Trump’s China Deal Risks Turning Tariff Relief Into Strategic Drift

The easiest way to lose a negotiation is to pretend every concession is reversible.

That is what worries me about the Trump administration’s new China opening. On paper, the agenda has a tidy transactional logic: cut some tariffs, reopen Chinese demand for U.S. farm goods, approve selected semiconductor exports, and use a Trump-Xi summit to cool a trade war that has hurt farmers, importers, and households. The White House says the package includes Chinese purchases of at least $17 billion a year in U.S. agricultural products in 2026, 2027, and 2028, plus renewed access for more than 400 U.S. beef facilities, according to its May 17 fact sheet¹. That sounds like a bargain. I think it is closer to a warning.

A tariff is just a tax on imports, usually paid first by the importer and then absorbed through thinner margins, higher prices, or supply-chain reshuffling. Export controls are the opposite sort of tool: rules that restrict sensitive goods or technology from going to a foreign buyer. The problem with the emerging China package is that it treats these tools as if they were poker chips of equal weight. They are not. Tariffs on shoes, furniture, or low-risk industrial inputs can be costly and blunt. Export controls on advanced AI chips and chipmaking equipment are part of the security perimeter. Taiwan arms sales sit in an even more dangerous category because they shape Beijing’s calculation across the Taiwan Strait, the contested waterway between China and democratic Taiwan.

I am not sentimental about tariffs. The U.S. International Trade Commission found that Section 301 tariffs on Chinese goods raised prices paid by U.S. importers nearly one-for-one, meaning the burden landed heavily on American firms rather than magically on Beijing, according to the commission’s 2023 report summary². Yale’s Budget Lab estimated that a 10 percent China tariff would raise the overall price level by just over 0.1 percent, equal to an average disposable-income loss of $223 per household, in its analysis of the February 2025 tariff proposal³. If Washington cut consumer-facing tariffs and tariffs on non-sensitive inputs, Americans would get real relief. Importers would too. Some manufacturers that rely on Chinese components would also breathe easier.

That is the strongest case for a limited deal. Broad tariffs are a clumsy way to compete with China. Strategic decoupling, meaning the selective reduction of dependence on a rival, should be selective. It should focus on technologies, minerals, military supply chains, and chokepoints that matter in a crisis, not every kitchen appliance or apparel shipment that crosses the Pacific.

But the current package does not look selective enough. It looks like the old Phase One mistake with a semiconductor rider attached. Phase One was the 2020 U.S.-China trade agreement in which China promised huge additional purchases of U.S. goods and some market reforms in exchange for a truce in the trade war. It failed the core purchase test. Peterson Institute economist Chad Bown found that China bought only 58 percent of the U.S. exports it had committed to buy in 2020 and 2021, and none of the extra $200 billion in purchases promised under the deal, in his postmortem of Phase One⁴. Agriculture did better than other categories, but “better” is not the same as enforceable.

The new farm pledge has the same soft center. The White House says China agreed to large annual agricultural purchases, while Reuters reported that China’s Commerce Ministry confirmed agricultural tariff cuts but left implementation questions unanswered and did not specify the products, in a May 20 account⁵. That mismatch matters. A public U.S. headline is not the same thing as a Chinese customs schedule. Farmers need actual soybean, beef, pork, poultry, and grain flows, not a summit souvenir.

The semiconductor piece is more troubling. A semiconductor supply chain is not just “chips.” It includes design software, advanced processors, fabrication equipment, materials, packaging, cloud access, and end users. Since 2022, the Commerce Department’s Bureau of Industry and Security, or BIS, has used export controls to restrict China’s access to advanced computing chips, supercomputing capacity, and semiconductor manufacturing equipment, with updates in 2023 aimed at limiting China’s ability to buy and build advanced chips for military-relevant uses, according to BIS’s public information page⁶. That policy makes sense because AI compute is not a normal commodity when it can support cyber operations, weapons design, surveillance, and military modernization.

