Microsoft is in active discussions to supply its custom Maia AI chips to Anthropic, according to sources familiar with the talks. The deal, if finalized, would mark a strategic win for Microsoft's Azure infrastructure, which has lagged behind AWS and GCP in offering proprietary silicon to external tenants.

Anthropic's compute requirements have escalated sharply. CEO Dario Amodei recently acknowledged persistent "difficulties with compute" as Claude and Claude Code adoption grows. The company's financial filings with SpaceX reveal a staggering $1.25 billion monthly commitment through May 2029 for general compute capacity, underscoring the severity of the bottleneck.

Historically, Anthropic has relied on Nvidia GPUs, but the company has been diversifying aggressively. In April, it signed a 10-year, $100B+ deal to use AWS Trainium chips. October brought a commitment to Google's TPUs. Adding Microsoft's Maia 200 would give Anthropic a third custom silicon option, but introduces integration complexity across three cloud providers.

The Maia 200, which Satya Nadella claims offers "over 30% improved tokens per dollar" versus Microsoft's existing fleet, is currently deployed in Arizona and Iowa data centers. It already runs OpenAI's GPT-5.2, but has not been generally available to Azure customers. A deal with Anthropic would change that, potentially accelerating Maia's broader rollout.

Key engineering trade-offs: Anthropic must weigh the performance-per-dollar gains against the operational overhead of managing model compilation and scheduling across three distinct chip architectures. The $5B Microsoft investment and $30B Azure spending commitment provide financial incentive, but the real test is whether Maia can deliver at Anthropic's scale. No deal has been finalized, and both companies declined to comment.

Source: CNBC