SK Hynix is reportedly discussing a potential arrangement with Intel to manufacture memory chips in the United States, according to reporting cited by Fortune. The companies are said to be considering the use of an Intel facility in Ohio as SK Hynix looks for additional manufacturing capacity. The discussions come as demand for memory remains closely tied to the expansion of AI data centers and as governments push semiconductor companies to increase production on U.S. soil.
Why memory capacity matters to AI infrastructure
SK Hynix is one of the world’s largest memory manufacturers and is a major supplier of DRAM, NAND and high-bandwidth memory. HBM has become important because modern AI accelerators depend on fast memory to feed large amounts of data into compute processors. Training and inference systems can use large quantities of HBM, making memory supply an important part of the AI infrastructure market.
The potential Intel arrangement would be notable because SK Hynix has not previously manufactured memory chips in the United States at the scale implied by the discussions. The company is already building a packaging facility in Indiana, but wafer fabrication is a different part of the production process. Packaging combines manufactured dies into finished products, while fabrication creates the semiconductor wafers themselves.
Using an existing Intel site could give SK Hynix access to physical infrastructure and an established semiconductor manufacturing environment. It could also provide Intel with another potential customer for its U.S. manufacturing footprint as the company works to expand its foundry business. The precise commercial structure and the type of memory that could be produced have not been finalized in the reporting.
The distinction between DRAM, NAND and HBM matters. DRAM is used as short-term working memory in computers and servers, while NAND flash is commonly used for persistent storage. HBM is a specialized memory technology built for high bandwidth and close integration with processors. AI accelerators use HBM because moving data quickly between memory and compute is critical for performance.
The HBM question
If a U.S. facility were used for advanced memory production, the arrangement could also affect supply-chain planning for AI hardware. Today, large AI systems depend on a relatively concentrated group of semiconductor manufacturers and packaging suppliers. Adding more U.S.-based capacity could give customers another production location, although building and qualifying new semiconductor capacity takes time.
The talks are also taking place against a backdrop of semiconductor trade policy. Fortune reported that SK Hynix faces pressure from both customers and governments to increase output, while the United States continues to use tariffs and industrial policy to encourage domestic manufacturing. Semiconductor companies So have to consider not only cost and technical efficiency but also the geographic location of production.
There are potential political and industrial complications for South Korea as well. SK Hynix is a major Korean technology company, and advanced memory is strategically important to the country’s economy. Moving some manufacturing activity to the United States could help satisfy U.S. customers and policy goals while raising questions about how much advanced production should remain in Korea.
- DRAM is working memory for computers and servers.
- NAND flash is persistent storage.
- HBM is improved for very high bandwidth near AI accelerators.
- Packaging and wafer fabrication are separate semiconductor stages.
Intel’s role is equally important. The company has been investing in its foundry operations and trying to attract external customers to its manufacturing network. A memory customer would broaden the potential use of Intel facilities beyond Intel’s own processors. However, memory manufacturing has specialized requirements, so any agreement would depend on whether Intel’s facilities and process technology can meet SK Hynix’s production needs.
For the AI industry, the most important issue is capacity. GPU and accelerator deployments can be constrained by memory supply and by compute silicon. Increasing HBM output has So become a major priority for the memory manufacturers. SK Hynix, Samsung and Micron are all expanding their advanced memory capabilities as demand from AI data centers increases.
The reported discussions are not the same as a completed manufacturing agreement. Details such as the facility, process technology, product type, investment amount and production timeline remain subject to confirmation. If the companies reach a deal, the next meaningful details will be the exact memory technology involved and when production could begin.
The potential partnership nevertheless shows how AI demand is changing semiconductor geography. Memory is no longer just a component inside a PC or smartphone. It has become a strategic part of data-center infrastructure, and manufacturers are considering where capacity should be located as customers and governments seek more resilient supply chains.
Source: https://fortune.com/2026/09/17/why-sk-hynix-may-tap-intel-make-memory-chips-us/.
Any move into U.S. memory fabrication would also require extensive qualification. Memory products have to meet electrical, thermal and reliability requirements before customers can use them in servers or accelerators at scale. Production cannot simply switch from one factory to another without validating the process and the resulting chips. That is one reason a potential agreement would be a long-term supply-chain project rather than an immediate response to current shortages.
The possible Intel connection is particularly relevant to HBM because AI accelerators are becoming increasingly dependent on high-bandwidth memory. The accelerator and memory have to work together at very high data rates, and advanced packaging connects those components physically. Expanding capacity So requires coordination among the memory manufacturer, foundry or packaging provider and the accelerator company.
The reported SK hynix-Intel discussions remain exploratory, and SK hynix has explicitly said that no specific plan or arrangement has been finalized. Reuters first reported on September 16 that the companies were discussing ways for SK hynix to manufacture memory chips in the United States, including a possible lease at Intel’s Ohio facility or a joint venture involving cloud companies. On September 18, Reuters also reported that SK hynix’s Solidigm unit was considering a U.S. NAND factory, with upstate New York among possible locations.
These are related but distinct developments. SK hynix is a major supplier of DRAM and high-bandwidth memory, while Solidigm, acquired from Intel in 2020, is focused on NAND flash storage. HBM is central to AI accelerators because it provides extremely high memory bandwidth, whereas NAND serves persistent storage. AI data centers require both classes of memory, but the manufacturing processes, economics and supply chains are different.
The U.S. manufacturing discussion is happening against a backdrop of rising AI infrastructure investment and pressure to localize semiconductor production. Domestic manufacturing can reduce exposure to shipping disruptions, tariffs and geopolitical restrictions, but it generally comes with higher construction and operating costs than established Asian production networks. Any new facility would also require qualification, equipment installation and a long production ramp before it could supply customers at meaningful volume.
SK hynix already has a U.S. footprint under development. The company is building an advanced packaging facility in Indiana focused on HBM-related work. A separate memory-fabrication project would So represent a larger expansion of the production chain. The unanswered questions are what products would be made in the United States, which facility would host them, who would finance the build-out and whether the economics work at the expected memory prices.
A second U.S. manufacturing option emerges
On September 18, Reuters separately reported that SK hynix’s U.S.-based Solidigm unit was considering a NAND flash memory factory in the United States, with upstate New York identified as one possible location. The report said the idea is still at an early stage. Solidigm was acquired by SK hynix from Intel in 2020 for $9 billion, making the development especially relevant to Intel’s history in the memory business. A U.S. NAND plant would address persistent-storage supply rather than HBM production, but both developments reflect the same pressure to expand memory capacity closer to major AI infrastructure markets.
Image source: DIGITIMES image credited to SK hynix.