Microsoft just built its own answer to Nvidia's inference tax. Maia 300: TSMC N2P, 8 HBM4 stacks (288GB), 18 TB/s per chip, 2.5x Maia 200 bandwidth. Multi-gigawatt rollouts start 2027. My take: inference is where the money burns, so every hyperscaler is building its own silicon. Your cloud bill will eventually reflect it. Which cloud will win the inference cost war: Nvidia GPUs or in-house chips? More AI news at www.contentbuffer.com #AI #Microsoft #Maia300 #AIChips #Inference #Azure #Nvidia