5 articles
Index of editorial tags and filtered coverage.
Microsoft announced on October 12, 2026, Azure Maia 300, its third-generation inference accelerator headed for production in 2027, and said it will scale the Maia family to multi-gigawatt deployments as the primary inference system for Microsoft products, targeting up to 40% more tokens per megawatt than leading GPU-based systems.
On 5 October 2026, Iterate Studio launched Lifeboat, an LLM inference engine with built-in confidential computing that, on company tests with an RTX PRO 6000 Blackwell running Qwen 30B-A3B, held 2,048 concurrent sessions and packs two to six times more agent density per GPU than standard engines.
Per Bloomberg (Sep 15) and TrendForce (Sep 16), Meta is testing MTIA 450 (Arke) for roughly H1 2027 deployment and preparing MTIA 500 (Astrid) for late 2027; Broadcom/TSMC inference chips, distinct from the Iris/14 GW story.
NVIDIA and TechTimes coverage report Vera Rubin NVL72 + AgentX at up to 30× throughput/MW vs GB300 on agentic coding (SemiAnalysis), and MLPerf v6.1 gains up to 3.7× on Qwen3-VL and 2.5× on DeepSeek-R1, with early-silicon and harness-governance caveats.
New accelerators promise up to 4× better energy efficiency for LLM inference.