Samsung's $34 Billion HBM Plant Won't Open Until 2028
Samsung is spending ₩46 trillion on HBM packaging at Onyang, but operations start June 2028. Combined with SK hynix and Micron expansions, glut risk builds for 2029.

Why it matters
- Samsung plans ₩46 trillion (~$34 billion) for HBM packaging at Onyang, with operations beginning June 2028.
- ₩40 trillion (87%) of the program goes to equipment; ₩6 trillion (13%) to construction.
- SK hynix announced ₩54.3 trillion for Yongin Y2 and Cheongju M17 in August 2026.
- The Information Network estimates HBM revenue growth moderating from 83.4% in 2025 to 24.8% in 2027.
- Global HBM revenue is projected at $63.0 billion in 2027, up from $21.1 billion in 2024.
Samsung has committed ₩46 trillion, roughly $34 billion, to a high-bandwidth memory expansion that will not begin operations until June 2028 — doing nothing to relieve the memory shortage it was announced into. The project, a five-line advanced packaging program at the Onyang complex in Asan, South Korea, was disclosed on November 12, 2026. Its real significance is not the size of the check. It is the timing: Samsung, SK hynix, and Micron are all building capacity that lands in the same 2028–2029 window, and that convergence is how memory gluts historically begin.
What is Samsung actually building at Onyang?
The facility is a back-end packaging operation, not a DRAM wafer fab. HBM (high-bandwidth memory, the vertically stacked DRAM used in AI accelerators) is manufactured in two stages: memory dies are produced in front-end wafer fabs, then assembled into stacked packages at facilities like Onyang.
That distinction matters. Adding packaging capacity does not add DRAM wafers. Finished HBM requires qualified dies, packaging equipment, and acceptable yields; a bottleneck in any one caps output regardless of spending elsewhere.
The numbers, per the announcement (with percentage calculations by The Information Network):
- ₩40 trillion — 87.0% — advanced packaging equipment
- ₩6 trillion — 13.0% — facility construction
- ₩46 trillion — total planned investment
The new building adds roughly 31,000 square meters of cleanroom across three floors, expanding the Onyang campus from 270,000 to 420,000 square meters, a 55.6% increase. The June 2028 start is a year earlier than originally planned. Two caveats temper that headline: the equipment spend phases in across the broader five-line program rather than one building, and "beginning of operations" means just that — full-volume production depends on equipment installation and customer qualification schedules that stretch beyond the ribbon-cutting.
The takeaway: Samsung is buying packaging capability for the end of the decade, not supply for this one.
Why won't it fix today's shortage?
Semiconductor facilities require construction, equipment installation, process qualification, and yield optimization before they contribute meaningful commercial output. Samsung frames the investment as a way to compete with SK hynix and Micron in next-generation HBM, where stacking more dies makes wafer thinning, bonding, inspection, and testing progressively harder and yields harder to hold.
The equipment share — 87% of the program — is the signal for the supply chain. Every additional stacked layer raises inspection and metrology intensity, which is why the announcement reads as a tailwind for toolmakers even if memory pricing weakens later.
The takeaway: the current shortage can persist through 2027 because physics and qualification schedules, not capital, set the clock.
Who else is building, and when does it arrive?
Samsung is not alone. In August 2026, SK hynix announced roughly ₩54.3 trillion for two fabrication projects: ₩35.2 trillion for its Yongin Y2 facility and ₩19.1 trillion for Cheongju M17. The company expects the M17 cleanroom to open in December 2028 and Y2 in June 2029, expanding long-term DRAM and NAND capacity for AI infrastructure demand.
Micron is developing additional DRAM capacity in Idaho and New York under multiyear programs. One caution: these are not dedicated HBM fabs. Their contribution to HBM supply depends on product allocation, technology requirements, and access to advanced packaging — so counting them as HBM capacity overstates the case.
| Company | Project | Investment | Milestone |
|---|---|---|---|
| Samsung | Onyang HBM packaging | ₩46.0T | June 2028 operations |
| SK hynix | Cheongju M17 | ₩19.1T | Dec. 2028 cleanroom |
| SK hynix | Yongin Y2 | ₩35.2T | June 2029 cleanroom |
| Micron | Idaho DRAM expansion | Multiyear | Phased ramp |
| Micron | New York DRAM expansion | Multiyear | Longer-term capacity |
Memory supply has always hinged on the collective capital decisions of a handful of producers. When several expand simultaneously, capacity additions can exceed demand, particularly if customers draw down inventories or end-market growth slows. Cleanroom opening dates are not production dates — a distinction the table's own sourcing note makes.
