**Samsung Electronics began shipping its 12-layer HBM3E memory chips on May 28, directly challenging Micron's lead in the high-bandwidth memory market that powers Nvidia's AI accelerators.
Back
Samsung ships 12-layer HBM chip, intensifying Micron rivalry in AI memory
Samsung Electronics began shipping its 12-layer HBM3E memory chips on May 28, directly challenging Micron's lead in the high-bandwidth memory market that powers Nvidia's AI accelerators.
Samsung Electronics Co. shipped its first 12-layer HBM3E memory chips on May 28, escalating a three-way battle with Micron Technology Inc. and SK Hynix Inc. for dominance in the $40 billion AI memory market.
"This is the industry's highest-capacity HBM product in volume production, giving AI customers 50% more memory per stack," said Lee Jung-bae, executive vice president of memory product planning at Samsung, in a statement.
The 12-layer HBM3E chips offer 36 gigabytes of capacity per stack with bandwidth exceeding 1.2 terabytes per second, according to Samsung. That compares with Micron's 8-layer HBM3E at 24 GB per stack, which began volume shipments in early 2025. Samsung's product uses thermal compression non-conductive film bonding, a packaging technique that reduces layer thickness by 20% compared with conventional methods.
The timing puts Samsung ahead of schedule in the HBM race, threatening Micron's recent dominance. Micron shares, which more than doubled over the past 12 months on AI memory demand, face renewed pressure as Samsung and SK Hynix ramp competing products. Samsung's HBM shipments come as Nvidia Corp. — the largest consumer of HBM — plans to invest $150 billion annually in Taiwan's AI ecosystem, CEO Jensen Huang said this week.
HBM Market Heats Up as AI Demand Surges
High-bandwidth memory has become one of the most contested segments in semiconductors, with all three major memory makers racing to supply Nvidia's H200 and B100 series accelerators. HBM stacks multiple DRAM dies vertically, connected through silicon vias, to deliver the massive data throughput required for training large language models.
Samsung's 12-layer HBM3E enters a market where supply has consistently fallen short of demand. Micron said in March its HBM supply for 2025 was already sold out, and SK Hynix — the first to ship HBM3E in volume — has been running at full capacity since late 2024. Samsung's addition of a third qualified supplier could ease pricing pressure for Nvidia and other AI chip buyers while compressing margins for memory manufacturers.
The South Korean chipmaker also resolved a potential disruption last week, reaching an unprecedented deal with its union that averted a strike by promising lucrative bonuses to memory chip workers. The agreement ensures production stability during the critical HBM ramp.
Investment Impact
For investors, the HBM race introduces competing forces. Samsung's entry as a volume supplier could expand the total addressable market for HBM — projected to reach $40 billion by 2027, according to industry estimates — by giving Nvidia and other AI chip designers more sourcing flexibility. But it also threatens the pricing power that Micron and SK Hynix have enjoyed during the supply-constrained period.
Micron trades at roughly 12x forward earnings, a discount to the Philadelphia Semiconductor Index's 18x average, reflecting investor caution about competitive pressure. Samsung's memory chip division, which swung to profit in early 2025 after a prolonged downturn, now faces the challenge of converting HBM volume into sustainable margin expansion.
This article is for informational purposes only and does not constitute investment advice.
