DEVOURED
Inside Microsoft's Frontier Pre-Training Stack
Microsoft unveiled its internal frontier training stack, detailing the 2.8T-parameter MAI-Base-2 model and plans for enterprise deployments via Foundry.
What: MAI-Base-2 is a mixture-of-experts model with 80B active parameters. Microsoft is preparing to offer these frontier-grade capabilities to enterprise clients through its Foundry infrastructure service.
Why it matters: This move indicates that Microsoft is shifting from purely selling API access toward providing custom-trained, private frontier models for large enterprise clients.
Decoder
- Mixture-of-Experts (MoE): A model architecture where only a fraction of the total parameters is activated for each input.
Original article
Microsoft detailed the training of MAI-Base-2, a 2.8T-parameter mixture-of-experts model with 80B active parameters that it said surpassed every open-weight model in its held-out pre-training evaluations. The deep dive covers its data strategy, compute efficiency, full-stack infrastructure, and Microsoft's plans to turn internally trained frontier models into enterprise offerings through Foundry.
Explore all 51 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi