SK Hynix's production workers' union has approved a revised pay agreement that raises the cash share of employee profit-sharing bonuses, ending weeks of uncertainty after members rejected the previous proposal by just 25 votes.
Micron has demonstrated what it calls the world's first 512GB DDR5 registered dual in-line memory module (RDIMM) across multiple server platforms, doubling the capacity of the 256GB RDIMM it began sampling in May as next-generation systems demand greater memory capacity and bandwidth.
US memory chip startup Kepler Computing has emerged from stealth after seven years, disclosing US$468 million in total funding and a plan to rethink how high-density memory is manufactured.
AI infrastructure customers are starting to reconsider how much high-bandwidth memory (HBM) needs to be placed alongside accelerators as external memory tiers create new options for storing key-value (KV) cache, according to Astera Labs.
The logic die at the bottom of a high-bandwidth memory (HBM) stack has become the semiconductor industry's newest contested territory. Within weeks, SK Hynix has been reported to be weighing Intel alongside TSMC, Samsung Electronics has been reported to be pairing its own foundry with TSMC, and Micron has confirmed that TSMC will manufacture its next-generation base die. What began as a single-vendor arrangement is turning into a multi-front sourcing contest — and the assumption that a second source automatically means a cheaper one is already being challenged.
DIGITIMES observed that Nvidia's newly announced NVHBM is not a brand-new memory technology, but rather an extension of existing custom HBM (cHBM). It is a managed cHBM design that moves the memory controller, which originally sat on the XPU, into the HBM logic base die, and pairs it with custom physical-layer (PHY) and die-to-die (D2D) interfaces. In strategic terms, Nvidia is extending its moat from GPUs to the memory architecture of custom ASICs, and is trying to define how future custom AI chips connect to memory, hook into scale-up networks, and come together as complete AI racks.
SK Hynix is reportedly considering Intel as an additional manufacturing source for high-bandwidth memory (HBM) base dies beginning with the HBM4E generation, a move that could reduce its reliance on TSMC and give the memory maker greater supply and cost flexibility.
Of the 94 listed Taiwanese semiconductor companies DIGITIMES tracks, 80 grew year-over-year in August 2026 and 14 declined. Median year-over-year growth was 27%. Sixty-three of the 94 posted faster year-over-year growth in August than in July, when the median was 21%.
As memory shortages worsen, Team Group is aggressively deepening its footprint in the industrial control market. Vice general manager Shao-An Hsia noted that with AI applications accelerating their expansion from the cloud to the edge, industrial and edge system manufacturers must pivot away from competing head-on with cloud giants for component supply. Instead, they need to rely on flexible specification adjustments and specialized hardware protection technologies. In driving edge AI adoption, memory suppliers will serve as "icebreakers" during the commercialization process.
Samsung Electronics and SK Hynix are preparing to move DRAM below the 10nm threshold in 2028, with the two memory makers pursuing different cell structures as conventional scaling becomes increasingly difficult.
As memory vendors continue signing long-term agreements (LTAs), the supply-demand imbalance is set to worsen in 2027. Gerry Chen, general manager of Team Group, revealed that module makers have recently received "friendly notices" from upstream memory manufacturers. These notices explicitly state that shipment allocations will be drastically cut in 2027, creating a massive supply gap for DRAM. Upstream manufacturers have reportedly sold out nearly all of their capacity for 2027 through 2028.
After months of tight supply and sharp price increases in 2026, the global memory market is expected to see its supply-demand gap narrow significantly in 2027 as DRAM capacity expands and demand patterns shift, according to Beijing-based Sigmaintell Consulting. The consultancy expects the memory and broader semiconductor markets to continue growing through 2028, but forecasts a potential pullback in 2029 as evolving AI applications reshape demand.
