Topic
Explore English articles about Data Centers.
Spot prices for DDR4 8Gb rose 3.27% in a week, yet transactions remained sluggish. Supply allocation and falling NAND prices reveal a fragmented demand picture in the memory market.
Microsoft has signed a multi-billion-dollar agreement to use Mistral's European GPU infrastructure, broadening its sourcing of AI compute. The timing of Vera Rubin's deployment and the supply capacity remain undisclosed.
TrendForce forecasts that the NAND supply shortage will ease in the second half of 2027. Higher-density conversions on existing lines, increased production by Chinese manufacturers, and falling demand from smartphones and notebook PCs will support the supply-demand reversal.
Z.ai reportedly completed and partially activated a 1GW-class data center built entirely with domestic AI chips. We examine the possible shift from its IPO-era cloud-procurement policy and the conditions needed to gauge real-world performance.
NVIDIA has released a detailed white paper on the Vera CPU along with SPECrate 2026 integer estimates. With fewer cores, it outperforms EPYC 9755 by about 3%, though official registration and independent verification remain pending.
Rubin cuts wait times across MoE weight transfers, long-context Attention, kernel dependencies, and NVLink synchronization. NVIDIA has fleshed out a design that keeps inference compute units busy and boosts throughput per watt.
We verify ADATA Chairman's prediction that 'the memory shortage will last for the next 10 years' against primary sources from Micron and SK hynix. We break down the structure by which HBM's wafer consumption and capacity expansion plans extending through 2033 will delay a recovery in supply.
SK Group Chairman Chey Tae-won warned that soaring memory prices could shrink device demand, and called for expanded supply ahead of a projected surge in AI-related customer demand in 2027. This article examines SK hynix's 1,100 trillion won investment strategy and the time lag before mass production comes online.
Reports indicate that three major memory makers, including Samsung, have scaled back in-house development of CXL controllers. Product development continues, with a shift toward a division of labor that incorporates externally designed control chips.
Starting from reports that Samsung has begun mass production of V10 V-NAND and supplying it to NVIDIA, this article organizes the changes that NVIDIA CMX—which places KV cache on shared flash—brings to demand and supply for AI NAND.
