| Mobile Web

Search results for Compute Express Link

Industry

HBM supply hits limits as memory race shifts from bandwidth to utilisation

Limits on expanding high-bandwidth memory are pushing the memory industry to focus on improving resource utilisation in AI servers rather than boosting bandwidth alone. With only about 50 to 60 percent of installed server memory used for computation, companies are pursuing technologies to pool and reallocate capacity, including CXL, PIM and high-bandwidth flash. Supply constraints and rising costs reinforce the shift toward a tiered memory approach combining different memory types.

Telecommunications & Media

Korean telcos\' AI data centre race turns to operations, focusing on cooling, connectivity and power efficiency

Korea’s three telecom operators are stepping up efforts to apply their in-house capabilities to AI data centres as competition shifts from capacity expansion to operational execution. As centres scale up, heat and power consumption rise, putting cooling efficiency, stable power supply and network connectivity at the centre of competitiveness. SK Telecom is building an AI data centre unit and developing server solutions and resource-linking architecture. KT is focusing on data centre interconnection and liquid cooling. LG Uplus is combining group technologies in cooling, power and AI-based infrastructure management.

Industry

AI bottleneck shifts to servers, pushing faster eSSD testing

Investment in AI data centres is expanding from high-bandwidth memory to enterprise SSDs, with prices for 30TB to 64TB drives up more than 50 percent. As capacity and NAND layers rise, burn-in and final tests take longer and inspection becomes a bottleneck, driving demand for higher-parallelism, PCIe 5.0-capable equipment. New testing needs are also emerging around CXL 2.0 memory pooling and switching, prompting shipments of dedicated testers.