The takeaway: the industry has synchronized its building schedule, which is precisely the pattern that preceded past downturns.
Is HBM demand growing fast enough to absorb it?
The Information Network estimates global HBM revenue grew from $21.1 billion in 2024 to $38.7 billion in 2025 — 83.4% growth — with projections of $50.5 billion in 2026 and $63.0 billion in 2027. The market nearly triples in three years.
The growth rate, however, decelerates sharply: 305.8% in 2024, 83.4% in 2025, 30.5% in 2026, and a projected 24.8% in 2027. In absolute terms the market still adds $11.8 billion in 2026 and $12.5 billion in 2027, so deceleration is not decline. But the question manufacturers must answer is whether a market compounding in the mid-20s can absorb the volume of capacity scheduled to switch on in 2028 and 2029.
Two further complications sit inside those revenue figures. First, HBM revenue reflects both volumes and average selling prices — strong pricing can prop up revenue growth even as unit growth slows, and additional supply can compress it. Second, HBM consumes substantially more wafer resources per bit than conventional DRAM, which is part of why conventional DRAM has stayed tight during the buildout. That support could erode if HBM production gets more efficient, packaging bottlenecks ease, or AI infrastructure investment slows.
The takeaway: demand is still growing, but at a decelerating rate that must be weighed against fixed delivery dates on billions of dollars of capacity.
Where does the financial risk actually sit?
The critical window is 2028–2029. Additional capacity does not mechanically produce oversupply — manufacturers can delay equipment installation, cut utilization, shift product mix, or postpone later phases. The deeper risk is that several producers will have sunk capital before demand conditions are clear. Once fabs are built and tools installed, high fixed costs pressure companies to run the assets, even at falling prices and shrinking margins.
The exposure splits unevenly across the value chain:
- Memory makers (Samsung, SK hynix, Micron) face pricing and utilization risk once new capacity ramps; for Micron investors specifically, the relationship between capex, revenue growth, and free cash flow becomes the metric to watch.
- Equipment suppliers — KLA in process control, inspection, and metrology; Applied Materials and Lam Research in front-end DRAM; specialized packaging toolmakers in stacking and bonding — recognize revenue during the build phase, before the resulting memory hits the market.
That timing gap means toolmakers can profit from this cycle even if memory pricing deteriorates afterward. But equipment demand is itself cyclical: if oversupply arrives, manufacturers cut orders and delay lines, hitting suppliers after the investment phase ends.
The takeaway: the money is safest earliest in the chain — and the risk accumulates the closer you sit to selling the chips themselves.
Investor takeaway
Samsung's ₩46 trillion commitment confirms the scale of capital now required to compete in advanced HBM packaging. It strengthens Samsung's position against SK hynix and Micron for the next product generations, and it will do essentially nothing for the shortage persisting through 2027. AI has permanently raised the importance of HBM and advanced DRAM, but it has not repealed the memory industry's sensitivity to synchronized capacity expansion.
The announcement is therefore both an opportunity and a warning: equipment suppliers collect during the build, memory producers shoulder the risk when the capacity ships. The industry's next turning point will not come during today's shortage — it will come when the capacity built to end it finally arrives, and the market discovers whether AI demand was large enough to swallow it.
Source: 247wallst.com
More from James Calloway
Show full bio
News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.
604 articles
Related articles
- Samsung Breaks Ground on HBM Packaging Plant, Fast-Tracked to 2028
- Inside Samsung's Massive Pyeongtaek Chip Complex, Where AI Demand Fuels a Building Boom
- TSMC Is Weighing 3D DRAM Production at NVIDIA's Request
- CXMT's Revenue Jumped Sevenfold. China's Memory Chipmakers Are Closing In
- Samsung Nears $146.6 Billion in Cash, a First for Korean Companies