Microsoft Unveils Maia 200 Inference Chip; Nvidia Shares Largely Unmoved
Microsoft announced the Maia 200, a new AI inference accelerator built on TSMC's 3nm process and equipped with native FP8/FP4 tensor cores, 216GB of HBM3e, and 272MB of on-chip SRAM. The company says the chip delivers over 10 petaFLOPS in 4-bit precision and over 5 petaFLOPS in 8-bit within a 750W power envelope, and will be used to run OpenAI's GP…