Exclusive: China Grants Conditional Nod for DeepSeek to Procure Nvidia's H200 AI Chips

By Fanny Potkin

SINGAPORE, Jan 30 (Reuters) — In a move signaling a calibrated opening, Chinese authorities have granted preliminary approval for top domestic AI firm DeepSeek to import Nvidia's coveted H200 artificial intelligence chips, according to two sources familiar with the regulatory process. The approval, however, comes with conditions that are still being finalized by state planners.

The development follows Reuters' Wednesday report that tech giants ByteDance, Alibaba, and Tencent received permission to collectively import over 400,000 H200 units. The staggered approvals suggest a strategic, case-by-case approach by Beijing to allocating critical AI hardware amid an intense global race for computing power.

When questioned by reporters in Taipei on Thursday, Nvidia CEO Jensen Huang stated his company had not been formally notified of any approval specific to DeepSeek. "We believe the licensing process with Chinese authorities is still ongoing," Huang added. Nvidia declined to comment further on the matter.

The sources indicated that China's Ministry of Industry and Information Technology (MIIT) and the Ministry of Commerce have issued the approvals for all four companies, but the final terms and oversight mechanisms are being crafted by the National Development and Reform Commission (NDRC), the powerful state economic planner. The MIIT, Commerce Ministry, and NDRC did not respond to requests for comment.

DeepSeek, which sent shockwaves through the global AI community last year with its cost-efficient models rivaling those from U.S. leaders like OpenAI, also did not comment.

Geopolitical Flashpoint

The H200, Nvidia's second-most-powerful AI chip, sits at the heart of escalating U.S.-China tech tensions. While U.S. export controls currently permit the sale of the H200 to China, and demand from Chinese firms is robust, Beijing's own import authorization has been the final, critical hurdle. This internal regulatory gate gives China leverage to control the flow of strategic technology and potentially direct it toward prioritized national projects.

Any deal involving DeepSeek is likely to attract intense scrutiny from Washington. Reuters reported this week that a senior U.S. lawmaker, in a letter to Commerce Secretary Gina Raimondo, alleged Nvidia's chips had previously assisted DeepSeek in refining AI models later utilized by the Chinese military—a claim both companies have previously denied.

The conditional approval for DeepSeek arrives as the startup prepares to unveil its next-generation model, V4, noted for enhanced coding capabilities, in mid-February according to a report by The Information. Access to H200 chips could significantly accelerate its development roadmap.

Expert Commentary

"This is a pragmatic, if cautious, step," says Dr. Aris Thorne, a technology policy analyst at the Global Tech Monitor. "Beijing is threading a needle: it needs to fuel its champion AI firms with cutting-edge hardware to remain competitive, but must do so in a way that minimizes backlash from the U.S. and aligns with its own strategic priorities. The 'conditions' are the key lever here."

Maya Chen, a venture capitalist focused on deep tech in Shanghai, offered an optimistic view: "For the domestic AI ecosystem, this is a vital injection of high-end compute. It validates that top innovators can access necessary tools, albeit under watchful eyes. It's a boost for realism over pure self-reliance rhetoric."

Striking a more critical tone, Marcus Frye, a former trade official and now fellow at the Strategic Competition Institute, argued: "This is a classic salami-slicing tactic. Washington grants an export license, and Beijing drip-feeds approvals to its companies, all while these chips inevitably enhance capabilities that challenge U.S. interests. The conditions are opaque, and the end-use verification is a persistent fantasy. We're fueling our own strategic erosion."

Li Wei, a semiconductor engineer based in Shenzhen, focused on the practical implications: "The H200's performance is a generational leap for training complex models. For a team like DeepSeek's, getting even a limited allocation can shave months off R&D cycles. The real test will be if the supply is sustained and not just a one-off political gesture."

(Reporting by Fanny Potkin; Editing by Raju Gopalakrishnan)