• BIS Under Secretary Jeffrey Kessler told a U.S. House hearing on 14 July that ‘very few’ H200 chips had been shipped to China or Hong Kong. No exact unit count, customer delivery, installation or use was disclosed.
  • The evidence chain remains conditional: case-by-case U.S. review, customer licences, Chinese implementation, orders, production, U.S. inspection, shipment, import and revenue are separate milestones.

The newest evidence is limited shipment—not broad delivery

Reuters reported that Kessler told lawmakers only a very small number of H200 chips had so far been shipped to China or Hong Kong. The statement adds limited-shipment evidence to May's report of no deliveries, but it still does not establish final delivery; nor does it identify units, purchasers, destinations or completed Chinese import procedures.

‘Shipped’ is a logistics milestone. It does not by itself establish receipt by the final customer, installation in a data centre, operational availability, workload use, payment or revenue recognition. The absence of those figures prevents a claim that access has become broad or commercially material.

The U.S. rule is case-by-case and conditional

BIS changed its review policy on 13 January so applications for H200, AMD MI325X and similar chips could be considered case by case if security conditions were met. Applicants must show that exports will not reduce production available to U.S. customers, that Chinese purchasers have compliance and screening procedures, and that products undergo independent third-party testing in the United States.

A review policy is not a licence for every buyer or consignment. NVIDIA later told investors that February licences allowed small amounts of H200 products for specific China-based customers and required U.S. inspection before shipment. The company said that inspection route would subject the chips to a 25% tariff when imported into the United States; that disclosure should not be simplified into a universal sales fee or a delivered-China revenue share.

The implementation record changed over time

Reuters reported in March that NVIDIA chief executive Jensen Huang said the company had licences from both governments, was in the process of restarting production and had many customer orders. In May, however, Reuters said the United States had cleared around 10 Chinese firms—including Alibaba, Tencent, ByteDance and JD.com—while no H200 delivery had occurred. Lenovo and Foxconn had distributor approvals, a different role from final use.

On 8 July, Reuters relayed a report that China planned to permit limited purchases by leading AI firms, potentially fewer than 200,000 chips and less than half their earlier requests. That was an attributed plan, not an official final allocation. The 14 July testimony establishes that some shipments began; it does not resolve the total Chinese quota or every customer's status. Separate documents reported H200 purchase permission for ZTE Kangxun Telecom and Maginfra, while a Kingsoft subsidiary's approval concerned AMD chips.

Orders, production and shipment are not revenue

NVIDIA's Q1 Form 10-Q said no Data Center Hopper products were shipped to China during the quarter ended 26 April. It separately said that, as of the filing date, no revenue had been generated under the H200 licensing programme and the company did not yet know whether China would allow imports. Those are dated filing facts, not statements that no shipment could occur later.

NVIDIA's Q2 outlook assumed no Data Center compute revenue from China. That conservative planning assumption is not an actual Q2 result and does not tell readers whether the very few July shipments will be accepted, billed or recognized. A purchase order can trigger production planning without establishing shipment; shipment can occur before delivery; and delivery can precede installation, use and accounting recognition.

What the limited opening changes—and what it does not

The limited shipments show that the export channel is operational for at least some approved cases. They do not demonstrate a general reopening of China's advanced-AI-chip market, unrestricted supply or a measurable change in NVIDIA's segment revenue. Customer vetting, U.S. testing, Chinese import decisions and the economics of the inspection/tariff route remain control points.

For Chinese AI operators, H200 access can matter only if permission converts into usable systems at sufficient scale, with networking, power, software and data-centre capacity. For NVIDIA, the commercial test is recognized revenue and margin after compliance and logistics costs—not announced orders, production-restart claims or units placed in transit.

What to watch

  • Exact H200 unit counts shipped, imported, delivered and installed.
  • Named customer and distributor licences, end-use restrictions and expiration terms.
  • Official Chinese import permissions or quotas rather than attributed plans.
  • Completion of U.S. inspection and treatment of the 25% tariff route.
  • NVIDIA disclosure of H200-program revenue, margin or inventory commitments.
  • Evidence of customer utilization rather than orders or physical movement alone.

Sources