Nvidia Reaffirms 260,000 Advanced GPUs for Korea After Huang Resigns

By Grace Kim

Nvidia has reaffirmed plans to supply 260,000 advanced graphics processing units (GPUs) to South Korea and continue cooperation with Korean companies despite Chief Executive Jensen Huang’s resignation, Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon said on the 13th.

Bae made the announcement in a Facebook post titled “Jensen Huang’s Resignation, but the Cooperation Continues.” He said Nvidia had reaffirmed the commitments at the corporate level, covering the 260,000 advanced GPUs and cooperation with Korean companies as previously announced.

The Ministry of Science and ICT’s second vice minister chaired a review meeting with Nvidia on the morning of the 13th. Nvidia and the ministry agreed to push ahead with existing cooperation projects, including the GPU supply, without disruption.

Nvidia announced on the 12th that Huang had resigned and that Chief Financial Officer Colette Kress had been named interim CEO. The leadership change raised concerns about possible disruption to GPU supplies and cooperation projects in Korea.

The review came a day later. For procurement teams and AI infrastructure operators, the ministry’s message was operational continuity rather than a renegotiation of the announced supply and partnership programs.

What does this change for Korea’s AI buildout?

It preserves the near-term procurement assumption for Korea’s AI computing buildout: Nvidia will proceed with the previously announced 260,000-GPU supply and cooperation with Korean companies, according to Bae. The ministry framed the result as continuity of existing commitments.

The announcement covered two figures serving different functions. The 260,000-GPU commitment covers Nvidia’s planned advanced hardware supply and related work with Korean companies. The government separately targets 50,000 GPUs for artificial intelligence (AI) research, development and policy programs, with direct and computing-center procurement channels.

Beyond the GPU supply, Nvidia and the ministry agreed to continue technology cooperation already under discussion:

• Connect domestic AI chips with Nvidia’s platform through NVLink, the company’s high-speed interconnect technology.
• Build AI computing infrastructure.
• Conduct joint research on physical AI.
• Train talent.
• Continue efforts to bring GTC, Nvidia’s annual developer conference, to Korea.

Government procurement

Bae said the government’s plan to secure 50,000 GPUs remains on track. The plan uses two procurement channels: direct purchases and the National AI Computing Center.

• Direct purchases: About 21,000 of roughly 35,000 GPUs have been secured.
• Scheduled balance: The remaining roughly 14,000 are scheduled for delivery by early 2027.
• Computing-center batch: A separate group of about 15,000 GPUs is being secured through the National AI Computing Center.
• Performance measure: Bae said the government expresses the quantities in computing performance equivalent to Nvidia’s B200 GPU.

The reported quantities use the B200 as a common performance yardstick, while the outstanding direct purchases retain the stated early-2027 delivery schedule.

Deployment priorities

The government plans to make secured GPUs available for AI research and development by companies, universities and research institutes. It also plans to use them for policy projects, including development of a sovereign AI foundation model and the “AI for All” initiative.

Bae said: “The government will work closely with domestic and foreign companies and institutions and continue to monitor GPU supply and demand in both the public and private sectors.”

He added: “We will make sure the GPUs we have secured can be used when needed for AI research and development at companies, universities and research institutes, and for government projects such as the sovereign AI foundation model and AI for All.”

The ministry’s review therefore preserves project continuity while leaving two future procurement milestones in view: roughly 14,000 direct-purchase GPUs due by early 2027 and a separate 15,000-unit computing-center batch still being secured.