Korea says Nvidia's 260,000-GPU pledge survives Huang's exit
Korea's science ministry says Nvidia's 260,000-GPU pledge stands after Jensen Huang's exit, but Musk warns many deals were "a handshake with Jensen, not a written contract."

Why it matters
- Nvidia pledged 260,000 GPUs to Korea in October: Samsung, SK Group, Hyundai Motor Group and the government each get 50,000, Naver Cloud 60,000.
- The government says it has secured compute equivalent to about 21,000 Nvidia B200 GPUs, with the rest due by early 2027.
- Nvidia paid 1.34 trillion won ($1 billion) into Naver's Nov. 6 capital increase for a 4.13 percent stake.
- Elon Musk wrote on X that many Nvidia deals were 'a handshake with Jensen, not a written contract.'
- The Ministry of Science and ICT reconfirmed the commitments with Nvidia officials on Friday, a day after Huang resigned as CEO.
Nvidia still plans to deliver 260,000 GPUs to Korea and keep its partnership projects running, the Korean government said, one day after CEO Jensen Huang resigned from the company he built into the world's dominant AI chip supplier.
The Ministry of Science and ICT held a meeting with Nvidia officials on Friday, according to the ministry, and came away with confirmation that the supply plan and the collaborative projects attached to it remain on track. The stakes are considerable: the GPU allocations sit at the center of Korea's national AI compute buildout, and they were negotiated largely during Huang's personal visit to Seoul in October last year.
What exactly has Nvidia promised Korea?
The October pledge splits 260,000 GPUs across five recipients. Samsung Electronics, SK Group, Hyundai Motor Group and the Korean government each secured 50,000 units. Naver Cloud agreed to receive the remaining 60,000.
The government's own allocation follows a two-track structure. Of its 50,000 GPUs, 35,000 will be procured directly, while the remaining 15,000 will be secured through the National AI Computing Center.
Progress is already visible, according to the ministry. Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon said the government has secured computing capacity equivalent to approximately 21,000 Nvidia B200 GPUs, with the remainder expected to arrive by early 2027.
The takeaway: the hardware pipeline is real and partly delivered, but the bulk of the government's compute still sits more than a year out.
What did the meeting actually settle?
The Friday meeting covered more than GPU counts. Bae laid out the confirmed agenda in a Facebook post.
"The second vice minister presided over a meeting with Nvidia and reconfirmed that the company plans to proceed with the supply of 260,000 GPUs and cooperation with Korean companies as previously announced," Bae wrote.
"We also agreed to continue ongoing projects, including connecting Korean AI chips with Nvidia's platform through NVLink, building AI computing infrastructure, conducting joint research on physical AI, fostering talent and bringing Nvidia's developer conference GTC to Korea."
That list matters beyond the raw hardware. The NVLink connection points to Korean AI chips gaining access to Nvidia's platform — a technical foothold for domestic silicon in Nvidia's ecosystem. Physical AI research, talent programs and a Korean edition of the GTC developer conference extend the partnership into research and workforce territory.
The takeaway: the deal isn't just chips; it's a bundle of infrastructure, research and ecosystem commitments, all of which now depend on Nvidia's follow-through.
Where does the Naver deal stand?
The Naver Cloud partnership has already produced tangible money. Nvidia pledged to invest in Naver as part of their joint AI data center business and paid 1.34 trillion won (about $1 billion) through Naver's capital increase on Nov. 6, acquiring a 4.13 percent stake in the Korean mobile platform company.
That investment is the most concrete, hardest-to-unwind piece of the Korea portfolio. A completed equity purchase with a recorded date and stake size doesn't rest on anyone's word.
The takeaway: capital that has already changed hands gives the Naver leg of the partnership a solidity that verbal promises lack.
Why are observers still worried?
The government's reassurance has a soft spot, and Elon Musk put his finger on it. Writing on X on Saturday, the Tesla and SpaceX CEO argued that Huang's way of doing business leaves gaps now that Huang is gone.
"Jensen drove supply chain brilliantly at Nvidia, but many of those deals were a handshake with Jensen, not a written contract," Musk wrote. "Without Jensen, at least some of those critical deals will be lost."
Musk also shared a response from Grok that cited Huang's own podcast interviews from March and April. In those interviews, Huang said Nvidia had no formal legal contract with TSMC despite conducting hundreds of billions of dollars in business with the chipmaker over the past 30 years. Huang described the relationship as governed by a form of "rough justice," with the two companies placing complete trust in each other.
Musk's own experience followed the same pattern. He wrote that Huang always kept his promises, adding: "We'd put together a purchase order at some point later, but I never for a second thought it wouldn't happen."
The picture that emerges is of a company that ran enormous commercial relationships on personal trust rather than paper. That model worked while Huang ran Nvidia. His resignation, described in Korean coverage as a disgraceful departure, removes the individual who held those commitments together.
The takeaway: if Musk is right, the question isn't whether Nvidia wants to honor the Korea deals — it's whether anyone wrote them down.
What happens to handshake deals now?
Korean industry figures are already flagging the review problem. "Whether formal contracts exist in Huang's absence would vary from deal to deal, but commitments made verbally may need to be reviewed to ensure they can result in actual deliveries," an industry official said.
That comment cuts to the mechanism. A pledge announced during a CEO's state visit — 260,000 GPUs, NVLink integration, a GTC conference — may exist as a memorandum or a press release rather than an enforceable agreement. Each commitment could sit at a different point on that spectrum, which means each one carries a different level of risk.
The timing pressure is asymmetric. The government expects the remaining GPUs by early 2027, so slippage wouldn't show up immediately. Naver's completed investment is safe. The open questions surround the delivery schedule, the NVLink integration work and the softer commitments like the GTC conference and talent programs.
The takeaway: the next concrete checkpoint is delivery, and the government will be watching whether words signed off in Huang's era turn into hardware on Korean soil.
For now, Seoul is treating the partnership as intact, and Nvidia's officials have told the ministry exactly that. Whether that holds may depend on how much of the October pledge ever made it out of the handshake stage and into contract form — a question only the two sides' paperwork, or the lack of it, can answer.
Source: feed.koreatimes.co.kr
More from Sophie Lindqvist
Show full bio
Staff writer covering regulation, policy and their impact on artificial intelligence and machine learning at AI In Context.
625 articles
Related articles
- Nvidia, SK Telecom and Rebellions plan AI computing system
- NVIDIA backs Point2 and plans three Korean AI startup investments
- Rebellions Plans R200 Integration for Korean Sovereign AI
- Nvidia Carries $164.5 Billion in Commitments Into CEO Vacuum
- Nvidia 'Honey Trap' Scalarm Puts South Korea's Chip Industry on Espionage Alert