SK Hynix subsidiary Solidigm is considering building a NAND flash memory factory in the United States, with upstate New York emerging as a leading candidate, as pressure from the Trump administration pushes South Korean chipmakers to expand semiconductor production on American soil.

Three people familiar with the matter told Reuters that Solidigm is evaluating the potential project, which would be separate from discussions between SK Hynix and Intel over producing memory chips at Intel’s Ohio facility.

No final investment decision has been made, and key details of the proposed plant remain under discussion, one of the sources cautioned.

SK Hynix confirmed that Solidigm is examining options but said there was no specific project yet.

“Subsidiary Solidigm is reviewing various options to strengthen its business competitiveness, but no specific plans have been confirmed at this time,” SK Hynix said.

A U.S. manufacturing base would give Solidigm an alternative to its sole NAND production facility in Dalian, China. It could also reduce the company’s exposure to potential U.S. tariffs and restrictions on the export of semiconductor manufacturing equipment to China.

The potential investment comes as Washington intensifies efforts to bring semiconductor production back to the United States, particularly as the rapid expansion of AI data centers tightens global supplies of memory chips.

For SK Hynix, however, building in America presents a difficult cost calculation. Semiconductor production is generally more expensive in the United States than in South Korea, while Seoul and Washington are still negotiating the terms of South Korea’s broader investment commitments and trade relationship with the United States.

Solidigm Could Offer a Lower-Risk US Expansion

Solidigm’s NAND business is expected to provide SK Hynix with a politically easier path into U.S. manufacturing than relocating its most advanced memory technologies.

One of the sources said producing NAND through the U.S. subsidiary could encounter less resistance from Seoul because NAND is viewed as less strategically sensitive than advanced DRAM products, particularly high-bandwidth memory, or HBM, which has become a critical component of AI accelerators.

That has become a matter of interest as Washington and Seoul compete for semiconductor investment.

The South Korean government has been encouraging SK Hynix and Samsung to accelerate plans for a semiconductor manufacturing cluster in the country’s southwest, potentially involving investments worth hundreds of billions of dollars. At the same time, the Trump administration has been pressing Korean and Taiwanese semiconductor companies to expand U.S. manufacturing.

U.S. Commerce Secretary Howard Lutnick has threatened tariffs of up to 100% on South Korean and Taiwanese companies unless they commit to increasing production in the United States.

The pressure puts SK Hynix in the middle of competing industrial policies. Washington wants more chip manufacturing on American soil, while Seoul wants to preserve South Korea’s position as a global semiconductor manufacturing hub.

South Korean President Lee Jae Myung said Seoul would seek to “protect South Korea’s national interests while respecting the views of companies” when asked about pressure on Korean chipmakers to establish factories in the United States.

“It remains a matter of controversy,” Lee said Friday, without elaborating.

The issue is particularly sensitive because South Korea has already committed to a $350 billion investment package in the United States as part of an agreement aimed at securing lower U.S. tariffs.

China Is Expanding Its Memory Ambitions

The potential Solidigm investment also comes as Chinese semiconductor companies are moving deeper into the memory market.

Chinese chipmaker CXMT plans to establish an R&D production line for NAND flash memory at a new facility in Beijing, according to a Reuters report, as it seeks to enter a market dominated by Samsung, SK Hynix and Micron.

China’s push adds another layer of pressure on SK Hynix.

A U.S. plant could help Solidigm diversify its manufacturing footprint while providing some protection against geopolitical restrictions. But building capacity in America would take years and require substantial capital, making it difficult to know what the market will look like when the factory begins production.

That uncertainty is relevant for NAND, a market that has experienced prolonged cycles of oversupply, falling prices and weak profitability.

SK Hynix acquired Solidigm from Intel for $9 billion in 2020 and has since dealt with years of losses at the business following the extended downturn in NAND. The company is now looking for ways to improve Solidigm’s competitiveness. Last month, SK Hynix said it was considering measures to strengthen the unit after reports that Solidigm was exploring a pre-IPO fundraising of about 5 trillion won, or roughly $3.6 billion.

SK Hynix said at the time that no decisions had been made.

A U.S. factory could therefore serve several purposes at once: diversifying Solidigm’s manufacturing base, reducing its exposure to China-related trade restrictions, and giving SK Hynix a stronger position in the U.S. market.

But the economics remain uncertain.

Memory has become increasingly important to the AI infrastructure boom. DRAM provides the working memory used by processors, while NAND is primarily used to store data. The rapid construction of AI data centers has sharply increased demand for memory and put pressure on global supply chains.

SK Hynix has benefited heavily from the surge in demand for HBM, which is used alongside AI processors. That business is different from Solidigm’s NAND operation, but the broader AI infrastructure boom has created a more favorable environment for memory manufacturers.

However, there is concern about demand remaining strong enough to justify new U.S. capacity several years from now. The debate has become more complicated after concerns emerged over the pace of frontier AI development. Any meaningful slowdown in AI infrastructure investment could eventually affect demand for the memory and storage components used in data centers.

