Amazon cuts Alexa+ AI costs by reducing Anthropic model use

Amazon's Alexa+ is projected to cost $1.7 billion in AWS cloud spending by 2026, prompting a sweeping effort to route fewer queries to Anthropic's Claude models.

Amazon's Alexa+ is shifting more queries to its in-house AI models and away from Anthropic's Claude, part of a cost-cutting push that could more than quadruple the number of transactions each unit of computing capacity supports, internal documents show.

"There's an urgency to reduce the cost per unit in AI," Chief Executive Officer Andy Jassy wrote in his 2025 shareholder letter, arguing that cheaper inference would "unleash AI being used as expansively as customers desire."

Internal forecasts reviewed by Business Insider showed AWS cloud costs for the upgraded Alexa+ were on pace to reach roughly $1.7 billion in 2026, nearly triple the prior year's level. The service was running about 60% above Amazon's target for cloud cost per monthly active user. Even after identifying roughly $450 million in potential savings, internal reviews concluded the business would not hit its financial targets.

The effort highlights a broader shift in the AI industry as competition moves from building smarter models to making them cheaper to run. Amazon's approach — routing simpler requests to lower-cost models while reserving frontier systems like Claude for complex tasks — mirrors a strategy adopted by OpenAI and Cursor. Investment firm William Blair wrote in a recent report that "multi-model routing is becoming standard architecture in software."

Amazon's internal roadmaps called for moving specialized Alexa "Experts" from Anthropic's Claude Sonnet to Amazon's own AI models while reducing other use of Claude across the digital-assistant service. The company also sought to avoid unnecessary inference by expanding caching — storing answers to common requests so the AI doesn't repeat the same work — and increasing deterministic handling for predictable queries that don't require a large language model.

The strategy is notable given Amazon's deep ties to Anthropic. The company has invested billions in the AI startup, partners closely with it, and stands to reap a significant windfall from Anthropic's IPO. Yet internal documents show Amazon has been actively looking for ways to reduce how often Alexa relies on Anthropic's models.

GPU Efficiency Gains

Beyond model selection, Amazon focused on squeezing more work from each GPU. Rather than simply adding more Nvidia chips, the company projected software upgrades would increase available computing capacity by roughly 50% while cutting response times by about 40%. Internal planning dashboards tracked projected customer growth, GPU utilization, available capacity and inference efficiency as Amazon prepared to scale Alexa+.

Amazon also evaluated both Nvidia GPUs and its own Trainium chips to further lower costs. The company even weighed delaying parts of Project Moonraker, an effort to give Alexa more advanced AI agent capabilities, which was expected to become the service's largest AI expense this year.

Investor Implications

For Amazon, the cost discipline signals improving margin potential for its Alexa division and reduced dependency on a third-party AI provider. Amazon shares trade at roughly 22 times forward earnings, and every percentage point of margin improvement in Alexa could translate to hundreds of millions in annual operating profit. For Anthropic, the reduced reliance could temper revenue expectations from one of its largest customers, though the startup's broader enterprise and consumer business continues to grow. The broader takeaway for investors: the AI inference market is commoditizing fast, and companies that control both the model and the infrastructure — like Amazon with its Trainium chips and AWS — hold a structural cost advantage.

This article is for informational purposes only and does not constitute investment advice.