Nvidia's China problem hasn't gone away. The latest black-market prices show buyers still want the banned chips badly enough to pay more than double, while Beijing keeps pushing them toward Huawei instead.
Jensen Huang has been warning Washington about this for months, and the numbers are getting harder to wave away. Nvidia's chief executive said at Citadel Securities' Future of Global Markets event last year that the company had gone from roughly 95% of China's advanced AI accelerator market to zero after US export controls tightened. He called the policy "already largely backfired." You don't have to like that argument to see the damage sitting in plain sight.
The Financial Times reported today that Nvidia's banned AI hardware is now selling on China's black market at prices that make the official restrictions look less like a wall and more like a toll booth. The DGX B300 system, which retails in the US for about $400,000, has climbed to more than Rmb8 million, or about $1.1 million, in China, according to traders interviewed by the FT. Six months ago, the same system was around Rmb4 million. Nvidia's RTX 6000 Pro workstation chip has jumped from about Rmb50,000 at the start of the year to as much as Rmb130,000.
Those aren't casual markups. They're proof that a parallel market is still doing its job, even after Washington, Taiwan and Malaysia tightened enforcement around re-export routes. The FT's reporting also noted that older A100 chips are being repurposed and that GPU rental prices in China now match or exceed US levels, a sharp change from the period when smuggled supply made Chinese compute cheaper. If you're reading the AI infrastructure market, that price signal matters more than the policy language.
The Super Micro case showed how large the channel can get. Reuters reported in March that US prosecutors charged three people tied to Super Micro Computer, including co-founder Yih-Shyan "Wally" Liaw, with helping divert about $2.5 billion of Nvidia-powered servers to China. The company itself was not charged, and Reuters later reported that Liaw resigned from Super Micro's board after the indictment. Prosecutors said the alleged operation used a Southeast Asian front company and falsified routing to move restricted hardware to Chinese buyers.
That case should have been a warning that chip controls are only as strong as the enforcement network underneath them. AI servers are valuable, movable and hard to police once they begin passing through brokers, cloud resellers and third countries. A policy that assumes Chinese buyers will simply stop buying Nvidia chips is not serious. They pay more, wait longer, rent capacity, strip older systems for parts, or use intermediaries. Frankly, that is exactly what a determined market does.
Huawei is the other side of the story. Tom's Hardware reported in late May that Huawei rotating chairman Xu Zhijun publicly thanked the US for its pressure, saying that if the United States hadn't forced China and its companies, they wouldn't have built this way. The comment came after questions about Huawei's LogicFolding chip architecture, which the company has presented as a way to reorganize chip layouts and improve data movement. That doesn't make Huawei equal to Nvidia. It does mean the ban has given Chinese chipmakers a permanent customer base and a patriotic reason to keep spending.
Huawei's Ascend line has become the main domestic alternative for many Chinese firms, helped not only by US restrictions but also by Beijing's own pressure on local companies to buy Chinese hardware. The Guardian, citing the Financial Times, reported in January that Chinese customs officials blocked Nvidia H200 shipments even after the US had cleared them under a licensing regime tied to a 25% tariff. Chinese authorities also warned domestic technology companies against buying the chips unless they had to. Washington opened the door a crack. Beijing told its own firms to look elsewhere.
Nvidia's position is awkward because its software still travels where its sanctioned hardware cannot. CUDA remains the language many AI engineers already know, and Chinese companies running smuggled, rented or older Nvidia gear are still living inside that ecosystem. But software loyalty weakens when the hardware supply becomes expensive and politically unstable. Huang has said China is only "nanoseconds behind" in chipmaking. That may be rhetoric, but it captures the risk: every month of restricted access gives Huawei, Cambricon and other domestic suppliers another reason to close the gap.
For investors and operators, the black-market premium is the cleanest signal in the story. A DGX B300 selling for more than twice its US retail price tells you Chinese AI demand has not been crushed. It has been repriced. Some buyers will keep paying that premium because they need Nvidia performance now. Others will decide the better bet is to fund domestic stacks, accept lower efficiency and avoid the next customs order or indictment.
The export controls have done some real work. They made top-end compute harder to get, raised costs and forced companies to make uglier procurement decisions. But if the goal was to keep Nvidia-class capability out of China altogether, the latest FT pricing data says that goal has not been met. Nvidia lost legal revenue, brokers captured the spread, and Huawei got the kind of urgency that normal industrial policy struggles to create.
That is the uncomfortable arithmetic. The chips are still getting in, just at black-market prices, and the companies that can't get them are building around the shortage.
Also read: Zhipu AI bets on a Shanghai listing to lock in its 2,000% post-IPO surge • Superhuman bets on trust as it buys AI detection startup GPTZero for its 19 million users and $30M ARR • Cerebras Systems beat its first earnings bar and still watched its stock fall 10 percent
