Anthropic is negotiating with Microsoft to supply custom Maia 200 AI chips for use on Azure cloud servers, amid growing compute needs for its AI models [1, 2, 3, 4, 5]. The company has not yet finalized a deal for the chip usage, though talks remain active [1, 5].
Microsoft announced the second-generation Maia 200 AI chip in January 2026 but has not yet made it publicly available through its Azure cloud platform [1, 3, 4, 5]. The chips currently run in Azure data centers in Arizona and Iowa. Microsoft claims Maia 200 improves token per dollar efficiency by over 30% compared to its previous silicon fleet. Satya Nadella, Microsoft CEO, said, "Maia 200 offers over 30% improved tokens per dollar, compared to the latest silicon in our fleet" [1].
Anthropic CEO Dario Amodei publicly acknowledged in early May 2026 that the company faces significant compute constraints due to rising user demand for its Claude assistant and Claude Code AI tools. "We have faced difficulties with compute due to the surge in demand for our Claude assistant and Claude Code tool," Amodei stated [1, 3, 5]. To address these limitations, Anthropic is seeking rental access to Microsoft’s Maia 200 chips primarily to support AI model inference workloads and enterprise AI services [4].
Historically relying on Nvidia GPUs, Anthropic is diversifying its chip suppliers. It has multi-billion dollar contracts with Amazon AWS Trainium (valued over $100 billion for 10 years) and Google TPU chips alongside the ongoing Nvidia fleet [1, 3, 5]. Additionally, Anthropic signed a contract to pay SpaceX $1.25 billion per month for computing power through May 2029, emphasizing the scale of its compute needs [1, 3, 5].
Anthropic committed to spending $30 billion on Azure cloud services following Microsoft's $5 billion investment in the startup announced in November 2025. The deal marked a strategic partnership between the two companies [1, 3, 5]. Microsoft has yet to officially offer Maia 200 chips publicly on Azure, but discussions to incorporate the chip for Anthropic’s workloads are underway [1, 3, 4, 5].
The next expected step is a final agreement on Maia 200 chip rental terms between Anthropic and Microsoft, which would enable Anthropic to scale AI model inference capacity soon.
