TLDR
Jeffrey Kessler told the House Foreign Affairs Committee on 14 July 2026 that H200 exports to China had been 'minimal' and 'very few', well below the volumes implied by licences granted. Washington's January 2026 rule allows case-by-case H200 approvals but caps China-bound shipments at 50% of US domestic sales and requires independent third-party performance testing. A loophole allowing newer Blackwell-series chips through Chinese-linked subsidiaries was closed by BIS guidance on 31 May 2026, while Beijing's own 30-working-day import licensing process adds a second bottleneck. The gap between approvals and actual deliveries shows that legal, regulatory and geopolitical friction now governs the pace of the US-China chip competition.
KEY TAKEAWAYS
Approvals on paper, trickle on the ground
Jeffrey Kessler, Under Secretary of Commerce for Industry and Security, gave the House Foreign Affairs Committee a blunt assessment on 14 July 2026. Kessler said that there have been minimal exports of any H200s to China so far, describing shipments as "very few."verifiedVerified Source: fidelity.com[2] That assessment cuts against any impression that Washington's decision to unlock case-by-case approvals had opened a meaningful pipeline of advanced AI chips to Beijing.
Kessler also told the committee that President Trump has turned export controls into a bargaining chip in broader negotiations with China, and weakened existing safeguards, including by approving licences for advanced AI chips destined for China.[2] The framing underlines the political volatility sitting behind every licence decision.
How the January 2026 BIS rule works
The Bureau of Industry and Security's final rule, which took effect on 15 January 2026, formally shifted the licence review policy for exports of the NVIDIA H200 and equivalent advanced computing chips to China and Macau from a presumption of denial to discretionary case-by-case review.verifiedVerified Source: govinfo.gov[1] The change was not a blanket permission; it came loaded with conditions that have proven difficult to satisfy quickly.
Exporters must certify that aggregate H200 shipments to China and Macau will not exceed 50% of the total product shipped to US end users, and each shipment must undergo independent third-party testing to verify performance.[1] BIS also requires rigorous Know Your Customer procedures and end-use controls before any licence clears. Those obligations stack into timelines that licence numbers alone do not capture.
The Blackwell loophole and how BIS shut it
Critics had flagged a specific vulnerability in the January 2026 framework: subsidiaries of Chinese firms operating outside mainland China might source NVIDIA's newer Blackwell-series chips without clear licence requirements, then route them onward. The concern was structural, not theoretical, given the global footprint of major Chinese technology conglomerates.
BIS closed the gap on 31 May 2026, issuing guidance that explicitly requires export licences for advanced AI chips, including the Blackwell series, sold to any foreign subsidiary whose ultimate parent company is headquartered in a Country Group D:5 nation, regardless of where the subsidiary itself is located.verifiedVerified Source: bis.gov[3] The guidance directly targeted structures designed to shift the effective point of sale away from the controlled jurisdiction.
The May guidance came roughly four and a half months after the January rule, suggesting BIS moved reactively once the routing risk became concrete. Whether any Blackwell chips reached Chinese-linked entities through subsidiary structures before 31 May remains unconfirmed in the public record.
Beijing's own bottleneck
Even where a US export licence exists, shipments face a second sovereign gate. Under China's Measures for the Administration of Prohibited and Restricted Import of Technologies, advanced semiconductor imports including the H200 require an import licence and a technical review by local commerce authorities within 30 working days of application.[4] That 30-day window is a minimum processing standard, not a guaranteed clearance date.
China's Ministry of Commerce administers the scheme under rules that date to December 2001 but apply fully to contemporary AI hardware.[4] An exporter who navigates BIS certification, third-party testing and the 50% cap still cannot book delivery without Beijing's separate sign-off, which helps explain the distance between approved licences and Kessler's "very few" on the ground.
What the stalemate means for Australia
Australia sits in a structurally awkward position in this standoff. The country's AI research base, cloud infrastructure sector and defence-adjacent technology programmes are all deeply dependent on high-end GPU supply, and constrained H200 availability globally compresses the pool accessible to allied nations competing for the same hardware.
Washington's export control architecture does not treat Australia as an adversary, but its knock-on effects on global allocation are real. Every chip diverted through approved China-bound licences, however few actually ship, tightens a supply chain that Australian cloud operators and research institutions draw from. Kessler's testimony on 14 July 2026 confirmed the gap between policy and physical reality has not closed.
SOURCES & CITATIONS
FREQUENTLY ASKED QUESTIONS
Why are H200 shipments to China so low despite export licences being approved?
What was the Blackwell loophole and is it still open?
When did the US shift its H200 export policy toward China?
Caleb Reed covers breaking news and sport for Bushletter. Fast and verb-led, he writes with a news-wire cadence and no patience for PR spin.
