According to foreign media reports on October 7, citing people familiar with the matter, Microsoft plans to launch the company’s first chip designed to support artificial intelligence (AI) at its annual developer conference next month. The move is the culmination of years of work and could help Microsoft reduce its reliance on Nvidia’s AI chips. Nvidia’s AI chips have been in short supply as demand surges.
Microsoft’s chips are similar to Nvidia’s graphics processing units (GPUs) and are designed for data center servers that train and run large language models, the software behind conversational AI features like OpenAI’s ChatGPT. Microsoft’s data center servers currently use Nvidia’s GPUs to power cloud customers including OpenAI and Intuit, as well as artificial intelligence capabilities in Microsoft’s productivity apps.
The chip, codenamed “Athena,” may be unveiled at Microsoft’s Ignite conference in Seattle on November 14. Athena is expected to compete with Nvidia’s flagship microprocessor H100 GPU to accelerate artificial intelligence in data centers. The custom chip has been secretly tested by teams at Microsoft and its partner OpenAI.
Microsoft began developing the Athena chip around 2019, seeking to cut costs while also hoping to increase its leverage in negotiations with Nvidia. Azure currently relies on Nvidia’s GPUs to power AI capabilities used by Microsoft, OpenAI and cloud customers. But with Athena, Microsoft can follow in the footsteps of rivals AWS and Google in offering its own AI chips to cloud users.
Athena’s performance details are unclear, but Microsoft hopes the chip will rival Nvidia’s H100. While many companies tout superior hardware and cost-efficiency, Nvidia GPUs remain the top choice for AI developers thanks to the company’s CUDA platform. Attracting users to new hardware and software will be key for Microsoft.
Microsoft’s in-house development of AI chips may also reduce its reliance on Nvidia amid tight GPU supplies. After beginning to work closely with OpenAI, Microsoft reportedly ordered at least hundreds of thousands of Nvidia chips to support OpenAI’s product and research needs. By using your own chips, significant cost savings can be achieved.
OpenAI may also be considering reducing its reliance on Microsoft and Nvidia chips. There have been reports recently that the artificial intelligence research laboratory is considering manufacturing its own artificial intelligence chips. Recent job postings on the OpenAI website also indicate that the company intends to hire talent to evaluate and co-design AI hardware.
While Microsoft and other cloud providers have no immediate plans to stop buying GPUs from Nvidia, it could be financially beneficial in the long run to convince their cloud customers to move more to in-house chips rather than Nvidia’s GPU servers. Microsoft is also working closely with AMD on its upcoming artificial intelligence chip, the MI300X. As AI workloads proliferate, this diverse approach offers a variety of options. Cloud computing rivals are employing similar strategies to avoid vendor lock-in.
Amazon and Google have strategically integrated their AI chips into promotions for their cloud businesses. Amazon provided financial support to OpenAI rival Anthropic on the condition that Anthropic would use Amazon’s artificial intelligence chips, called Trainium and interentia. Meanwhile, Google Cloud announced that customers including artificial intelligence image developer Midjourney and Character AI are using the company’s tensor processing units.
As artificial intelligence chips become an essential part of data centers, the rewards for betting on the space could be high. With this development, Microsoft will also join the ranks of competitors fighting for market share in the accelerating field of artificial intelligence chips. With Athena, Microsoft can offer cloud customers more choices while charting a more independent course on next-generation AI infrastructure.
