Nvidia's H20 China sales resume after Jensen meets Trump — Zak Kukoff breaks down the security trade-offs

Summary

Nvidia scored a major policy reversal on July 14th when the Trump administration greenlit exports of the H20 chip to China, ending a ban imposed in April that had forced Nvidia to absorb an $8 billion revenue impairment. The H20 was originally engineered specifically to comply with Biden-era export restrictions, making the April ban a particularly costly regulatory whipsaw for the company.

Jensen Huang's direct lobbying campaign was central to the outcome. Nvidia's registered lobbying spend has scaled from roughly $500,000 in 2023 to $640,000 in 2024 and nearly $1 million in Q1 2025 alone, but the more significant lever was Huang's personal relationship with the White House. A bipartisan letter criticizing his Beijing trip attracted only two signatories — Senator Banks on the right and Senator Warren on the left — indicating how little appetite exists in Congress to publicly oppose Nvidia.

The Chip Security Act, introduced in May and shaped with Nvidia's input, is moving through the House with strong momentum. Its Senate trajectory remains uncertain. The Act centers on chip diversion tracking and a still-unresolved question of whether a remote kill switch would be embedded in exported hardware.

The national security case against the export is straightforward and largely conceded even by proponents. H20s diverted to PLA or military use is treated as a near-certainty rather than a hypothetical. Zak Kukoff argues the more immediate risk is not kinetic — it is cyber warfare, where LLMs are increasingly being deployed to target adversaries' water, electrical, and financial infrastructure. He also flags the soft-power vector: AI-enhanced content influence operations, an evolution of what TikTok already executes without requiring large transformer models.

China's strategic posture toward the H20 deal is transactional rather than dependent. The H20 serves as a bridge architecture to Huawei's Ascend platform — Beijing wants the chips to accelerate its own domestic AI stack, not to entrench reliance on Nvidia. The Apple-in-China parallel is instructive: technology transfer tends to build indigenous capability that ultimately displaces the supplier. Nvidia compounding that dynamic by opening an R&D center in Shanghai amplifies the technology diffusion risk beyond chip hardware alone.

The broader trade context matters. The H20 export approval is partly a function of improving US-China trade negotiations, with China described as a more cooperative counterpart than Japan at this stage. Rare earths and tariffs on American goods remain live issues in the broader bilateral framework, and the chip concession likely reflects movement across that wider negotiating table.

On the military AI question, autonomous decision-making in live combat theaters is already operational — drone targeting and related systems are deployed in Ukraine and the Middle East today. The trajectory Kukoff outlines is removal of the human from the decision loop entirely, not just from the cockpit. A PLA equipped with H20-powered supercompute clusters is meaningfully more capable in a Taiwan scenario than one still fighting conventional warfare, a point that cuts directly against the optimistic framing that commercial chip dependency creates geopolitical stability.

On TikTok, Polymarket currently prices a sale announcement in 2025 at 35%. There is no active lobbying constituency pushing the ban to enforcement. Zuckerberg's practical priorities in Washington are data center permitting reform, accelerated nuclear energy approvals, and federal preemption of the current patchwork of state and local regulations — not a TikTok ban that would draw antitrust scrutiny toward Meta. The watch item to track over the next several weeks is whether H20 exports actually ship, what licensing requirements attach, and whether the Chip Security Act clears the Senate.