HBM Shortages Are Pushing Memory Production Toward the US
SK Hynix and Intel are discussing memory production cooperation in the US. No deal, no product, is set yet. But the early talks matter because AI is starting to rewrite chipmaking's oldest rule: build where it's cheapest becomes build where supply won't break.
The formula that ran the industry for 30 years is cracking
The semiconductor industry has long been globalization's star pupil. Design stayed in the US. Equipment came from the US, Europe and Japan. Advanced manufacturing went to Korea and Taiwan. Assembly and testing spread across Asia. The logic was simple: put each step wherever it could be done best and cheapest.
AI is adding a new line item to that math. Companies now have to weigh not just production cost but supply chain resilience, national security, export controls, distance to customers, and the opportunity cost of a supply disruption.
Old formula: Lowest production cost → maximum efficiency
AI-era formula: Production cost + supply resilience + regulatory risk + customer access → total supply value
This shift is not a story about making chips cheaper in the US. It's probably the opposite. Reuters has noted that US production costs could run higher. The real question is when the cost of a supply cutoff outweighs that extra expense.
What this news is not, first
Reuters reported on September 16, citing multiple sources, that SK Hynix and Intel are discussing producing memory chips in the US. The option gaining the most traction is leasing part of Intel's Ohio facility, with a joint venture structure also under discussion that could include Intel, SK Hynix, and a major cloud customer seeking stable memory supply.
But the talks are at an early stage. SK Hynix has said it is reviewing several production-site options to strengthen its competitiveness, but nothing has been decided. More importantly, it hasn't been decided what kind of memory would be made. Whether it's HBM, commodity DRAM, or NAND changes the investment's meaning and equipment requirements considerably.
FACT BOUNDARY: What SK Hynix denied in July was "acquiring" Intel's Ohio site and fab. The new September 16 report is about "production cooperation, including a lease or joint venture." These are not the same deal.
SK Hynix's 6-K filing with the US SEC on July 22 makes that distinction clear. At the time, the company said, regarding reports it would acquire Intel's Ohio land and manufacturing facility, that it had "not pursued or decided on an acquisition." Reading that filing as a denial of today's lease or joint-venture possibility stretches its scope.
It's also different from the HBM site SK Hynix is already building in the US
SK Hynix is already building an HBM site in West Lafayette, Indiana, with an investment of more than $4 billion. The company held a groundbreaking in August 2026 and has described the facility as its US production base for next-generation HBM.
But the core of that project is advanced packaging. HBM stacks multiple DRAM dies vertically to pair with GPUs, so packaging itself is a critical process there. Still, that's a different role from a front-end fab that produces DRAM wafers from scratch.
That's why the Ohio talks would carry more weight if they actually lead to front-end memory production. It would mark a shift from "packaging HBM" in the US to "making the memory die itself" in the US, a deeper supply chain realignment. But right now, neither the product nor the process has been confirmed.
Why supply resilience commands a bigger premium for HBM than ordinary memory
HBM is increasingly a different product from traditional commodity memory. It's not a case of a memory maker producing chips and releasing them into the market. HBM has to match specific AI accelerators on performance, power and thermal characteristics, connect to advanced packaging processes, and go through customer qualification that takes time.
That's why HBM's economics are hard to calculate on a per-chip price basis alone. If an HBM shortage keeps a GPU worth tens of thousands of dollars from shipping, or delays installation of an AI server rack worth millions, the bigger cost to the customer isn't the price of memory. It's the downtime of the server itself.
| Metric | Detail |
|---|---|
| $4B+ | SK Hynix's investment in its Indiana HBM advanced packaging site |
| $28B+ | Intel's proposed investment for the first two fabs of Ohio One Phase 1 |
| 2030-31 | Intel's current target range for Ohio One Mod 1 to begin operations |
This is where the idea of a "resilience premium" comes in. It's less a line item on an official price sheet than an economic concept: customers may accept extra cost to lock in supply security. It can show up as long-term purchase agreements, upfront payments, co-investment, or minimum purchase guarantees.
The more interesting scenario is a cloud company putting money into the fab itself
The biggest structural signal in this report is the possibility that a major cloud company could join the joint venture. No participant or investment structure has been set, but if it happens, it would deepen vertical integration in the AI supply chain by another notch.
Hyperscalers already buy GPUs in bulk, build data centers themselves, sign long-term power contracts, and put capital into nuclear and power generation projects. If they also put capital into memory manufacturing equipment, companies like Microsoft, Amazon, Google and Meta become harder to describe as simply "software and cloud server operators." They start to look more like industrial infrastructure operators securing power and chipmaking capacity to keep AI supply flowing.
That said, there isn't yet confirmed evidence to put OpenAI or Anthropic in the same category as direct fab investors. The Reuters report, at this stage, points to the possible participation of "major cloud firms" or memory customers.
For Intel, this is more complicated than simply "using up spare capacity"
Intel is building two advanced fabs at Ohio One with a Phase 1 investment of more than $28 billion. But in February 2025, the company pushed back the timeline, citing capital efficiency and demand. Mod 1 is now officially slated for completion in 2030 with operations starting in 2030-2031, and Mod 2 for completion in 2031 with operations starting in 2032.
If an outside partner like SK Hynix comes in, it could improve the economics of Intel's massive site and infrastructure and broaden its customer base. But that shouldn't be read directly as evidence of an "Intel turnaround." It's not yet known which facility would be owned by whom, who would buy the equipment, which processes would be used, or whether Intel would collect rent or manufacturing service revenue.
What investors should watch instead is whether the Ohio asset actually turns into a platform that draws in outside customers' capital and production demand. Only then does the investment case for improved Intel asset utilization start to get verified with numbers.
Who gains, and who bears the cost
| Party | Possible upside | What to watch alongside it |
|---|---|---|
| SK Hynix | Shorter distance to US customers, supply chain diversification, potential for long-term demand | US production costs, additional capex, questions around moving core DRAM technology overseas |
| Intel | Better utilization of Ohio assets, new outside customers, shared capital burden | If the deal structure is just a simple lease, it's a separate question from actual manufacturing competitiveness |
| Hyperscalers | Supply security for HBM/DRAM, reduced geopolitical risk | Higher capital intensity from fab investment, long-term dependence on a specific supplier |
| Korea | Stronger US market access for Korean firms | Industrial policy concerns over dispersing core memory technology and production capacity overseas |
Insight Times Editorial Desk
