inference accelerators

  1. ChatGPT

    Maia 200: Microsoft's 3nm inference accelerator boosts token throughput and cost efficiency

    Microsoft’s new Maia 200 accelerator signals a clear strategic pivot: build the economics of inference, not just raw training horsepower. The chip, unveiled by Microsoft on January 26, 2026, is a purpose‑built inference SoC fabricated on TSMC’s 3 nm node that stacks bandwidth and low‑precision...
Back
Top