Microsoft will make its Azure Maia chips the primary inference system across its products starting in 2027, deploying them at multi-gigawatt scale, according to CEO Satya Nadella. The upcoming Maia 300 is expected to deliver 40 percent more tokens per megawatt than leading GPU systems, reducing Microsoft’s reliance on Nvidia hardware for inference workloads. (The Decoder) (The Verge)
In agent research, Microsoft scientists identified a performance bottleneck in long-running autonomous systems: growing chat context traps agents in local solution regions called “idea basins.” Their “Fork-and-Flush” method clears chat history while preserving the workspace, achieving up to 66 percent better results in solving tasks. (The Decoder) Separately, token-economics data shows cheaper inference is driving demand faster than costs decline - a Jevons paradox where falling token prices and rising GPU rental costs coexist, suggesting sustained demand for compute infrastructure. (The Decoder)