SK Hynix’s shares nevertheless rose 6.4% on Friday, compared with a 2.7% gain in the benchmark KOSPI index, as investors responded to broader semiconductor developments.

OpenAI is offering robotics engineers base salaries of up to $500,000 a year as the artificial intelligence company accelerates efforts to build robots capable of operating in the physical world.

The company now has 27 robotics-related positions listed on its careers page, up from 11 in May, according to a review by Business Insider. The roles span hardware engineering, software, machine learning, data collection, prototyping and testing, providing a clearer picture of an effort that goes beyond developing AI models to building the physical systems that will use them.

Publicly advertised base salaries for the positions range from $177,000 to $500,000 a year, excluding equity. The highest-paid opening is for a machine-learning engineer focused on distributed data systems, responsible for the infrastructure needed to process and move large quantities of robotics training data across computers.

The hiring spree comes as OpenAI increases its focus on what the industry calls physical AI, the use of sophisticated AI models to control machines and interact with the physical world.

OpenAI is hiring actuator design engineers to develop the motors and mechanisms that move robotic joints, as well as software, firmware and machine-learning engineers. It is also seeking a laboratory technician to help “develop, build, test, and iterate on robotic systems” and a lawyer dedicated to the robotics team.

Another position would oversee OpenAI’s “data collection facilities,” highlighting one of the central challenges in developing capable robots.

Unlike large language models, which can be trained on enormous quantities of text and other information already available online, robots require data generated through interactions with the physical world. That can include demonstrations of humans performing tasks, robots manipulating objects, and machines responding to changing environments.

Companies developing robots therefore have to generate much of that data themselves or obtain it from external providers. OpenAI’s decision to recruit personnel specifically around data collection suggests that it sees the data pipeline as an important part of its robotics infrastructure.

Guy Hoffman, a Cornell mechanical and aerospace engineering professor who leads the university’s Human-Robot Collaboration and Companionship Lab, reviewed the 27 job postings for Business Insider and said OpenAI appears to be assembling a “custom robot design team.”

The postings also offer an indication of the scale of OpenAI’s ambitions.

One listing says the robotics team is focused on “unlocking general-purpose robotics and pushing towards AGI-level intelligence” while exploring “a broad range of robotics form factors.” Another describes a longer-term vision in which “everyone” could have a personal robot capable of performing whatever tasks they need.

That ambition represents a significant expansion from OpenAI’s earlier robotics work.

OpenAI Moves Beyond Software

CEO Sam Altman has become more direct about the company’s plans to build physical machines.

“We will definitely do a humanoid,” Altman told investor Alex Heath on the Sources podcast earlier this month. “We will do other form factors as well.”

The robotics team is led by Aditya Ramesh, an OpenAI researcher known for creating DALL-E and later working on Sora, the company’s video-generation system. Ramesh has also worked on models designed to simulate aspects of how the physical world operates.

OpenAI previously experimented with robotics before shutting down its robotics project in 2020. That effort became known for a robotic hand capable of solving a Rubik’s Cube. The company’s return to robotics comes as substantial capital flows into physical AI, with technology companies and startups seeking to combine advanced AI models with machines that can perceive, reason and act in physical environments.

OpenAI has already been reported to be training a robotic arm to perform household tasks as part of its humanoid efforts. If the company ultimately develops its own humanoid robots, it will enter a market that includes well-funded efforts from Tesla and Figure AI. Tesla is developing its Optimus humanoid robot, while Figure has previously worked with OpenAI on AI models for robots. OpenAI Startup Fund, a venture fund affiliated with OpenAI, is also an investor in Figure.

The job listings provide few definitive clues about the eventual design of OpenAI’s machines. References to laser range finders and batteries suggest to Hoffman that the company may be considering an untethered mobile robot, although the postings do not establish whether such a machine would travel on wheels or legs.

The hiring pattern also provides clues about areas where OpenAI may be placing less emphasis.

“There is not a lot of electronics work sought, so I don’t think there will be a focus on sensors beyond off-the-shelf cameras, microphones, and laser range finders,” Hoffman said.

Laser range finders use light to measure distance and can help robots map environments and navigate around obstacles.

OpenAI’s strategy appears to be centered on combining its AI expertise with purpose-built robotic systems and the data infrastructure required to train them. That could give the company greater control over the interaction between its models, robotic hardware and training data, rather than relying entirely on third-party manufacturers.

The approach also opens a new cost center. Training physical AI requires not only computing infrastructure but physical facilities, hardware, human demonstrations, testing environments, and large-scale data collection. Altman has pointed to data centers as one potential early use for OpenAI’s robots. Machines operating in such environments could eventually perform physical tasks alongside the company’s broader AI infrastructure.

“There will of course be data center robots that have, like, different form factors,” Altman said. “Someday, I think everyone should have a personal robot.”

The hiring surge indicates that the vision is moving from a largely conceptual ambition toward a broader engineering operation. OpenAI is recruiting across the physical hardware, software, data and testing layers needed to turn AI models into machines that can operate outside a computer screen.