SK Hynix Wants to Make Memory Chips in Ohio. The Real Story Is the Three Governments in the Room.

Here's the version of this story that gets the headlines: SK Hynix, the world's dominant maker of AI memory chips, is in talks with Intel to fabricate chips on American soil for the first time. AI data centers are starving for high-bandwidth memory, Washington wants chipmaking back, and Intel has an idle-ish $100 billion question mark in Ohio. Everyone shakes hands, ribbons get cut, shortages end. That tidy version dies on the first hard question, which is: who actually has to say yes?

The answer, according to three people who spoke to Reuters on condition of anonymity, is that this is not a bilateral negotiation at all. It is a triangular one, with two governments sitting at the table alongside two chipmakers — and a third pressure from the market itself. That's the real story.

The deal nobody can quite describe yet

Reuters reported Sept 16 that SK Hynix and Intel are exploring two structures. Under one, SK Hynix leases part of the chipmaking complex Intel began building in Ohio in 2022. Under the other, it forms a venture with Intel and major cloud firms — hyperscalers who are exactly the customers desperate to lock in memory supply. One source called the talks exploratory and stressed no decisions have been made.

The product line is undecided too. Reuters could not learn whether the Ohio output would be DRAM, NAND, or the high-bandwidth memory (HBM) that SK Hynix effectively owns as the AI era's leading supplier. South Korea's MK, corroborating the wire, was blunt that HBM and advanced DRAM manufacturing are designated as "National Core Technology" in Seoul. That one word, national, is where the story stops behaving like a supply chain story.

The Ohio mirage

Intel announced in 2022 it would invest up to $100 billion to build potentially the world's largest chipmaking complex in Ohio, with production slated for 2025. The site's two plants have been delayed to 2030 and 2031. For a chipmaker under enormous financial strain — and now part-owned by the U.S. government — an operating partner that could put a real product on those lines looks like a lifeline, not a favor. Semafor reported in July that Intel was actively seeking a partner to run the fab, and that SK Hynix was among the names on the shortlist, alongside the caveat that the interest was very early.

The numbers behind the need are staggering in both directions. Intel shares are up 161% over the past year on the strength of Washington's strategic stake and the reshoring push. Apple CEO Tim Cook told the WSJ that supplier price increases from companies like SK Hynix had "become unsustainable" for the iPhone maker. The AI memory crunch is real enough that customers are now trying to buy their way into the fab line itself.

The Seoul veto nobody talks about

Here is the part that doesn't make the press release. Manufacturing in the U.S. costs materially more than in South Korea — higher labour and construction costs, plus a supply chain that lives in Asia. But cost is not the hard constraint. The hard constraint is Seoul's Industrial Technology Protection Act, which requires government review before "national core technology" like HBM manufacturing moves overseas. South Korea's trade ministry confirmed to Reuters that any decision is the company's call, but that a transfer of national core technology would trigger legal review.

That review is not a rubber stamp. Seoul has been pushing SK Hynix and rival Samsung to accelerate a new cluster of chip plants in the country's southwest — the Yongin mega-cluster — and it wants to spend time and money there, not see the crown jewels exported. The politics run through the balance sheet too: of the $350 billion investment commitment South Korea made to Washington last year, $150 billion is earmarked for shipbuilding and the remaining $200 billion remains undecided. Two sources told Reuters that Seoul is using SK Hynix's U.S. investment as leverage in those broader talks, while Commerce Secretary Howard Lutnick has threatened up to 100% tariffs on South Korean and Taiwanese firms that don't commit to producing more on American soil.

The only reason this is even close to making sense for SK Hynix is the push-pull described by SK Group chairman Chey Tae-won in July: "We are facing significant pressure from customers and governments around the world to produce locally. If possible, we need to build a plant in the United States."

The company isn't new to U.S. soil by half measures. It listed on the Nasdaq in July, and in August broke ground on a $4 billion advanced packaging facility in Indiana, where SK Hynix CEO Kwak Noh-jung laid out a vision of HBM wafers made in Korea, shipped to Indiana for packaging and testing, and delivered to customers from a fully integrated U.S. supply chain by 2030. That packaging plant is a toe in the water. Ohio would be the whole leg.

Why base die matters

There's a quieter reason this pairing is less bizarre than it looks, reported by South Korea's Herald Business at the end of August. SK Hynix is considering using Intel Foundry — not just TSMC — to make the base die chips that anchor its next-generation HBM. Former SK Hynix CEO Lee Seok-hee joined Intel as a senior vice president for its foundry business in June, and Intel CEO Lip-Bu Tan has made no secret of wanting to integrate compute and memory. Diversifying foundry supply beyond TSMC, which is at full capacity and planning price increases on advanced nodes, gives SK Hynix leverage and gives Intel's loss-making foundry a marquee external customer. That thread is still speculative, but it explains why two companies that look like rivals keep finding reasons to sit in the same room.

The tension that decides it

Strip away the deal mechanics and you're left with the actual strategic tension. SK Hynix needs to satisfy three masters at once: Washington wants U.S. output and is brandishing 100% tariffs; Seoul wants to keep its crown-jewel technology and leverage the investment; and the market wants capacity yesterday, with customers like Apple and the hyperscalers effectively begging for memory. Nobody gets everything they want. If SK Hynix ships HBM-making to Ohio, it risks the Seoul review and the political fallout. If it ships only commodity DRAM, it risks not pleasing Washington enough to matter. If it ships nothing, it keeps betting the whole memory future on Asia in a world that is actively hostile to that bet.

The genuinely hard question for anyone watching this space isn't whether a deal gets signed. It's which master blinks first.

What this means for anyone buying hardware

For an industry that has spent two years chasing AI compute, the memory constraint is the part of the supply chain that keeps quietly reshaping timelines, prices, and roadmaps. SK Hynix's Indiana plant won't even begin HBM mass production until the second half of 2029, and the company itself sees memory shortages persisting through 2030. That is the production-grade reality: even the most optimistic reshoring story measures in years, not quarters. Enterprises planning AI infrastructure are planning against a shortage that the politics may delay further, not one this Ohio deal immediately solves.

The supply chain complexity behind standing up memory capacity inside a three-government standoff is precisely the kind of problem that doesn't show up in a spec sheet. At DMC, we help hardware companies navigate exactly these constraints — sourcing strategy, cost and tariff modeling, and production ramp planning when demand outruns the fab. If your roadmap assumes memory will be there when you need it, stress-test that assumption before the next negotiation writes it into stone. Want help modeling your hardware dependencies? Let's talk.