SK Hynix ships next-generation HBM4E memory samples on schedule, reinforcing memory's central role in the AI hardware race.

What happened: SK Hynix, a major memory supplier to Nvidia, has delivered 12-stack HBM4E samples to customers on schedule. The new chips reach a maximum data processing speed of 16 Gbps per pin, with power efficiency improving by more than 20% from previous models and heat resistance up 17% compared with HBM4.

Why it matters: High-bandwidth memory is a critical bottleneck in AI hardware. Faster and more efficient HBM directly supports better AI training and inference (the step where an AI produces an answer) performance. SK Hynix faces competition from Samsung and Micron in supplying this memory to AI processor makers.

What to watch: The samples have been delivered on schedule, signaling that SK Hynix is keeping pace with the demand cycle for next-generation AI chips. Samsung and Micron are also competing aggressively in the same market.