Microsoft plans to announce its in-house artificial intelligence chip at its annual developers’ conference next month, the Information reports.
The data center chip will be used to train and run large language models as well as other AI workloads.
Codenamed 'Athena,' the chip is planned to compete against Nvidia's H100 GPU, which is both expensive and in short supply. Microsoft is still debating whether it will make the chip available to Azure cloud customers, and may first use it for internal work.
The company began working on the chip in 2019, putting it behind rivals Google and Amazon which already have mature AI chip platforms. Google uses its TPU family for its own generative AI models, while AWS has pitched Trainium and Inferentia as cheaper alternatives to Nvidia.
Microsoft also invested in OpenAI back in 2019, building huge Nvidia GPU-based supercomputers for the company. But now the startup is reported to be considering developing its own AI hardware.