Amazon is reworking how it pays for artificial intelligence from partner Anthropic, shifting from a compute-hour pricing model to token-based billing starting next year, according to a report from The Information. The change could meaningfully raise what Amazon owes for Anthropic models across Amazon Web Services (AWS), though the company disputes that costs will increase.
Amazon shares rose 3.58% to $241.02 on the day of the report, suggesting investors are more focused on whether AWS can manage the economics of its generative AI business than on the exact billing label.
How the billing shift works
Under the current compute-hour arrangement, Amazon essentially rents chunks of computing capacity from Anthropic. That makes costs relatively predictable once the capacity is booked. Token billing, by contrast, charges based on how much text a model reads and writes. Longer prompts, longer answers, and spikier traffic can all push costs higher.
The Information reported that the shift could meaningfully raise what Amazon owes for Anthropic models across AWS. Amazon told MT Newswires it is “incorrect” that the expanded collaboration will increase its costs, while also saying it continues to deepen the Anthropic partnership.
Amazon is also evaluating alternatives, including OpenAI models and its in-house Nova lineup, according to the report. That flexibility could prove important as AI usage scales.
What it means for AWS margins
For AWS, this is a margins story. Paying per token turns AI spend into a direct function of customer behavior. Profitability depends on keeping “tokens per task” from creeping higher as products get more capable and chatty.
That’s where flexibility matters: if AWS can route requests to the cheapest model that still meets quality needs—across Anthropic, OpenAI, and Nova—it gains a practical lever to protect gross margin and pricing power in its generative AI services as usage ramps.
The broader market is watching how cloud providers handle AI costs. Enterprises have been pushing back on AI token pricing, with some shifting to open-source models to control expenses. Amazon’s move could signal a broader industry trend.
Investor takeaway
For everyday investors, the key question is whether Amazon can maintain or improve AWS margins as AI usage grows. Token billing introduces more variable costs, but Amazon’s ability to mix and match models gives it some control.
The stock’s positive reaction suggests the market is giving Amazon the benefit of the doubt. But the shift highlights a broader challenge for AI companies: as models become more capable, they also become more expensive to run. How Amazon manages that trade-off will be a key factor in its AI profitability story.
Meanwhile, Chinese AI rivals could challenge OpenAI and Anthropic's IPO hopes, adding another layer of competition. And the US dollar's strength could affect multinational tech earnings.
