SK Hynix and Intel Discuss First U.S. Memory Production
SK Hynix is in discussions with Intel to manufacture RAM chips in the United States for the first time. The talks were reported Sept. 16, 2026, citing sources familiar with the matter. Reuters Nothing has been signed. The structure under discussion is still fluid.
Two models are on the table, according to the same reporting. One would have SK Hynix lease space inside Intel's planned factory in Ohio to produce the chips. The other would create a joint venture that could include cloud-service providers as participants. TechCrunch
SK Hynix responded the same day with a holding statement. It said it is "exploring various options to strengthen its global competitiveness, but no specific plans or arrangements have been finalized." It added that no decisions have been made regarding the two scenarios described in the reporting. Its official newsroom post carried the title "No plans have been confirmed regarding the reported talks between SK hynix and Intel on memory chip production in the U.S." SK Hynix Newsroom
Investors moved quickly on the report. SK Hynix's U.S.-listed shares gained 3.3% following the news. Intel shares also rose initially. Those gains eased after SK Hynix said no plans had been confirmed. Investing.com
The talks are separate from SK Hynix's committed U.S. footprint. The company is building a $3.8 billion advanced packaging and research facility for AI chips in West Lafayette, Indiana, with mass production expected to begin in 2029. That site will take DRAM wafers fabricated in South Korea and package them into HBM chips.
The two companies have transacted before. In 2020, Intel sold its NAND flash-memory business to SK Hynix for $9 billion, a deal that included Intel's NAND SSD business. SK Hynix later received China's approval to take over Intel's NAND business.
The broader context here is worth spelling out for process and supply-chain readers. Packaging in Indiana is not the same as fabricating memory in Ohio. One adds U.S. capacity at the back end, where HBM stacks are assembled for accelerators. The other would put front-end DRAM production on U.S. soil for the first time for SK Hynix.
Looking at what this means for operations, the lease model and the joint venture model solve different problems. A lease inside an Intel shell reduces greenfield risk and time to cleanroom. A joint venture with cloud-service providers would distribute capital intensity and tie output more directly to contracted demand. In my view, that second detail is the more telling one, because it points to buyers wanting firmer visibility into HBM supply rather than leaving procurement to spot cycles.
In my view, if either path proceeds, the practical effect would be incremental U.S. optionality for memory, not autarky. That still matters. Diverse packaging plus domestic DRAM output would give system vendors and accelerator builders more knobs to turn on qualification, logistics and continuity planning. For an industry that has concentrated leading memory fabrication in a small number of sites, added geography is enabling in itself.