Yet BIS announced in January 2026 that license applications for Nvidia H200, AMD MI325X, and similar chips for China would be reviewed case by case if security requirements are met, according to its license-review policy revision⁷. Those are not toaster chips. They are advanced AI accelerators. I can imagine licensing legacy chips for civilian autos, appliances, medical devices, or industrial machinery. I cannot see why H200-class and MI325X-class approvals should be treated as part of a trade thaw unless Washington is prepared to say openly that it is relaxing the national-security line.

The usual answer is end-use control: approve sales only to trusted customers, monitor where the chips go, and punish diversion. That is better than nothing, but it is not a wall. In March, the Justice Department charged three people with allegedly conspiring to divert high-performance U.S.-assembled AI servers to China through false documents, staged dummy servers, and transshipment schemes, according to the DOJ’s indictment announcement⁸. The charges are allegations, not convictions. Still, they show the mechanism of risk: once sensitive hardware enters a commercial maze, paperwork can become a weapon.

Taiwan is the biggest red flag. The United States does not formally recognize Taiwan as a country, but under the Taiwan Relations Act it has long provided the island with defensive arms and maintained the capacity to resist coercion that would threaten Taiwan’s security. That posture is not charity. Taiwan is central to the global semiconductor economy, and a war or blockade in the Taiwan Strait would hit everything from AI servers to smartphones to cars. After the Trump-Xi summit, AP reported that Taiwan had not been notified of any pause in a planned $14 billion U.S. arms sale, even after Trump described arms sales to Taiwan as “a very good negotiating chip” in dealings with China, in a May 22 report⁹. Japan had already been watching nervously; The Japan Times reported before the summit that Japanese officials were concerned Trump could weaken U.S. engagement in East Asia while prioritizing a deal with China, in a May 5 report¹⁰.

That is where the bargain crosses from clever to careless. Tariff relief can be staged. Farm purchases can be measured. Semiconductor licenses can be counted. But deterrence is psychological as much as material. If Beijing hears that Taiwan arms sales are now part of the dealmaking menu, the damage is not limited to one delayed package. The signal itself changes.

Allies are reading the same signal. The European Commission has said China’s export controls on rare earths and other critical raw materials caused disruptions for EU and global supply chains, and it is pushing diversification through its RESourceEU agenda, according to a European Parliament answer from the Commission¹¹. India, another hoped-for partner in supply-chain diversification, has had its own tariff turbulence with Washington; Reuters reported in January that U.S.-India trade talks had stalled after a dispute over tariffs and process, in a report on the breakdown¹². If Washington tells allies to de-risk from China while Washington itself trades AI-chip access and Taiwan ambiguity for purchase pledges, the message is not strategy. It is mood.

The counterargument deserves respect. Keeping every China tariff in place because China is a strategic rival would be lazy policy. A disciplined agreement could cut low-value tariffs, require monthly customs-verified farm purchases, publish product-level commitments, use automatic snapbacks for missed benchmarks, preserve advanced-chip controls, and explicitly exclude Taiwan arms policy. I would support that kind of bargain. It would turn tariff leverage into consumer relief without confusing commercial easing with strategic surrender.

But that is not the bargain now visible. The visible bargain contains immediate U.S. concessions, ambiguous Chinese farm implementation, case-by-case review for advanced AI chips, and presidential language that makes Taiwan sound negotiable. My judgment is that Washington is at serious risk of losing both sides of the trade: it may give up tariff leverage before China delivers durable purchases, while also weakening the security clarity that export controls and Taiwan arms sales are supposed to provide.

The test is simple. Within the next six months, watch three indicators: whether the final text legally excludes H200-class and MI325X-class AI accelerators from the trade package, whether China publishes product-level monthly purchase schedules that can be checked against customs data, and whether the $14 billion Taiwan arms sale moves forward without linkage to Beijing. If any two of those fail, this will not be a shrewd reset. It will be Phase One with higher stakes.