Sources: CXMT plans to start HBM3E production despite yields of just 20% to 30%; its HBM3E performance is closer to that of SK Hynix's previous-gen HBM3 chips
THE SO WHAT
CXMT pushing HBM3E into production at 20–30% yields shows how strategic AI memory has become—China is willing to eat brutal economics to close the gap. For buyers, this means more geopolitical complexity in your HBM supply chain and a longer period where top-tier performance remains scarce and expensive.
READ THE SOURCE
MORE FROM THE WIRE
Deep & Emerging TechPrinceton Digital Sees Strong Asia Data Center Demand
India, Australia, Japan, and South Korea emerging as AI data center hotspots means the gravity of compute—and therefore AI-native products—will tilt further toward these markets. If you operate in-region, treat power, land, and hyperscaler relationships as strategic inputs on par with talent and capital.
Deep & Emerging TechSK Hynix Profit Beat Adds Fuel to AI Rebound Trade
Memory vendors beating expectations are a direct readout on AI infra demand—HBM and related products are now macro drivers, not niche lines. If your roadmap depends on cheaper or more abundant AI compute, factor in that the capital markets are rewarding scarcity, not racing to commoditize it.
Deep & Emerging TechProxima Fusion signs two Japanese deals to build its fusion supply chain
Proxima Fusion locking in Japanese partners years before a plant runs shows fusion is shifting from physics experiment to industrial supply-chain problem. If you touch specialized materials, cryogenics, or high-power systems, there’s a window now to become a qualified supplier before fusion demand spikes.
Deep & Emerging TechIndustry insiders say PCBs and chip substrates are key supply constraints for consumer electronics makers as AI demand increasingly eats up production capacity
PCB and substrate bottlenecks show how AI demand is crowding out "mundane" components that the rest of electronics depends on. Hardware teams need to lock in multi-year supply and consider design-for-availability—your constraint may be a board house in Taiwan, not a GPU allocation from your cloud.