Nvidia’s H200 Return to China Was Real—but Far Smaller Than Its License Allowed

Nvidia did re-enter China with its H200 artificial-intelligence processor—but only in a limited, tightly controlled way. During the quarter ended July 26, 2026, the company sold a small number of H200 chips to Chinese customers, yet those sales represented less than 1% of its data-center revenue. 217

The episode showed why regulatory permission did not amount to a normal reopening of the Chinese market. U.S. authorities had licensed some exports, but Chinese restrictions and case-by-case approvals delayed shipments and kept actual volumes well below the permitted ceiling. 25

Which Chinese companies could buy Nvidia’s H200?

Reports identified ByteDance and Tencent as the clearest early recipients. Each was reported to have received roughly 10,000 H200 processors in recent weeks. 3340

Other reports described a broader licensing framework in which approved Chinese companies could purchase as many as 75,000 H200s each. One account listed companies including Alibaba, ByteDance, Tencent and JD.com among roughly 10 firms in an earlier licensing round. 36

Those names and limits should be treated as reported figures rather than a fully confirmed public authorization list. The higher-authority earnings materials confirm that Nvidia made only small shipments and did not sell the full amount allowed under its U.S. license, but they do not provide a complete customer-by-customer breakdown. 217

Why shipments fell short of the permitted volume

The central constraint was that U.S. and Chinese approval processes operated separately. U.S. licensing allowed Nvidia to ship H200 processors to selected customers, but Beijing continued to impose restrictions on purchases and how the chips could be used. Reports also described requirements for individual regulatory approval and limits on where some processors could operate. 239

That political and regulatory friction delayed deliveries. It also reduced the commercial value of the U.S. license: a product can be exportable in principle while remaining difficult for customers to order, receive or deploy in practice.

The result was a narrow foothold rather than a normalized China business. Nvidia said H200 sales to Chinese buyers contributed less than 1% of fiscal-Q2 data-center revenue. 2

Why Nvidia recorded a $400 million H200 charge

Nvidia recorded a charge of about $400 million during the first half of fiscal 2027 for excess H200 inventory and purchase obligations. Reporting attributed the diminished demand for that inventory to restrictions on U.S.-licensed H200 shipments into China. 5

The charge underlined the operational cost of planning around a market that could change faster than the supply chain. Nvidia had inventory intended for a China opportunity, but regulatory uncertainty meant that the company could not reliably convert the U.S. authorization into sales at the expected pace.

This was not the same issue as Nvidia’s earlier H20-related charge: the SEC material supplied here concerns a separate $4.5 billion fiscal-2026 H20 charge. 1 Keeping the two figures separate is important when assessing Nvidia’s China exposure.

Nvidia’s fiscal-Q3 outlook assumed zero China compute revenue

Despite the initial H200 shipments, Nvidia’s forward outlook did not assume a recovery in Chinese data-center-compute revenue. Chief Financial Officer Colette Kress said the company’s forecast reflected “ongoing geopolitical uncertainty” and included no China data-center-compute revenue. 644

Nvidia nevertheless guided fiscal-Q3 revenue to $108 billion, above the roughly $92.2 billion analysts had expected for fiscal Q2 and well ahead of the company’s earlier $91 billion fiscal-Q2 guidance midpoint. 7834

The message for investors was clear: China remained a potential upside opportunity, not a dependable part of the near-term model.

The China story was small beside Nvidia’s earnings results

Nvidia reported fiscal-Q2 revenue of $96.2 billion, up 106% from a year earlier. Data-center revenue reached $89.0 billion, up 117% year over year. 1737

The company’s revenue exceeded the approximately $92.2 billion analyst forecast, and its outlook for the following quarter also topped expectations. 34 Nvidia shares initially fell after the release before later rising by roughly 4% in after-hours trading, as investors focused on the strength of the broader AI demand outlook. 3234

That context explains why the limited China shipments did not materially change the quarter. Even a politically important return to China was financially minor compared with Nvidia’s much larger data-center business elsewhere.

Rubin made the supply decision more consequential

Nvidia was also managing a transition to its next-generation Vera Rubin platform. Reports said the company stopped production of China-bound H200 processors and redirected TSMC manufacturing capacity toward Rubin as China sales stalled. 410

That shift sharpened the trade-off. Nvidia could preserve inventory for a restricted, uncertain H200 market, or use scarce manufacturing capacity for a newer platform aimed at stronger demand outside China. The reported decision to prioritize Rubin suggested that the near-term opportunity cost of serving China had become less attractive.

For Nvidia, the bigger strategic test was therefore not whether a few H200 shipments could enter China. It was whether the company could keep scaling its newest systems while navigating export controls, Chinese approvals and the possibility that major customers will seek alternative or internally developed AI hardware.

What the H200 episode means

Nvidia’s fiscal-Q2 China return was best understood as a limited reopening, not a market reset:

• ByteDance and Tencent reportedly received about 10,000 H200s each, while broader per-company limits of up to 75,000 were also reported. 3336
• Deliveries remained far below the permitted amount because Chinese restrictions and approval requirements constrained demand and timing. 239
• China H200 sales contributed less than 1% of Nvidia’s fiscal-Q2 data-center revenue. 2
• The company recorded a roughly $400 million H200 inventory and purchase-obligation charge during the first half of fiscal 2027. 5
• Nvidia’s fiscal-Q3 outlook assumed no China data-center-compute revenue. 644
• The company’s $96.2 billion quarter and $108 billion fiscal-Q3 forecast showed that its global data-center business remained the dominant earnings driver. 1734

The practical conclusion is that China could still matter strategically, but it was not yet a reliable source of near-term revenue. Nvidia’s stronger commercial priorities were global AI infrastructure demand and the successful production ramp of Vera Rubin.