Mon, Oct 12
6 itemsMicrosoft detailed its third-generation Maia AI accelerator and plans to scale in-house silicon to multi-gigawatt deployments, while Samsung reportedly locked up most of its 2027 memory output and Intel and AMD both pushed new data-center hardware.
New5
Microsoft unveils Maia 300, plans multi-gigawatt in-house AI chip push
Microsoft detailed Maia 300, its third-generation Azure AI accelerator, with 96 accelerators per rack and superpods of up to 12,288 chips. It will scale Maia to multi-gigawatt deployments from 2027 while still buying Nvidia and AMD GPUs.
Why it mattersMicrosoft becoming a major buyer and builder of AI silicon shifts accelerator demand and pressures merchant GPU vendors' share of cloud inference.
Samsung reportedly commits 75-80% of 2027 memory supply to long-term deals
Samsung has committed 75-80% of its 2027 memory supply, including HBM, to long-term agreements, Korea Economic Daily reported, with Nvidia, Google and Microsoft among buyers. Samsung has not confirmed the allocation or named customers.
Why it mattersLocking up most 2027 memory output could tighten spot supply and raise costs for buyers without long-term contracts.
Intel's Crescent Island AI accelerator launches as GPU D-Series
Intel's Crescent Island inference accelerator will reach market as the GPU D-Series family, offering 1.3 PFLOPS of FP8 compute and up to 480 GB of LPDDR5X at 1.5 TB/s bandwidth.
Why it mattersIntel's first serious inference accelerator gives data-center buyers another alternative to Nvidia and AMD in AI inference.
AMD shows Helion rack design with up to 32,768 Zen 6 cores
At the OCP Global Summit, AMD unveiled Helion, a rack-scale reference design for agentic AI housing up to 32,768 Zen 6 'Venice' CPU cores, shipping next year in CPU-only and all-in-one configurations.
Why it mattersAMD is positioning CPUs, not just GPUs, as the backbone of rack-scale agentic AI infrastructure.
Nanya posts record quarter as DRAM prices jump over 60%
Nanya Technology reported record third-quarter 2026 earnings, with gross margin hitting 85.7% as DRAM average selling prices rose more than 60% sequentially after low-priced long-term agreements expired.
Why it mattersSurging DRAM margins signal a memory upcycle that raises component costs across servers, PCs and devices.
Follow-ups1
GlobalFoundries unveils 7nm-class FDX Fusion, production planned for 2028
GlobalFoundries is returning to advanced FD-SOI development with its 7nm-class FDX Fusion process, with production planned for 2028, seven years after abandoning FinFET.
Why it mattersA second advanced FD-SOI option could give chip designers an alternative to leading-edge FinFET and gate-all-around nodes.