"Microsoft is planning to make "significant investments" in its own AI chip cluster to become "self-sufficient in AI," Microsoft AI CEO Mustafa Suleyman said during an all-employee town hall meeting on Thursday. Microsoft's AI strategy has so far largely relied on a partnership with OpenAI, although the companies appear to be drifting apart lately and they're locked in tense contract renegotiations right now. Suleyman's comments suggest Microsoft wants to forge its own path in AI, while still supporting OpenAI with cloud-computing services."
"Instead of relying solely on OpenAI, Microsoft is using open-source models, partnering with other AI developers, and building its own models, Suleyman said. The software giant unveiled MAI-1-preview in late August. This is Microsoft AI's first foundation model trained end-to-end by the company, and offers a glimpse of future offerings inside its Copilot service. This model ranks 24th among text models on LMArena, a widely followed leaderboard, so Microsoft has a lot of work to do still."
""It's critical that a company of our size, with the diversity of businesses that we have, that we are, you know, able to be self sufficient in AI, if we choose to," Suleyman said. Microsoft plans to make "significant" investments in its own AI chip cluster to help the company build its own models, he added. Suleyman noted that MAI-1-preview was only trained on 15,000 Nvidia H100s, which he said was a "tiny cluster" in the grand scheme of things."
Microsoft plans significant investments in an internal AI chip cluster to achieve self-sufficiency in AI and the capacity to build world-class models in-house. The company is complementing its historical partnership with OpenAI by adopting open-source models, partnering with other developers, and training its own models. The MAI-1-preview model marks Microsoft AI's first end-to-end foundation model, trained on 15,000 Nvidia H100s and currently ranking 24th on an industry leaderboard. Microsoft recognizes the need for larger clusters for frontier training while remaining pragmatic about using external models where appropriate.
Read at Business Insider
Unable to calculate read time
Collection
[
|
...
]