Intel's Xeon 6 platforms will support 8000 MT/s RDIMMs, offering a 25% memory speed increase and 6% lower latency for CPU-side AI and long-context workloads. Separately, Z.AI completed a 1GW AI datacenter in China, utilizing Chinese-made chips for GLM model training. Google's 'Frozen v2' initiative aims for specialized inference chips for Gemini, targeting 2028 deployment to enhance AI serving efficiency.
Intel is enhancing CPU-side AI performance with faster memory, while China is building sovereign AI compute at scale with domestic chips, reducing reliance on restricted foreign silicon.