Microsoft Maia 200 May Gain Anthropic as First Major Customer

According to industry reports, Microsoft is reportedly in discussions with artificial intelligence company Anthropic to supply its in-house custom AI chips, a move that could mark a significant step forward in Microsoft’s position in the AI silicon supply chain. If finalized, the agreement would represent a notable milestone, as Microsoft has so far lagged behind cloud competitors Amazon and Google in offering dedicated AI accelerator chips to customers.

Microsoft introduced its second-generation Maia AI chip, the Maia 200, in January this year. However, the chip has not yet been made available to customers through the Azure cloud platform. At the time of launch, Microsoft stated that the Maia 200 processor is designed to support advanced workloads, including the operation of OpenAI’s GPT-5.2 model.

Sources familiar with the matter indicate that Anthropic has not yet reached a final agreement with Microsoft regarding the adoption of Maia chips. Earlier media reports this week also highlighted ongoing negotiations between the two companies. Microsoft’s share price remained largely unchanged on the day of the reports.

In November last year, Microsoft announced a planned $5 billion investment in Anthropic, while Anthropic committed to spending $30 billion on Azure cloud infrastructure. Despite this partnership, Anthropic continues to diversify its compute resources across multiple cloud providers, including Amazon and Google.

Anthropic co-founder and CEO Dario Amodei stated at an event earlier this month that the company is facing “compute constraints.” Demand for computing power has accelerated as its Claude assistant and Claude Code AI programming tools have gained rapid popularity this year.

Earlier this week, SpaceX disclosed that Anthropic has agreed to pay approximately $12.5 billion per month for computing resources through May 2029, underscoring the scale of its long-term infrastructure requirements.

Historically, Anthropic has relied heavily on NVIDIA GPUs for training and deploying generative AI models. In April, the company announced a major agreement to use Amazon Web Services’ custom Trainium chips under a 10-year contract valued at more than $100 billion. In addition, Anthropic previously confirmed plans to adopt Google’s Tensor Processing Unit (TPU) technology for AI workloads.

Anthropic declined to comment on its discussions with Microsoft, and Microsoft has not issued an official response regarding the matter.

Microsoft CEO Satya Nadella stated during the company’s April earnings call that the Maia 200 chip delivers more than a 30% improvement in token output efficiency per dollar compared with Microsoft’s previous-generation silicon. He also noted that the chip is already being deployed in Microsoft data centers in Arizona and Iowa.