Anthropic Negotiates Use of Microsoft's Self-Developed AI Chips

According to a report by The Information, citing two informed sources, Anthropic is in negotiations with Microsoft to rent servers equipped with Microsoft's self-developed AI server chip, Maia, in order to alleviate its rapidly growing computational demand. The negotiations are still in the early stages, and it is uncertain whether a final agreement will be reached.

This news has caused Microsoft's stock price to rise by nearly 2% pre-market.

For Microsoft, securing a deal with Anthropic would mark a significant breakthrough in its self-developed chip business. Microsoft's self-developed chip project had experienced delays in the past, and the company now hopes to replicate the successful path of Google and Amazon by reducing reliance on Nvidia's AI chips through the development of their own chips.

Anthropic has previously committed to spending $30 billion on Azure, and by further adopting Microsoft's self-developed chips, the cooperation between the two parties would extend from model distribution and cloud computational power leasing to a deeper level of adaptation with AI hardware.

For Anthropic, using Microsoft's chips also aligns with its long-term adherence to a "multi-chip" strategy. The company has been leasing chips from different manufacturers such as Amazon, Google, and Nvidia for model development and operation, avoiding over-reliance on a single hardware supplier.

The positioning of Microsoft's Maia chip also better fits the current market demand. It mainly focuses on improving the inference speed of existing models, rather than helping customers train new models, thus it is expected to achieve breakthroughs in the inference scenario first. Microsoft said in January that the latest generation Maia 200 has been deployed in Azure data centers. Azure executive Scott Guthrie revealed that Maia 200 has significantly reduced the operating costs of the Copilot AI tool.