Microsoft unveils Maia 300 AI inference chip, targets Nvidia efficiency
Microsoft has revealed the Maia 300, an AI accelerator it developed in-house, focused on inference. The company says the chip can handle 40% more tokens per megawatt than “leading GPU-based systems,” which appears to be a comparison to Nvidia’s offerings.
The announcement comes as Microsoft continues expanding its AI capabilities, including recent Copilot updates and new streaming voice models. The Maia 300 is reportedly built for Microsoft’s cloud services, suggesting an interest in competing in AI hardware, where Nvidia holds a strong position.
No performance benchmarks or rollout dates have been shared, but the move could indicate a shift in how cloud providers approach AI acceleration. Whether the chip meets its efficiency goals—and what it means for the broader market—remains to be seen.
Sources: siliconangle.com
“Microsoft’s custom AI inference chip may signal an effort to reduce dependence on Nvidia while improving cloud AI workload efficiency.”
Read the original reporting
The outlets below did the original reporting.
- Microsoft debuts multichiplet Maia 300 inference accelerator — siliconangle.com
Related briefs
- Dell symposium spotlights runtime AI governance shift
- AI infrastructure bottlenecks shift from chips to data pipelines
- Ex-Oxford Nanopore CEO raises £4m for AI materials discovery startup
- WindBorne secures $50M US Air Force contract for balloon-based weather data
- Dell shifts AI sales playbook for Silicon Valley’s DIY buyers
This brief was drafted automatically from the sources above and published under our editorial policy. Spotted an error? Tell us.