Samsung’s $34 Billion HBM Bet Raises Memory Glut Risks

Memory & Storage

Samsung's $34 Billion HBM Packaging Bet Lands in 2028

Samsung's ₩46 trillion Onyang HBM packaging program starts operations in June 2028. SK hynix and Micron capacity lands in the same window, raising glut risks for the memory industry.

By
Sophie Lindqvist
Filed
Channel
Memory & Storage
Read
4 min read

Samsung has committed ₩46 trillion, roughly $34 billion, to high-bandwidth memory packaging at its Onyang complex in Asan, South Korea, and the first line will not start operations until June 2028. The five-line program, announced November 12, 2026, allocates ₩40 trillion to equipment and ₩6 trillion to construction. It will do nothing for today's memory shortage.

That timing gap is the real story. Samsung, SK hynix and Micron are all spending aggressively during a supply crunch, and their new capacity arrives together. Industry data points to a convergence of operational milestones in 2028 and 2029.

What is Samsung actually building?

The Onyang project is back-end packaging, not DRAM wafer fabrication. HBM dies come from front-end fabs; packaging plants stack those dies into finished products. Adding packaging lines does not add DRAM wafers. Finished HBM output still depends on qualified dies, packaging tools and acceptable yields, and a shortage in any one link caps shipments.

The numbers show the capital intensity of stacking, bonding, inspection and testing. Equipment takes 87.0% of the program; construction takes 13.0%. The equipment spend phases across all five lines, so the figure is not an immediate commitment to the newly announced building.

The facility adds roughly 31,000 square meters of cleanroom across three floors. Campus floor area grows from 270,000 to 420,000 square meters, an increase of about 55.6%. The June 2028 start comes a year earlier than originally planned. But start of operations is not full-volume production. Meaningful supply could arrive later, pending equipment installation and customer qualification.

How much capacity are rivals adding?

Samsung is not alone. SK hynix announced in August 2026 approximately ₩54.3 trillion for two projects:

  • ₩35.2 trillion for the Yongin Y2 fab, cleanroom opening June 2029
  • ₩19.1 trillion for the Cheongju M17 fab, cleanroom opening December 2028

Both expand long-term DRAM and NAND output for AI infrastructure demand. Micron is building additional DRAM capacity in Idaho and New York under multiyear programs. Those are not dedicated HBM fabs. Their contribution to HBM supply depends on product allocation, technology fit and packaging availability.

Memory supply has always reflected the collective spending of a few producers. When several expand at once, capacity can exceed demand, especially if customers cut inventories or end markets slow. HBM complicates the picture because wafer capacity splits between conventional DRAM and HBM, but the economics are unchanged: new capacity must eventually earn its return.

Is HBM demand growth keeping pace?

The Information Network estimates global HBM revenue grew from $21.1 billion in 2024 to $38.7 billion in 2025, up 83.4%. Projections reach $50.5 billion in 2026 and $63.0 billion in 2027, growth of 30.5% and then 24.8%. The percentage rate moderates even as absolute gains hold near $12.5 billion in 2027 versus $11.8 billion in 2026.

The numbers raise a question. Will AI accelerator demand from Nvidia and others absorb the capacity arriving in 2028 and 2029, or will the industry repeat its pattern of building into shortages and shipping into gluts? Revenue growth also blends volume and pricing, so added supply could pressure prices even as unit shipments climb.

HBM's wafer appetite cuts the other way too. Its complexity reduces effective bit output per fab, which can keep conventional DRAM tight despite higher spending. That supports current pricing. It could change if packaging bottlenecks ease or AI investment slows.

Who carries the risk, and when?

The danger window is 2028 to 2029. Once fabs and equipment are in place, manufacturers face pressure to run them. High fixed costs mean lower utilization raises unit costs, and falling prices squeeze margins quickly. Micron investors in particular should watch capital expenditure against revenue growth and free cash flow.

Equipment suppliers sit on different timing. KLA, Applied Materials and Lam Research recognize revenue as facilities are built and equipped, before the resulting memory hits the market. KLA benefits through process control, inspection and metrology; Applied Materials and Lam supply front-end DRAM tools; specialized packaging vendors gain from stacking and bonding demand. That benefit is cyclical. Excess capacity later means delayed orders and postponed lines.

The current shortage can persist through 2027 because facilities take years to build and qualify. The consequential question is whether the capacity now under construction, led by Samsung's ₩46 trillion program, exceeds demand when it finally switches on.

Source: 247wallst.com

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,001 articles

Related articles

« Previous articleNext article »