NVIDIA once again finds itself in the center of the game. According to Reuters, the company is considering increasing production of its H200 chip after orders from China have exceeded what its current capacity can cover. But this time the result will not be decided in Washington, but in Beijing, where the government must authorize the entry of the hardware. The Chinese response will determine whether the window opened by the United States translates into real sales or remains a gesture caught between opposing interests.
What has changed in Washington. The shift began in Washington on December 8, when Donald Trump announced that the United States would allow H200 exports to commercial customers approved and validated by the Department of Commerce, with a 25% tax on each sale. The measure marked a turning point with respect to previous restrictions and introduced a more flexible control model: the US Government will supervise shipments from Taiwan, subject processors to a security review before authorizing their departure to China and apply the corresponding surcharge. NVIDIA celebrated the announcement as a balance that, according to its own statement, seeks to make national security compatible with commercial activity, while in the markets its shares rose around 2% in subsequent operations.
Avalanche of orders. The signal that has led NVIDIA to consider increasing production is clear. According to the aforementioned agency, H200 orders from China already exceed the current manufacturing capacity of the chip. And, as we noted last week, technology groups such as Alibaba and ByteDance have contacted the company to explore volume purchases, aware that availability is very limited. NVIDIA has informed these clients that it is considering adding capacity, although without commitments or figures, in a context marked by scarcity and the priority that other, more advanced generations have today.
The interest in the H200 is also explained by its place in the NVIDIA catalog. It is the most powerful chip of the Hopper generation and a clearly superior alternative to the trimmed models designed for China, although it falls behind Blackwell, a generation with which, Trump explained, NVIDIA’s American customers are already moving forward. That position makes it an awkward balance: it’s not state-of-the-art, but it’s advanced enough to make a difference in training large-scale models.
What China decides. Beijing is not limited to giving a yes or no. According to sources cited by Reuters, the internal debate revolves around how to allow access to H200 without weakening the momentum of its domestic semiconductor industry. The authorities are studying imposing specific conditions on each order and reviewing the final destination of the chips, in a context in which manufacturers such as Huawei or Cambricon continue to be a priority for the country’s industrial policy.
NVIDIA H200
Capacity and bottlenecks. Increasing H200 production is not an immediate or easy decision. The chip is manufactured at TSMC using its 4nm process, an advanced capability that is hotly contested today. NVIDIA is prioritizing Blackwell production and preparing the transition to Rubin, while competing with other large customers, such as Google, for space in the Taiwanese manufacturer’s most advanced lines. That context explains why the company has warned its customers of tight supply even if it ultimately decides to add capacity.
National security and industrial pressure. The H200 debate goes beyond NVIDIA. In Washington, fear persists that the sale of advanced chips will contribute to strengthening China in sensitive areas, while the Administration itself has defended that completely cutting off access to American chips could reinforce the efforts of local manufacturers. The solution adopted by the Trump Administration seeks that balance, but keeps alive a controversy that conditions both exports and the real possibility of expanding production.
With demand pressing and supply at a minimum, the outcome is now being played out in the offices of the Chinese regulator. If Beijing authorizes the purchases, NVIDIA will have to decide to what extent it can reallocate capacity without compromising its industrial priorities. If it doesn’t, the H200 will join the list of advanced chips caught between politics and strategy. In both scenarios, the episode confirms that access to hardware has become as determining a variable as the chip design itself.
Images | NVIDIA + Photoshop
In WorldOfSoftware | Microsoft has reduced its ambition with AI. It has been realized that almost no one uses Copilot, they say in The Information
