Nvidia’s China business moved from waiting room to reopening signal this week. At GTC in San Jose on March 17, CEO Jensen Huang said the company had received purchase orders from Chinese customers for H200 chips and was restarting manufacturing. Reuters then reported, citing sources, that Beijing had approved H200 sales and that Nvidia is preparing a China-saleable version of Groq’s AI chips for inference. Taken together, the reports suggest Nvidia is not just preserving a reduced China presence under export controls. It is trying to rebuild a more meaningful foothold in China’s training and inference stack.
The H200 story is bigger than a routine export-license update
The H200 matters because Reuters described it as Nvidia’s second-most powerful AI chip, not another low-end compromise product built only to stay barely inside the rules. That changes the tone of the China story. For the past year, Nvidia’s China narrative has largely revolved around the H20, the lower-capability part designed to satisfy tighter U.S. restrictions. By contrast, the H200 sits much closer to the center of Nvidia’s data-center AI business. If that chip is moving again, even under a narrower approval window, China is re-entering the company’s higher-value compute conversation.
Huang’s own comments added the clearest commercial signal. CNBC reported that he said Nvidia had received purchase orders and was “restarting our manufacturing,” while Reuters quoted him as saying the company had been licensed for “many customers in China” and had received purchase orders from “many” companies. Tom’s Hardware also highlighted the same point: this was not just a policy discussion or a forward-looking hope. Nvidia was describing orders already in hand and a supply chain that was being fired up again.
That is a notable shift from where the company was only weeks ago. Reuters said Nvidia disclosed in a late-February SEC filing that the U.S. had granted a license in February for “small amounts of H200 products to specific China-based customers.” CNBC separately noted that after Nvidia’s Feb. 25 earnings report, CFO Colette Kress told analysts that a small number of H200 products had been approved for sale to China by the U.S. government, but the company had not yet generated revenue from them. In other words, the market had moved from regulatory possibility to early commercial motion in a short span.
The bottleneck was no longer only in Washington
What makes the Reuters report especially important is that it reframed the last missing hurdle as Beijing rather than Washington alone. According to Reuters, despite strong demand from Chinese firms and U.S. approval for exports, Beijing’s hesitation to allow imports had been the main barrier to H200 shipments into China. That matters because it means Nvidia’s China business was stuck not simply on U.S. export policy, but on a dual-approval system in which Chinese regulators also held practical veto power over whether the chips could move.
Reuters reported that Nvidia had now received licenses for many customers in China from Beijing, citing a source familiar with the matter. It also recalled that in January, China granted preliminary approval to ByteDance, Tencent, Alibaba and AI startup DeepSeek to import H200 chips, though the conditions for China’s approvals were still being finalized at the time. That earlier signal made it clear there was demand from top Chinese buyers. This week’s reporting pushed the story further by indicating that the approval process may finally be turning into something operational.
The language still needs to stay careful. Reuters attributed the Beijing-approval details to sources, not to a full public announcement by Chinese authorities or Nvidia. A spokesperson for the Chinese embassy in Washington told Reuters they were “not aware of the specifics” and directed questions to the competent authorities. One Chinese company source also told Reuters they did not know whether final Chinese government approval had been granted, but that Nvidia had told them they could now place purchase orders. That is exactly why the safest framing is that Reuters reports Beijing’s nod, not that every element of the process has been publicly formalized.
Groq extends the story from training chips into inference
The second Reuters report is what turns this from a single-chip comeback narrative into a broader China AI-market strategy. Reuters said Nvidia is preparing a version of its Groq AI chips that can be sold in China, citing two sources familiar with the matter. The report added that Nvidia licensed technology from Groq, the AI chip startup, late last year in a $17 billion deal and showed a new lineup of products built around those chips at GTC this week. That gives the story a second leg beyond H200.
The H200 line speaks to training or higher-end data-center compute. The Groq line points directly at inference, the phase where AI systems answer questions, generate text, write code or handle user-facing tasks. Reuters said Nvidia plans to use Groq’s chips for inference and, in products shown this week, combine them with forthcoming Vera Rubin chips that cannot be sold in China. For China, that matters because inference is where actual application demand keeps scaling, from enterprise copilots to consumer AI assistants to cloud services.
