SK Hynix takes Intel's Ohio shell off Intel's hands — the first US memory fab, a counter-cyclical play that re-routes AI memory supply

SK Hynix and Intel are in talks for a deal that would put memory wafers on US soil for the first time — and the Reuters exclusive on September 16, 2026 lays out two specific shapes the agreement could take: [SK Hynix] would either lease part of [Intel]'s long-planned Ohio One campus, or form a joint venture with [Intel] and major cloud firms to underwrite the buildout. Both sides have left the door open since [SK Hynix] walked back an earlier outright-acquisition story on July 22, and Ohio Senator Jon Husted publicly confirmed the talks weeks later, but neither company has signed anything yet.

The Ohio site is the deal's natural anchor. [Intel] broke ground in September 2022 on what was originally framed as a $20B project, then raised the planned investment to more than $28B in March 2024 for two leading-edge fabs in Licking County — a campus sized for roughly eight fab modules in total. Construction has been pushed to 2030–2031 after repeated delays, the US Department of Commerce has awarded [Intel] up to $7.865B in CHIPS Act direct funding (with $1.5B explicitly earmarked for the Ohio One project), and the first module is slated to produce the 14A node and other future [Intel] processes. Reuters says it could not determine which chip types — DRAM, NAND, or HBM — [SK Hynix] would actually produce there, but the front-end memory footprint is the prize: a US front-end fab lets [SK Hynix] skip the back-end-only model it is currently building in Indiana.

[SK Hynix] is not acquiring [Intel]'s Ohio fab, despite the breathless July headlines from JoongAng Daily. The cleaner trade for both sides is a lease or operating partnership: [SK Hynix] inherits a permitted, partially-built, US-funded shell; [Intel] keeps the asset on its books, keeps its CHIPS Act money, and recovers some of the carrying cost on a campus whose first wafer was supposed to ship in 2025. The 1.2-million-pyeong site is large enough for eight fab modules, more square footage than [SK Hynix] could realistically greenfield in five years, and Ohio's power, water, and permitting groundwork is already done.

The economics push [SK Hynix] toward renting too. [SK Hynix] posted KRW 79.32 trillion in Q2 2026 revenue (up 257% year-over-year) and a 76% operating margin — a once-in-a-cycle windfall, but one tied directly to AI-driven HBM pricing that history suggests will normalize. Spending tens of billions of dollars on US land and a greenfield cleanroom would convert that windfall into a multi-decade depreciation anchor at exactly the wrong moment. Leasing a half-built [Intel] campus lets [SK Hynix] reserve a US footprint, qualify for incremental CHIPS funding, and lock in customers like the cloud hyperscalers mentioned in Reuters's venture scenario — without writing the capex check itself.

Through2025 the CHIPS Act story was a logic story: [Intel]'s Ohio and Arizona fabs, Samsung Electronics's $17B Taylor, Texas campus (now planned to add a second fab by 2030 inside a $55B Central Texas investment), and Micron's Idaho and New York DRAM clusters built to qualify for federal grants. The September 16 [SK Hynix]-[Intel] talks are the first time the federal framework is being actively redirected toward a foreign memory leader with no US-listed equity, no US board representation, and no CHIPS Act award of its own — beyond the $450M in direct funding SK Hynix already received for advanced packaging in 2024.

What makes the deal structural, not transactional, is the way it fuses [Intel]'s logic process technology with [SK Hynix]'s memory dominance. Reuters reports that [SK Hynix] is also evaluating [Intel] Foundry for next-generation HBM4E base dies, alongside TSMC's12nm-class process that currently supplies the logic layer under the HBM4 stack. If [SK Hynix] anchors HBM4E base dies at [Intel] 18A or 14A while also operating memory fabs at Ohio, the CHIPS-funded [Intel] campus becomes the substrate for the entire HBM stack — base die plus memory — instead of a logic-only asset. That is a quietly radical reorganization of who builds what in the US.

Front-end memory fabs are tooling-intensive in a way packaging fabs are not. A greenfield DRAM/HBM module needs immersion and EUV lithography, deposition and etch clusters, and a continuous drumbeat of advanced chemicals — a stack where ASML, Applied Materials, Lam Research, Tokyo Electron, and KLA split the wallet. An [SK Hynix] lease at Ohio converts what would have been a 100% [Intel] logic tool order book into a hybrid memory-plus-logic order book starting in 2027–2028, exactly when the equipment complex has been guiding to a digestion year. [ASML] booked EUR 9.33B in Q2 2026 revenue alone, and [Tokyo Electron]'s fiscal Q1 (June 2026) sales hit JPY 732B — both up sharply year-over-year — so the marginal question is whether an Ohio memory tenant accelerates or merely smooths the next leg of demand.

The nearer-term tie-up is already public. [Applied Materials] signed a long-term R&D partnership with [SK Hynix] in March 2026 at its new EPIC Center in Silicon Valley, focused on next-generation DRAM, HBM, and 3D advanced packaging — the same set of problems a US memory fab will need solved. [SK Hynix]'s procurement pipeline has historically centered on Korean and US suppliers, and the supply-chain lift is explicitly downstream of any Ohio deal: roughly 100 Indiana packaging suppliers are already under evaluation, and Ohio would extend that footprint to front-end tools.

The September 16 deal does not break the HBM oligopoly in the near term. [SK Hynix] still controls roughly 50% of the global HBM market, with [Samsung Electronics] climbing toward 33% in Q2 2026 (up from 21% in Q1) after passing Nvidia HBM4 qualification, and [Micron] holding roughly a fifth. Reuters's reporting makes clear that the Ohio site will not produce wafers until at least 2028–2030 — long after the current HBM4 ramp, and possibly not in time to materially affect the HBM4E cycle that follows. Pricing power for HBM4 in2026 and 2027 is therefore unchanged; what changes is the geography of incremental capacity.

[Micron] is the most exposed incumbent. Its Idaho and New York fabs are funded by a $250B+ US capex plan through 2035 and are explicitly being built to address HBM domestic-content requirements from US hyperscalers. A [SK Hynix] Ohio fab directly into the same customer wallet means [Micron] loses some of the 'only US option' premium it has been trading on since the Trump administration expanded the federal stake in domestic memory production in June 2025. [Samsung Electronics] is less exposed — its $17B Taylor campus and follow-on Fab 2 are logic-focused, not memory — but loses the HBM4 narrative advantage if [SK Hynix] can credibly deliver US-made HBM at scale inside Nvidia's Vera Rubin roadmap window.

In the next several quarters, the deal's market impact runs through three near-term readouts: an Ohio site announcement (likely Q4 2026 or Q1 2027 given that the two sides have not begun formal negotiations), incremental disclosure on the HBM4E base-die decision at [Intel]18A, and any signal on which cloud firms — Reuters names them generically — would anchor the joint-venture scenario. Each is a real catalyst for [SK Hynix], [Intel], and the equipment complex, but none changes unit volumes until 2028 at the earliest.

Over a1–3 year horizon, the structural implications are larger. A signed Ohio deal would harden the US government's quiet pivot from funding logic to underwriting memory, force [Micron] and [Samsung Electronics] to either accelerate their own US memory ramps or concede the 'Made in USA' HBM premium, and create a captive US customer base that [SK Hynix] can serve without exporting finished stacks from Korea. Risks cluster around three points: the deal never closes because the parties cannot agree on economics; the federal government withholds CHIPS-related approvals because of foreign-ownership sensitivities around a $45B+ Korean footprint; and the Ohio site remains structurally delayed past 2031, leaving [SK Hynix] carrying lease costs on an asset that ships no wafers.