SK hynix Breaks Ground on $4B Indiana Fab: First 'Made in USA' HBM Arrives in 2029 — and the Memory Shortage Won't End Until 2030

On Thursday, August 27, SK hynix held a groundbreaking ceremony at Holloway Gymnasium on Purdue University’s campus in West Lafayette, Indiana, for a facility that has no precedent in the United States: the country’s first high-bandwidth memory (HBM) advanced packaging plant. The company will invest over $4 billion in the site, open its cleanroom by October 2028, and begin mass production of next-generation HBM in the second half of 2029 — the first “Made in USA” HBM in history, delivered into a market that SK hynix’s own CEO now expects to remain supply-constrained through the end of 2030.

The ceremony drew an unusually political guest list for a memory chip factory: Indiana Governor Mike Braun, US Senator Todd Young, Purdue President Mitch Daniels, South Korean Ambassador to the US Kang Kyung-wha, former USFK commander Robert Abrams, and SK hynix CEO Kwak Noh-jung. Senator Young tied the project explicitly to national strategy: “This project will create good-paying jobs, strengthen our national and economic security, and ensure the technologies of the future are developed and manufactured here in the United States. I was proud to help make this investment possible through the CHIPS and Science Act.”

Why packaging, and why Indiana

HBM is the memory that makes AI accelerators possible. Stacking DRAM dies vertically and placing them millimeters from the GPU’s compute silicon gives models the enormous memory bandwidth that inference requires — and the manufacturing step that creates those stacks, advanced packaging, has become the tightest bottleneck in the AI supply chain. Every leading-edge GPU shipped next year will carry HBM built and packaged in East Asia, overwhelmingly in South Korea. Indiana changes that equation for the first time.

The plant’s design reflects a pragmatic division of labor rather than full onshoring. Cutting-edge wafers will still be produced in South Korea and flown to Indiana, where they undergo advanced packaging and testing before being supplied to US customers as finished HBM. The US gets the strategically sensitive back-end capacity and the jobs; Korea keeps the front-end fab economics. Alongside the production lines, SK hynix will build an “Advanced Packaging R&D Testbed” where customers, universities, and partners can co-develop next-generation packaging technologies and run rapid prototyping and performance validation — an arrangement formalized in an MOU with Purdue University signed at the ceremony.

The site itself is a 133-acre plot in Purdue Research Park. The project is expected to support around 1,000 direct jobs at the fab during commercial operations, roughly 7,000 direct and indirect jobs across construction and the broader supply chain, and a supplier ecosystem of more than 100 companies being considered for materials, components, and equipment. The CHIPS and Science Act underpins the deal: NIST’s CHIPS Program Office lists up to $458 million in direct funding for the roughly $3.87 billion West Lafayette investment.

There was also a quieter diplomatic touch. Indiana deployed roughly 140,000 troops during the Korean War, and SK Group used the ceremony to join the ‘USFK Veterans Job Platform’ operated by the Korea Chamber of Commerce and Industry and the Korea-US Alliance Foundation, committing to recruit discharged US Forces Korea service members.

The real headline: shortage through 2030

If the factory is the setting, the forecast is the story. Speaking to reporters after the ceremony — as reported by the Korea Herald and Reuters — CEO Kwak Noh-jung said the global memory shortage will persist through the end of 2030, with supply-demand only beginning to balance after that point. “After 2030, we expect supply-demand to become more balanced, with the overall AI industry continuing to grow,” he said.

That is a strikingly long runway of scarcity from the executive with the best seat in the house. SK hynix is the dominant HBM supplier to Nvidia and effectively the bellwether of AI memory economics. Its CEO saying publicly that he sees “no clear sign of a downturn in the coming years” — and that any slowdown would ease gradually rather than collapse as in past memory cycles — is the clearest signal yet that hyperscaler demand has outrun the entire industry’s capacity to respond, even with SK hynix itself committing $38 billion to two new fabs in Korea.

Kwak’s structural argument deserves attention. The memory business, he said, is being reshaped from standardized commodity chips toward partially or fully customized products — HBM4, for instance, embeds logic circuitry in its base die, blurring the boundary between memory and logic. Customization means closer coordination with customers, which in turn gives chipmakers better demand visibility than the industry has ever had. In his telling, the boom-and-bust cycles that defined memory for decades should become less severe, because the industry is no longer building blindly against a commodity spot market.

He also kept doors deliberately open. On further US investment: “We are always open, but nothing has been decided,” citing power, water, and government incentives as the criteria for any additional site. On Solidigm, the company’s NAND subsidiary: a US IPO “remained under review.” On Kioxia, where an SK hynix-backed vehicle became the largest shareholder earlier in August: “We are carefully looking at how we can grow the market together.” And a newly established AI Co. will serve as an investment vehicle for startups aligned with SK hynix’s next businesses.

What it means

For AI builders, the message is blunt: memory is the constraint that outlasts the GPU race. The industry has spent two years fixated on accelerator supply — Nvidia’s racks, TSMC’s CoWoS capacity, gigawatt data centers — but HBM sits underneath all of it. A CEO-level forecast of shortage through 2030 implies that whoever holds memory allocation holds the real throttle on AI deployment through the end of the decade. Expect long-term supply agreements, prepayments, and memory-adjacent M&A to intensify.

For US industrial policy, Indiana is a proof point with fine print. The CHIPS Act has largely succeeded at attracting front-end logic fabs; advanced packaging was the acknowledged gap, and this is the first HBM packaging capacity on US soil. But the wafers still come from Korea, the equipment and materials ecosystem is still being assembled, and the first product is three years away. “Made in USA” here means packaged-in-USA — a meaningful but partial decoupling of the AI supply chain.

For Purdue and the Midwest, the testbed is the long game. A co-located R&D line where next-generation packaging can be prototyped with a tier-one manufacturer is a rare asset for a US university, and SK hynix explicitly plans to recruit talent across the Midwest. If it works, West Lafayette becomes something the US has lacked: a packaging engineering cluster.

For the market, note what wasn’t said. No new mega-fab announcement for the US, no Solidigm IPO confirmation, no Kioxia integration plan — just openness to all three. SK hynix is keeping optionality while its core business prints money in a shortage it expects to last four more years.

The Indiana fab breaks ground amid intensifying competition to expand AI infrastructure. SK hynix frames it as “beginning a new future for US-Korea AI.” The more practical reading: the most critical component of the AI buildout is now being partially built where most of its customers are — and its maker believes demand will still be waiting when it opens in 2029.