Join War on the Rocks and gain access to content trusted by policymakers, military leaders, and strategic thinkers worldwide.
The White House is reportedly lobbying Congress to kill the GAIN AI Act. If accurate, this represents one of the more baffling policy reversals of the Trump administration’s second term — an administration that has otherwise demonstrated remarkable clarity about using American technological advantages as instruments of statecraft.
The confusion stems from widespread misunderstanding about what the GAIN AI Act actually does. Critics characterize it as protectionist overreach that would starve allies of advanced chips. The reality is far different — and far more aligned with the administration’s stated objectives than the White House appears to realize.
What GAIN AI Actually Does
As written, the GAIN AI Act is designed to deny U.S. chip sales to China — plus 24 other countries designated as state sponsors of terrorism or against whom the United States enforces an arms embargo. It is not designed to hoard AI chips in the United States.
The act achieves this by specifying that any U.S. company seeking a license to export “advanced integrated circuits” to an entity in a “country of concern” must first ensure that a “U.S. Person” did not want to buy it first. The logic is straightforward: If advanced AI accelerators end up being a scarce commodity for the foreseeable future — a good bet, given unprecedented global demand — then Chinese buyers seeking to displace American AI leadership should be last in line to buy them.
Denying chips to China makes strategic sense — which is why it has been the bedrock of American export control policy for nearly a decade, beginning under President Donald Trump’s first term. First, many Chinese customers of U.S.-designed AI chips are partnering with the Chinese military to build systems that directly threaten the United States and its allies. Second, computational power is the one bottleneck Chinese AI companies complain is slowing their ability to compete with U.S. companies in the global AI services market. Third, every chip American companies sell to customers in China today frees up inventory for Huawei to sell abroad — accelerating the rise of a “Digital Silk Road” that contests U.S. technological leadership.
As foolish as it may be to sell China the very tools it needs to displace American power, the GAIN AI Act does not even stop this from happening. It simply ensures that American companies have their computational needs met before selling chips to the last people on earth who should have access to them.
But the bill’s most significant provision — widely overlooked — is what it does for allied AI exports, the cornerstone of the Trump administration’s foreign policy. Section C creates a new category of “trusted United States persons” who would be entirely exempt from existing Bureau of Industry and Security license requirements for regional stability (EAR 742.6) and advanced computing end-use (744.23). This exemption would dramatically streamline AI chip exports to U.S.-owned and operated data centers worldwide.
To qualify as “trusted,” the bill requires companies to meet certain security standards designed to prevent rampant chip smuggling. Essentially, the bill promises to grant American data center providers a “get-out-of-jail-free” card from the Commerce Department in exchange for doing four things they are mostly doing already: implement robust physical and cybersecurity controls in data centers they operate; forbid entities in “countries of concern” to hold more than 10 percent ownership of their companies; keep the majority of their total compute in the United States; and submit to annual audits.
These are not onerous requirements. For U.S. cloud providers building data centers in Abu Dhabi, Riyadh, Tokyo, or Singapore, GAIN AI would eliminate much of the licensing friction that currently slows American AI infrastructure exports. Microsoft and Amazon did not support this bill by accident. They recognized it for what it is: A pathway to faster deployment of American technologies in strategic markets. But American industry is not unified, as chip design firms face different incentives than cloud providers. NVIDIA — which has opposed any measure to restrict its sales abroad, and denies that the Chinese military uses its chips — benefits substantially from selling AI accelerators to Chinese customers willing to pay premium prices. The bill’s “right of first refusal” for domestic purchasers would redirect some of these high-margin sales away from Chinese buyers — a commercially rational concern that nonetheless conflicts with the president’s broader strategic objective of denying crucial resources to the United States’ chief competitor.
The Strategic Logic
The administration is right to prioritize exports to allies with energy grids capable of immediately ingesting large numbers of U.S. AI accelerators. Allies like the United Arab Emirates, Saudi Arabia, Japan, and Singapore have powered infrastructure ready for trusted U.S. persons to install chips today. Facilitating their ability to build data centers would help establish American infrastructure dominance in those markets before Huawei’s alternatives become competitive enough for export — which could happen within a few years if China plays its cards right.
