Amazon pushes back on report that Anthropic deal changes will raise its AI costs

Amazon’s stock rose slightly after-hours and in early trading following a report that its deal with Anthropic is being reworked in a way that could increase the cost of Amazon’s artificial intelligence services. In response, Amazon denied the specific allegation, according to the market report.

The report, carried by Yahoo Finance, said Amazon is moving away from paying Anthropic based on computing hours toward a token-based pricing model that would take effect beginning next year. A “token” is a unit of text processing used in large language models, so token-based pricing typically ties costs more directly to the amount of text generated or processed rather than the underlying compute time.

According to the same account, the change is expected to be linked to Amazon’s broader AI strategy, in which it increasingly supplies models and AI features to customers through AWS and its own consumer and business products. Amazon’s denial was aimed at the conclusion that the restructuring would lift AI costs, rather than directly addressing whether the business terms themselves are changing.

Token-based pricing can shift cost exposure depending on how workloads are run. If usage involves longer outputs or more token consumption, token pricing can raise costs relative to time-based compute pricing. Conversely, if systems are optimized to reduce token usage, the economics can improve. That means the direction of cost impact can depend on workload mix, model selection, and how requests are paced, none of which the market report detailed.

Amazon did not provide additional terms in the market update, such as the exact rate card, minimum commitments, or whether any caps, volume discounts, or performance-based adjustments are involved. It also did not disclose how the company’s internal accounting would map token-based charges to the cost line items for AI products sold through AWS.

For Amazon, these contract mechanics matter because generative AI workloads can be expensive and because AWS has been trying to convert model capability into recurring customer spend. Even small shifts in how third-party model providers price access can affect margins, especially when AI is used at scale across customer applications, enterprise deployments, and internal tooling.

Separately, Amazon’s corporate communication channels continue to emphasize its ongoing investment in AI and cloud infrastructure, though they do not usually comment on specific counterparty pricing arrangements in near real time. The company’s newsroom and other public materials are more likely to focus on product launches and platform updates than on confidential contract details.

What remains unclear from the market report and Amazon’s denial is the full scope of the restructuring. The information provided does not specify whether the change is exclusive to Anthropic, whether it applies to all model tiers or only certain variants, or whether the accounting impact will be offset elsewhere within the pricing chain. Investors and customers will likely look for additional detail in future disclosures, including earnings commentary and AWS-related updates.