Microsoft is developing its own AI processor for training large models, according to sources speaking to tech site The Information. The processor, codenamed Athena, has been in development since 2019 and is currently being tested by a limited number of Microsoft employees and AI developer OpenAI.
Athena is designed for training large language models (LLMs) such as GPT-4, and to help mitigate the risk of AI model interference. Microsoft’s decision to create its own processor was driven by the need for alternatives to the AI processors it currently purchases from other suppliers, such as Nvidia’s A100 and H100 accelerators.
By creating its own AI processor, Microsoft hopes to save money and time on its AI developments. Other hyperscalers such as AWS, Google, and Facebook are also developing their own processors to reduce their reliance on third-party suppliers.
It is unclear whether Microsoft will make its proprietary AI processor available to a wider audience, such as Azure users. The company has its own internal roadmap for the processor, and further details may be revealed in due course.