Reuters also supplied the most important caution labels. The China-ready Groq chips were described by one source as not being downgraded versions or chips built specifically for the Chinese market. Instead, the source said the new variant could be adapted to work with other systems and could be available in May. Those verbs matter. “Could be available” and “expected in May” are defensible. Saying the launch is finalized or fully announced would go beyond what the reporting supports.
This inference angle also comes with a more competitive market structure. Reuters noted that while Nvidia dominates AI training, it faces much tougher competition in inference, including from Chinese companies such as Baidu that already produce their own inference chips. Huang reinforced that broader thesis at GTC when he said “the inference inflection has arrived” and argued that the revenue opportunity for Nvidia AI chips could reach at least $1 trillion through 2027. In other words, Nvidia is not just trying to reopen China for one training chip. It is trying to keep a role in the next phase of AI demand as the market tilts toward inference-heavy deployment.
China is reopening for Nvidia, but only on narrower political terms
The bullish reading is obvious: China once mattered enough to generate 13% of Nvidia’s total revenue, Reuters said, while CNBC noted that the country once accounted for at least one-fifth of Nvidia’s data-center revenue. Those are not side-market numbers. If H200 shipments restart and a Groq-based inference product follows, Nvidia gets back into a market large enough to affect both revenue mix and strategic positioning.
But this is not a return to the old world. CNBC reported that Nvidia’s current-quarter guidance still assumed no data-center revenue from China. The same report said U.S. license requirements remain burdensome, with shipment caps and mandatory testing still in place. Reuters, for its part, showed how much of the current progress still rests on source-based reporting and customer-by-customer approvals. So the cleaner way to read the moment is not “China is fully back.” It is that Nvidia’s China business may no longer be frozen, but it remains tightly managed by politics on both sides.
That nuance is why this week’s developments matter more than a headline about one executive speech. The combination of Huang’s public comments, Reuters’ reporting on Beijing approval, and the Groq inference angle suggests Nvidia is rebuilding a layered China strategy within a much narrower policy corridor. H200 addresses the near-term need to resume higher-tier compute sales. The Groq plan addresses the longer fight over inference infrastructure. Put together, they show Nvidia trying to remain relevant in China even as the rules of engagement have changed.
What changed, and what could happen next
What changed this week is that Nvidia’s China story moved from suspended demand to early evidence of restart. Before this, the company had approvals in limited form, but little visible commercial traction. Now there are reported purchase orders, reported manufacturing restarts, source-based indications of Beijing approval, and a second product line being positioned for China’s inference market. That does not amount to a full reopening, but it is more than symbolic.
What happens next will determine whether this becomes a durable China re-entry or just a narrow regulatory thaw. The first thing to watch is whether H200 shipments move from purchase orders to recognized revenue. The second is whether more named Chinese customers appear beyond the companies Reuters previously tied to preliminary approval. The third is whether the Groq-based China variant really arrives on something close to the reported timetable and whether it can compete against both local Chinese inference chips and other global alternatives.
For now, the most defensible conclusion is also the most interesting one. Nvidia has found a way to push back into China’s AI market without waiting for a broad political reset. If the H200 approvals hold and the Groq inference plan materializes, the company will have done more than restore a sales channel. It will have shown that even under export controls, the contest for China’s AI stack is still open, only more constrained and more strategically selective than before.
Sources
- Reuters — “Nvidia gets Beijing’s nod for H200 chip sales, adapts Groq chip for China, sources say”
https://www.reuters.com/world/china/chinese-authorities-approve-nvidias-h200-ai-chip-sales-source-says-2026-03-18/ - Reuters — “Exclusive: Nvidia preparing Groq chips that can be sold in Chinese market, sources say”
https://www.reuters.com/world/china/nvidia-preparing-groq-chips-that-can-be-sold-chinese-market-sources-say-2026-03-17/ - CNBC — “Jensen Huang says Nvidia has received orders from China and is ‘restarting our manufacturing’”
https://www.cnbc.com/2026/03/17/nvidia-ceo-jensen-huang-says-chipmaker-has-received-orders-from-china.html - Tom’s Hardware — “Jensen says Nvidia has received orders from Chinese customers for H200 GPUs, licenses from US gov’t — H200 manufacturing restarting”
https://www.tomshardware.com/tech-industry/nvidia-has-received-pos-from-chinese-customers