GAIN AI follows exactly this logic. The bill’s “trusted United States person” exemption eliminates licensing friction for U.S. companies deploying chips to their own data centers abroad, while maintaining hard barriers against Chinese access. It recognizes that “America First” in AI infrastructure does not necessarily mean that every chip stays on U.S. soil — but that American technology should establish irreversible footholds in strategic markets.
Why Is the White House Opposed?
In July 2025, Trump signed an executive order establishing the American AI Exports Program, designed to “support the global deployment of United States-origin AI technologies” and “decrease international dependence on AI technologies developed by our adversaries.” His most trusted technology advisors correctly understand the importance of exporting the American AI stack.
GAIN AI advances these objectives. It denies advanced chips to China while streamlining exports to aligned partners, and creates statutory backing for exactly the kind of strategic technology deployment the President envisions. The bill even includes dynamic updates to technical thresholds — long requested by industry — by empowering the Commerce Department to adjust which chips should be considered “advanced” as hardware evolves.
If the administration’s priority is maximizing short-term revenue for U.S. semiconductor design companies — including through sales to Chinese customers who often pay premium prices and negotiate less aggressively than U.S. hyperscalers — then opposing GAIN AI makes sense. NVIDIA’s record revenues could see near-term pressure if high-margin Chinese sales are displaced by domestic customers with a “right of first refusal.”
But this would mark a major departure from the administration’s broader China policy. Every other technology control regime Trump has maintained or expanded — from Entity List designations to ICTS authorities — has demonstrated a willingness to accept economic costs in service of strategic national security objectives. It does not follow that exports of semiconductors, the world’s most strategic technology, should be exempt from this logic.
The Alternative Is Worse
The alternative to GAIN AI is not the status quo. Congressional appetite for China chip restrictions remains intense and bipartisan. If GAIN AI fails, the likely replacement is the SAFE Act — a draft bill reportedly circulating among lawmakers, which has not yet been formally introduced — which would require the Department of Commerce to deny all license applications for any AI chips more powerful than the current H20 threshold for at least several months, with no trusted person exemptions and no flexibility for allied deployment.
The SAFE Act is precisely the blunt instrument that GAIN AI’s opponents claim to fear: a statutory mandate that removes executive branch flexibility and treats all non-U.S. markets identically. It would accomplish Chinese denial while simultaneously creating new obstacles for the strategic allied deployments championed by both industry and the administration’s own AI Exports Program.
Fumbling on the One-Yard Line
What makes this situation particularly frustrating is that GAIN AI seemed to have already navigated the hardest political terrain. The bill emerged from the House Select Committee on the Chinese Communist Party with bipartisan and bicameral sponsorship, earned support from major U.S. cloud providers, and survived months of negotiation between China hawks and commercial interest groups. The final bill is a rare policy compromise that advances U.S. strategic and national security objectives while addressing industry’s legitimate concerns about allied market access.
Watching it stall at the one-yard line — reportedly at the White House’s behest — suggests either profound confusion about what the bill does, or an active decision to prioritize the profitability of chip design firms over the strategic technology competition the administration claims to take seriously.
Neither explanation is reassuring. If it’s confusion about the legislation’s intent and operating mechanisms, then administration officials should spend more time examining updated provisions and recognize how they squarely advance the president’s objectives. But if opposition to the bill is motivated by corporate capture, then the administration should own the fact that it will permit the sale of advanced chips to a declared adversary because they pay premium prices.
GAIN AI is not perfect. But it is better legislation than any alternative likely to emerge from a frustrated Congress, and dramatically better than the incoherent middle ground of wanting to promote American AI exports while simultaneously enabling their Chinese competitors.
Ryan Fedasiuk is a fellow for China and Technology at the American Enterprise Institute and an adjunct professor at Georgetown University’s Security Studies Program. He previously served as an advisor for U.S.-China Bilateral Affairs at the U.S. Department of State.
Image: Official White House Photo by Molly Riley via Wikimedia Commons.