Samsung ships next-gen HBM4 chips to AI clients

Samsung’s announcement puts the company in the early wave of HBM4 commercial supply at a time when demand for high-bandwidth memory is being pulled higher by the global buildout of AI data centers. HBM chips are designed to move large volumes of data at high speed between memory and an accelerator, a key bottleneck as model sizes and compute intensity rise.

Samsung said its HBM4 delivers a consistent processing speed of 11.7 gigabits per second, which it described as a 22% increase from its predecessor, HBM3E. The company also said HBM4 can reach a maximum of 13 Gbps, which it framed as a way to reduce data bottlenecks in AI and high-performance computing workloads.

The shipment claim follows a string of recent Samsung statements aimed at showing progress after it was slower to respond to the advanced HBM segment than competitors. Samsung, the world’s largest memory-chip maker, has been working to close a perceived gap versus SK Hynix in earlier HBM generations, including in supplying top-tier parts used in leading AI systems.

Samsung executives have been foregrounding customer feedback as a proof point for readiness. A day earlier, on Feb. 11, 2026, Samsung’s chip-division chief technology officer Song Jai-hyuk said customer feedback on HBM4 had been “very satisfactory” and said Samsung expects strong memory-chip demand to remain in place through this year and into 2027, driven by AI.

Samsung also laid out its next step on the product roadmap. The company said it plans to deliver samples of HBM4E chips in the second half of 2026, signaling that the next iteration is already on schedule while HBM4 begins reaching customers.

The market reacted quickly to the Feb. 12 shipment update. Samsung shares closed up 6.4% on Thursday, while SK Hynix ended up 3.3%, according to the same report.

Samsung’s announcement arrives into a market where the competitive bar has shifted from “who can design” to “who can ship at scale with usable yields.” SK Hynix, widely seen as the current leader in HBM supply, said in January that it aims to maintain an “overwhelming” share in next-generation HBM4 chips, which it said are in volume production, and added that it is targeting HBM4 yields similar to its current-generation HBM3E chips.

Micron is also pressing into the same window. Micron’s chief financial officer has said the company is in high-volume production of HBM4 and has begun customer shipments, according to the Reuters report, adding another contender to a race that was previously framed as largely Samsung versus SK Hynix for top-tier HBM sockets.

The backdrop is a supply chain strained by the pace of AI infrastructure spending. The surge in accelerator deployments has pulled forward demand for the highest-end memory stacks, tightening timelines and making qualification cycles and shipment schedules a key differentiator for suppliers. Samsung’s push to ship HBM4 is aimed at translating its technical claims into share gains in the segment tied most directly to AI capex.

For customers, the immediate question is how quickly HBM4 supply can scale relative to demand for next-generation accelerators and server platforms, and how supplier roadmaps line up with deployment cycles across 2026. Samsung has now put a public marker on initial shipments and a second-half 2026 sampling plan for HBM4E, while SK Hynix and Micron have indicated HBM4 is already in volume production and shipping, setting up a year where execution and yield discipline are likely to decide who captures incremental HBM growth.