The offer, as described by two people familiar with the matter, was a trade: Washington lets SK Hynix ship American chipmaking equipment into its plant in Dalian, China, and the company builds a new NAND factory in the United States.
Reuters reported the arrangement on the evening of October 14, attributing it to two people with knowledge of the discussions. The next day SK Hynix denied that it was pressuring the American government, saying equipment installation at Dalian is proceeding under existing authorizations and consultations and that it is not seeking to expand capacity there.
Both statements can be true, which is what makes the episode worth reading closely. A company can decline to describe a request as pressure while still making clear, through intermediaries and lawyers, what it would like to receive. What the report describes is a trade in the plain sense: a factory that a government wants, offered in exchange for a permit that a company wants.
The specific permit concerns Solidigm, the American subsidiary that SK Hynix acquired when it bought Intel’s NAND business. Solidigm has applied to move equipment made by Lam Research, Applied Materials and KLA into unused space at the Dalian plant. Those three suppliers are the main vendors of wafer fabrication tools in the United States, and their equipment is covered by export controls that apply to China.
The regulatory backdrop changed earlier this year. The Commerce Department withdrew the arrangement that let Samsung and SK Hynix apply for item-by-item waivers for their China plants, leaving permission only for equipment needed to maintain existing operations. That decision put both companies in the position of running fabs whose tools they can repair but not upgrade.
For a memory maker, the difference between maintaining and upgrading a fab is the difference between a plant and an asset. NAND capacity is measured in layers, and each generation adds more of them. A facility that cannot receive the tools for a denser generation has a fixed life expectancy, and its operators plan accordingly.
Dalian is one of the largest NAND sites in the world, inherited by SK Hynix through the Intel transaction and expanded since. The idle space Solidigm wants to fill was part of that expansion. Filling it with equipment already ordered would raise output without building anything new, which is precisely the kind of move the current rule is written to prevent.
The American interest on the other side of the table is domestic production. The United States has subsidized memory manufacturing through the CHIPS program and has had limited success attracting NAND capacity, which is concentrated in Asia. A new American NAND plant would be a visible result for an administration that has made semiconductor self-sufficiency a stated goal.
Whether SK Hynix would actually build it is a separate question. American memory plants cost more to build and operate than Asian ones, and the company’s capital plans are committed to high-bandwidth memory, the profitable part of the DRAM market. A NAND plant in the United States would be justified by policy rather than by demand, and companies announce such projects in the language of partnership while negotiating the terms in private.
Solidigm has its own agenda. The subsidiary has selected Goldman Sachs and Morgan Stanley to prepare an American initial public offering of about $10 billion, according to reports. An IPO gives the unit its own capital and its own story for investors and gives SK Hynix a way to monetize an asset it has not fully integrated. It also gives the parent a reason to resolve the legal status of Dalian, since a subsidiary preparing to list needs a clear picture of what it owns and can operate.
Analysts who follow the memory sector said the reported exchange, if accurate, reflects the reality that export controls now function as a bargaining position rather than a wall. Every restriction creates a counterpart on the other side, and that counterpart has something to trade: capacity, capital, an announcement, a location. The question for both governments is what they are willing to pay for it.
There is also a timing dimension that the denial does not address. Equipment orders are placed years in advance, and tool lead times for advanced memory are measured in quarters. A permit granted in 2027 does not help a fab in 2027, which means companies raise these questions early and loudly, then describe the resulting conversation as routine.
Samsung faces the same constraint and has handled it differently, directing investment toward American fabrication while keeping its China plants running on existing tools. The comparison is useful because it shows what a company can do without a permit: build new capacity where the rules allow and let old capacity age. SK Hynix’s reported proposal is an attempt to improve on that outcome.
What the report does not establish is who made the proposal, at what level it was discussed, or whether the American side engaged with it. SK Hynix’s denial covers the characterization of the request rather than its existence. The decision on the Solidigm application will be the clearest statement either side makes.
