From London: As Australians woke on Tuesday morning, Bloomberg was reporting yet another potential twist in the longest-running soap opera in global technology policy. The Trump administration is considering capping Nvidia's H200 artificial intelligence chip exports to China at 75,000 units per customer, according to people familiar with the matter, adding a fresh layer of complexity to a saga that has already whipsawed markets and strained allied confidence in Washington's strategic consistency.
The proposed ceiling would apply equally to AMD's competing MI325X accelerator, meaning Chinese technology giants would face a combined quota across both products. The 75,000-unit limit is less than half of what firms such as Alibaba and ByteDance privately told Nvidia they would like to purchase. Bloomberg also reported that while total shipments to China could theoretically reach as many as one million units under earlier caps set during the regulatory process, the lion's share of current applications comes from a small number of Chinese tech giants, which under per-customer caps could collectively receive hundreds of thousands at most.
The commercial stakes for Nvidia are enormous. The company reported $193.7 billion in annual data centre revenue in its most recent earnings call. Yet China, which once accounted for at least one-fifth of Nvidia's data centre revenue, has effectively been a zero-revenue market for H200 chips despite Washington's formal approvals. Nvidia's CFO Colette Kress told investors last week that while small amounts of H200 products for China-based customers had been approved by the US government, the company had "yet to generate any revenue," adding: "We do not know whether any imports will be allowed into China."
The backstory is worth tracing, because the policy reversals have come thick and fast. President Trump announced on 8 December 2025 that the United States would allow Nvidia to sell its powerful H200 chip to China. He noted that the chips would only be sold to approved customers and that 25 per cent of the revenue would go to the US government as a tax. The US Department of Commerce followed up in January with formal rules permitting case-by-case export licensing. Yet Beijing has shown little enthusiasm for rubber-stamping the imports, with Chinese customs authorities at one point telling agents the chips were not permitted to enter the country.
The H200 matters because of what it can do. The chip is not Nvidia's best, but it is more than six times more powerful than the best US AI chip previously available in China, and better than any chip that leading Chinese designer Huawei plans to produce for at least two years. These are the computational engines behind advanced AI systems that increasingly drive autonomous weapons, including drone navigation systems, automatic gun emplacements, and targeting algorithms in modern warfare. That dual-use reality sits at the heart of a fierce internal debate in Washington between those who see the deal as sound economics and those who regard it as a strategic gift to a rival military power.
Nvidia CEO Jensen Huang has been the most prominent voice for engagement. Nvidia's logic is that keeping Chinese AI companies reliant on American technology will prevent Huawei from building the revenue and developer ecosystem necessary to compete on the global stage. Commerce Secretary Howard Lutnick, recounting a conversation with Huang, described the arrangement as "positive for us overall," though he acknowledged that "there are plenty of people who disagree."
Those people have a case worth hearing. Analysts at the Center for a New American Security calculate that the volume of chips permitted under current rules is roughly twice what Chinese fabs are expected to produce domestically in 2026, twice the capacity of the world's largest data centre, and nearly equal to OpenAI's entire deployed compute worldwide at the end of 2025. The Council on Foreign Relations has warned that large-scale H200 exports could cause Chinese AI developers to close the gap with leading US labs and enable China to develop an "AI Belt and Road" initiative that competes with US cloud providers around the world.
For Canberra, the implications of this policy turbulence extend well beyond semiconductor markets. As a consequence of joining the AUKUS security partnership, Australia has restructured its export control regime to align with US priorities. That alignment was premised on Washington maintaining a consistent, predictable approach to technology controls. The reality has proven rather different. Export controls work when they are consistent, predictable, and clearly tied to national security. They fail when they become bargaining chips or revenue generators. Australian policymakers are now confronted with a partner whose decisions on sensitive technologies appear to shift with each news cycle, driven as much by Nvidia's lobbying and tariff revenue considerations as by any coherent strategic framework.
There are, to be fair, legitimate competing interests at play here. Nvidia's argument that US technology dependency is itself a form of strategic leverage over China is not without merit. A China that runs on Nvidia chips is, in theory, a China that Washington can observe, pressure, and potentially constrain. The counter-argument, that providing China with powerful AI hardware at scale simply accelerates capabilities it will eventually develop domestically anyway, is equally hard to dismiss. It remains unclear whether Beijing wants to ban the H200 outright so that domestic chip companies can flourish, or is still negotiating restrictions, or whether these measures could be used as a bargaining tactic in talks with Washington.
What is clear is that the current state of play serves almost nobody particularly well. Nvidia has licences it cannot use. Chinese firms have demand they cannot satisfy. US security hawks believe too much has been conceded. And allies like Australia are left watching a foundational pillar of shared technology policy being redesigned in real time, driven by quarterly earnings calls and presidential deal-making instincts as much as by strategic doctrine. The per-customer cap now under discussion is an attempt to split the difference, but whether it holds, and whether it achieves anything coherent on either the commercial or the security front, remains genuinely uncertain. Reasonable people in Washington, Beijing, and Canberra can draw very different conclusions from the same set of facts. That, perhaps, is the most telling sign yet that this particular chapter of the US-China technology contest is far from resolved.