The HBM King Is Shopping for a Second Factory — Intel Is the Surprise Candidate

Here's the angle nobody's talking about: the king of AI memory is quietly shopping for a second factory — and the surprise candidate is Intel.

South Korea's Herald Business reported today that SK hynix is weighing Intel Foundry for part of its HBM4E base-die production — a deliberate move away from sole-sourcing TSMC. SK hynix gave its standard no-comment, the polite "difficult to confirm" that rarely kills a story. Intel shares rose on the report.

First, the refresher, because Intel's story gets buried in earnings noise. Intel builds the chips inside most of the world's computers — and it's trying to become the factory other chip companies rent.

That second business, Intel Foundry, is the whole ballgame now: a US-based, US-owned alternative to TSMC for the companies building the AI era's silicon. This report is the strongest signal yet that the strategy is landing.

Here's why it matters for AI. High-bandwidth memory, or HBM, is the specialized memory stacked directly next to Nvidia's and AMD's accelerators, feeding them data faster than anything else on the market — and it's the industry's most talked-about bottleneck.

Whoever controls HBM supply shapes AI timelines. In the HBM4 era, the base die — the bottom layer that talks to the processor — is the key differentiator.

Why should you care? Because the AI buildout has a concentration problem. TSMC makes the world's most advanced chips, and the memory giant that feeds Nvidia and AMD has been handing it the base dies too.

Every company racing to build AI infrastructure is quietly terrified of that single point of failure, and Washington is too — that's why the CHIPS Act bet billions on a US foundry revival. A credible second factory on US soil doesn't just help Intel. It de-risks the entire sector.

SK hynix made base dies in-house through HBM3E, then turned to TSMC for HBM4, which is now in mass production on a 12nm-class process. But TSMC's HBM4 base dies reportedly cost three to four times more than SK hynix's own core dies — and the gap widens for HBM4E.

Passing those costs along is hard, because most HBM is locked into long-term supply agreements. SK hynix eats the difference between what TSMC charges and what customers will pay. Bring in Intel, and it gets cost flexibility plus supply stability: leverage it has never had with TSMC.

The exact Intel node is unconfirmed. Analysts float Intel 3 or 4 for the base dies, or even 18A, with older nodes seen as the more financially sensible pick for a chip that doesn't need bleeding-edge density. Either way, this is real validation — the HBM king is hedging its biggest supplier risk, and Intel is the hedge.

And it's not just base dies. At Hot Chips this year, SK hynix validated designs against Intel's EMIB-T packaging — TSV-based vertical power delivery purpose-built for 3D memory.

Intel also hired former SK hynix CEO Seok-Hee Lee to run advanced packaging; the guy who knows HBM better than almost anyone now sits across the negotiating table. Add it up, and this relationship has been quietly building for a year.

The timing suits Intel. Its 14A process is ahead of plan on defect density — the CFO called it the best progress since 22nm — and customers are reportedly "competing for capacity." Fortinet is in as the first named external foundry customer under CEO Lip-Bu Tan, and Tan put his money where his mouth is, buying about $10 million worth of stock in Intel's $20 billion offering.

The honest counterweight: Intel Foundry revenue grew 31 percent last quarter but still lost about $2.1 billion, and Tan has cut the VP ranks from 450 to 200. This is a turnaround with real scars, and HBM4E base-die volume is a 2027-ish story — nobody should be booking near-term revenue from this.

Still, step back and look at the shape of the trade. Shares have more than tripled off their 52-week low, sit roughly a third below the June high, and rose again on today's report. The market is starting to price Intel as what the AI buildout actually needs: the second foundry, the US onshoring play, the one with packaging chops, a process roadmap, and a CEO buying stock alongside everyone else.

The bottom line: SK hynix doesn't need Intel. It wants options. And when the HBM king starts quietly shopping for a second factory, that's the kind of customer that changes everything.

Disclosure: The Signal holds no position in INTC. Positions may change. This is not financial advice.