounts offset part of the growth generated by increased usage. At the same time, large infrastructure commitments and costly short-term computing capacity contributed to an operating loss. This article examines what those numbers mean, why cheaper models may still make strategic sense, and the challenge Anthropic faces in aligning customer demand with the cost of supplying AI.

Lazy bag

Anthropic is using lower prices and more efficient models to encourage wider adoption of Claude, but the financial benefit of increased usage is not yet keeping pace with infrastructure spending. In its amended SEC filing, the company said price reductions and contractual discounts offset approximately $600 million in revenue growth. Compute and infrastructure expenses increased from $5.56 billion to $9.86 billion, roughly 77%, while revenue grew 47%. Short-term compute commitments contributed to a move from a $974 million operating profit in the second quarter to a $567 million operating loss in the third.

Main Body

Anthropic’s pricing strategy reflects a central question for businesses selling advanced AI: can lower prices generate enough additional usage to compensate for the revenue sacrificed on each task? The company has been reducing the cost of using Claude while releasing models that it describes as more capable or more efficient. Its aim is to make AI practical across a broader range of customer activities, including high-volume workloads that could be uneconomical at higher prices.

The financial effect was visible in Anthropic’s amended SEC filing. The company 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.” This wording matters. The $600 million was not revenue that disappeared after being recorded; it was the amount by which lower prices and discounts reduced the growth that higher usage would otherwise have produced.

At the same time, Anthropic reported a steep rise in the cost of compute and infrastructure. Those expenses increased from $5.56 billion to $9.86 billion, an increase of roughly 77%. Revenue growth was 47%. The comparison points to pressure on the economics of expansion: the company’s infrastructure costs grew faster than revenue over the period described.

One way to illustrate the relationship is to compare infrastructure spending with revenue. Anthropic spent roughly 58 cents on compute and infrastructure for every dollar of revenue, compared with 48 cents a quarter earlier. That ratio should not be mistaken for a direct measure of the cost or profitability of an individual Claude prompt. The expense category includes computing used for research, model training, and internal operations, in addition to the compute required to serve paying customers. It therefore captures a broader set of activities than customer inference alone.

Anthropic’s model releases show how it is trying to reduce the cost of using AI. Claude Opus 5.5 was released with lower token prices than Opus 5. Anthropic also introduced Haiku 5.5 in October, presenting it as a much cheaper option for high-volume work. The company has said 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 savings do not necessarily come from applying a uniform discount to every use. They can reflect a combination of lower listed prices and improvements in model efficiency: a model may complete a task using fewer resources, or a lower-cost model may be suitable for work that previously required a more expensive one. Anthropic also described lower cache-read pricing for Fable in its model announcement. The practical result is that customers may be able to complete similar work at lower cost, even when the source of the savings varies by workload.

There is a clear strategic rationale for such reductions. If a model becomes cheaper and more capable, existing customers may use it for more tasks, and new customers may find it accessible. Anthropic says users have historically expanded consumption following releases of this kind. Higher usage can support revenue growth and make the technology more valuable to customers. However, if prices fall faster than usage rises—or if the extra usage requires expensive capacity—the company may struggle to improve its margins in the short term.

That challenge is complicated by the timing of infrastructure procurement. Anthropic typically reserves compute capacity 12 to 18 months before that capacity becomes available. As a result, a significant share of near-term spending is relatively fixed. The company cannot necessarily reduce its infrastructure costs immediately when prices change or when demand shifts. Planning capacity well in advance can help secure resources for future growth, but it also exposes the business to costs that may be difficult to adjust quickly.

The third quarter included $5 billion of compute under short-term commitments. Anthropic said these arrangements cost significantly more than comparable capacity secured through multiyear contracts. That more expensive capacity helped push the company from a $974 million operating profit in the second quarter to a $567 million operating loss in the third. The comparison highlights the role of procurement conditions and contract duration in reported results. It does not, by itself, establish that lower Claude prices were the only cause of the operating loss.

The company’s longer-term commitments make the question of capacity especially important. Anthropic’s IPO filing—an indication that it is preparing to go public—put its infrastructure plans under greater scrutiny. Updated disclosures put aggregate lease and purchasing commitments at roughly $895.4 billion, substantially all related to compute and infrastructure. Commitments on this scale make future demand, financing, and delivery capacity important to the company’s ability to carry out its plans.

Reuters has reported that the filing raised questions about whether infrastructure partners can secure the financing needed to deliver all the capacity Anthropic expects. Google and Nvidia appear to be among the companies positioned to benefit from Anthropic’s growth, because its core business depends on their infrastructure. That reliance also illustrates that the AI business extends beyond the model maker: building and operating advanced systems depends on a wider network of computing resources and partners.

Competition adds another constraint. The source describes OpenAI as preparing its own IPO and offering a competing model with frequent usage resets and a more generous usage quota. For customers choosing between AI services, price is only one consideration; access limits, model capabilities, reliability, and the amount of work permitted under a plan also matter. Competitive offerings can make it harder for Anthropic to raise prices, even when its own costs are rising.

For now, demand for Claude is still growing rapidly. The open question is whether Anthropic can keep lowering the cost of its models—and the cost of completing customer tasks—faster than the expense of building and running the infrastructure behind them declines. If increased usage generates enough additional revenue and efficiency improves, lower prices may support a larger business over time. If infrastructure costs remain elevated while pricing stays under competitive pressure, the same strategy could continue to weigh on operating results. The balance between affordability, adoption, and the cost of capacity is therefore central to Anthropic’s business outlook.

Key Insights Table