Export Controls: National Security Tool or Industrial Policy Lever?

In December 2025, the Trump administration relaxed regulatory controls on the export of Nvidia H200 logic semiconductors to China.  The H200 devices are utilized in artificial intelligence (AI) applications. Following that action, the Bureau of Industry and Security (BIS) of the U.S. Department of Commerce is approving export licenses for China-bound shipments on a case-by-case basis for Nvidia’s H200 and AMD’s M1235X AI logic devices.

This action has raised concerns among some analysts that new production of China-bound H200 chips would divert Nvidia manufacturing capacity from production of comparable or superior AI devices from U.S. customers, resulting in a net loss of more computing power for the United States as China gains through its acquisition of H200 devices.  The implication is that the U.S. government should act to modify Nvidia’s manufacturing arrangements and sales patterns either by further restricting exports of H200s to China or by directing the company to modify its production plans to redirect resources toward chips for domestic users.

The larger policy issue is whether the executive branch should use semiconductor export controls solely as a narrowly tailored national-security tool—or as a broader instrument of industrial and economic policy that effectively directs the commercial and operational decisions of private companies.  In this instance, additional expansion should be rejected. H200 exports do not pose a meaningfully trade off against newer U.S.-focused chips like Nvidia’s Vera Rubin due to distinct supply chains. 

Semiconductor export controls were originally instituted by the Biden administration in October 2022 to address certain narrow national security concerns arising out of China’s advances in AI. The controls have been progressively modified and expanded, not always for national security purposes (e.g. mechanisms for raising revenue in the case of the tariff on H200 exports to China, achieving industrial policy objectives, and use as bargaining chips in trade negotiations) and they have induced China to accelerate its unprecedented efforts to achieve self-sufficiency in chips. A further expansion of government intervention should be rejected. In this context, it is important to note that: 

• Nvidia has received licenses to ship limited quantities of H200 devices to China but reportedly does not expect any meaningful shipments in the near term, reflecting restrictions by both governments.
• In January 2026, Chinese customs officials instructed agents that H200 chips were “not permitted” to enter China despite issuance of export licenses by BIS.
• The Chinese government told domestic tech firms to halt orders for Nvidia H200 devices.
• Although the government subsequently allowed a handful of domestic firms to import some H200 chips, it is using informal guidance to limit imports, with an eye toward encouraging users to turn toward domestic alternatives. China does not view acquisition of H200 chips as a national security advantage and sees emphasizing domestic alternatives as the best long-run guarantee of its own national and economic security.
• Proposed U.S. policies would place quantitative caps on the number of H200s that can be shipped to designated Chinese customers.
• Evolving U.S. and Chinese restrictions have placed Nvidia’s manufacturing and shipping operations for the H200 on a stop-and-start basis because of abiding uncertainty about how many shipments will be allowed, and to which customers.
• Reflecting the difficulties associated with the H200, Nvidia is prioritizing capacity for newer, more advanced architectures for U.S. clients (e.g., the Vera Rubin) so that limited production of the somewhat older H200 is not a direct trade-off for current U.S. customer needs.
• The current administration is imposing a 25% tariff on H200 chips, which are shipped from Taiwan to the U.S. for collection of the tariff, then sent back to China, a cost which creates an incentive for Nvidia to prioritize domestic U.S. sales.
• The H200 and the Vera Rubin differ substantially and engage separate and distinct supply chains, so there is no practical way to shift production from one device type to the other.
• The Vera Rubin will be manufactured at the 3nm process node, while the H200 is being made at TSMC 5-4nm node.
• The H200 uses TSMC’s CoWoS packaging, while the Vera Rubin will use C0WoS-L packaging, which increases the package area and CoWoS load per unit, but which will require a restructuring and expansion of substrate and advanced packaging capacity at TSMC.
• The Vera Rubin will be based on Samsung/Hynix  HBM4 architecture while the H200 uses TSMC HBM3.
• Consequently, there is no tradeoff between the H200 and the Vera Rubin in manufacturing terms – Nvidia can produce both, in response to demand signals. Currently, the ongoing export controls in the U.S. and regulatory guardrails in both the U.S. and China appear to be stalling any significant commercial momentum of H200 chips.

More fundamentally, as a general proposition, the U.S. government should not become involved in directing the commercial and operational decisions of private companies – in this case a world-leader driving U.S. competitiveness in semiconductor design and in AI. The U.S.’s own experience with this is illustrative. In the mid-twentieth century, U.S. government agencies regulated the operational and pricing decisions of private firms in industries like transportation and telecommunications. That practice was abandoned beginning in the 1970s, based on widespread recognition that reliance on market mechanisms – tempered only by the antitrust and consumer protection laws -- yielded better outcomes than the decisions of government regulators. There is no reason to turn back the clock now.

In short, the premise that H200 exports necessarily come at the expense of U.S. access to frontier AI compute is misguided, given the distinct manufacturing and packaging constraints that separate H200 production from newer, U.S.-oriented architectures. Policymakers should therefore resist calls to further expand export controls or to direct Nvidia’s production choices and instead rely on market signals—backstopped by existing security safeguards—to sustain U.S. technological leadership.