Written by Max A. Charney
SAN FRANCISCO (Reuters) – Microsoft said on Thursday it plans to offer its cloud computing customers a platform of AMD artificial intelligence chips that compete with Nvidia components, with more details to come at next week's Build developer conference. It is scheduled to be announced.
A preview of the new Cobalt 100 custom processor will also be announced at the conference.
Advanced Micro Devices' flagship AI chip, the Microsoft cluster of MI300X AI chips, will be sold through Azure cloud computing services. It will offer customers an alternative to Nvidia's H100 family of powerful graphics processing units (GPUs), which dominate the data center chip market for AI but are in high demand and difficult to obtain.
To build AI models or run applications, companies typically need to chain together, or cluster, multiple GPUs because the data and calculations don't fit on a single processor.
AMD, which expects AI chip revenue to reach $4 billion this year, says the chips are powerful enough to train and run large-scale AI models.
Similar to Nvidia's top-of-the-line AI chips, Microsoft's cloud computing division sells access to its own AI chip called Maia.
Separately, the Cobalt 100 processor that Microsoft plans to preview next week will offer 40% better performance than other processors based on Arm Holdings' technology, the company said. Snowflake and others started using it.
The Cobalt chip, announced in November, is being tested to power Microsoft's enterprise messaging tool Teams and is positioned to compete with Amazon.com's in-house Graviton CPU.
Amazon announced this week that social network Pinterest and fintech company Robinhood Markets have started using its Graviton chip.
(Reporting by Max A. Charney and Stephen Nellis in San Francisco; Editing by Edwina Gibbs)
