Anthropic, Microsoft in Talks for AI Chip Deal

Microsoft is reportedly in advanced discussions to supply its proprietary artificial intelligence chips to Anthropic, the rapidly expanding AI research company. This potential deal, confirmed by sources familiar with the matter, could mark a significant win for Microsoft in the intensifying competition to provide specialized AI silicon to leading AI developers.

While Microsoft announced its second-generation Maia AI chip in January, aiming to enhance its Azure cloud offerings and power advanced models like OpenAI’s GPT-5.2, it has yet to widely deploy these custom chips externally. A partnership with Anthropic, a major player in the generative AI space, would underscore Microsoft’s strategy to leverage its hardware investments to secure key cloud clients. This move comes as rivals Amazon and Google are also aggressively pursuing similar strategies with their own custom AI silicon.

Details regarding the exclusivity or the full scope of the potential agreement with Anthropic remain unclear, and the company has not yet finalized a deal with Microsoft. The Information first reported on the ongoing discussions.

Shares of Microsoft saw minimal movement following the news.

The relationship between Microsoft and Anthropic is already substantial. In November, Microsoft committed to a significant investment of $5 billion in Anthropic, alongside Anthropic’s pledge to spend $30 billion on Microsoft’s Azure cloud services over time. However, Anthropic’s extensive computing needs also extend to cloud services from Amazon and Google, highlighting the complex multi-cloud strategies prevalent in the AI industry.

Anthropic’s co-founder and CEO, Dario Amodei, has openly acknowledged the company’s “difficulties with compute.” This challenge has become increasingly acute as its advanced AI assistants, including Claude and the AI-assisted programming tool Claude Code, have gained considerable traction throughout the year, demanding ever-greater computing power.

The escalating demand for computational resources was further underscored by a recent disclosure from SpaceX, revealing that Anthropic will pay an estimated $1.25 billion per month through May 2029 for computing power, a substantial commitment that points to the critical nature of securing reliable and scalable infrastructure.

Historically, Anthropic has heavily relied on Graphics Processing Units (GPUs) from Nvidia for training and deploying its sophisticated generative AI models. However, the company has been actively diversifying its hardware strategy. In April, Anthropic announced a decade-long partnership with Amazon Web Services (AWS), committing to spending upwards of $100 billion on AWS’s custom Trainium chips. Prior to that, in October, Anthropic had also signaled its intention to utilize Google’s Tensor Processing Unit (TPU) chips.

Both Anthropic and Microsoft declined to comment on the specifics of the potential AI chip deal.

Microsoft CEO Satya Nadella highlighted the performance advantages of the Maia 200 chip during the company’s earnings call in April, stating that it offers “over 30% improved tokens per dollar, compared to the latest silicon in our fleet.” Nadella also confirmed that these chips are currently operational within Microsoft’s data centers located in Arizona and Iowa.

Correction: An earlier version of this article incorrectly stated the name of Amazon Web Services’ custom chips. In April, Anthropic confirmed its use of AWS’s Trainium chips in a long-term agreement valued at over $100 billion.