The One Hard Measure Backfired. Everything Else Is Voluntary.

The Trump administration has taken exactly one binding action on AI: cutting off China's access to advanced American chips. Nvidia CEO Jensen Huang warned what would happen next.

China's AI moves on with or without U.S. chips. It has to compute to train and deploy advanced models. The question is not whether China will have AI, it already does. The question is whether one of the world's largest AI markets will run on American platforms. — Jensen Huang

It happened. Cut off from Nvidia, Chinese firms built their own. Huawei produced the Ascend 950, a domestic AI chip that now powers DeepSeek's models. DeepSeek permanently slashed V4-Pro prices by 75 percent, making its models 12 to 19 times cheaper than equivalent offerings from OpenAI and Anthropic ^[1]. The company is now developing its own inference chips and has raised $7 billion at a valuation north of $52 billion to do it ^[2]. The result: Chinese AI models now process 12.96 trillion tokens each week, more than four times the 3.03 trillion handled by American models, all six of the world's most-used models are Chinese, and they deliver comparable performance at roughly one-tenth the cost ^[3]. The export controls were meant to preserve American dominance. Instead they created a competitor that now dominates global AI usage on price, and the cheaper AI that competitor enabled is feeding the very data-center buildout whose consumer costs the administration's other policies are scrambling to contain. That is the pattern: every time the administration's security goal and its economic goal collide, the result is a voluntary pledge that neither side's hawks consider adequate. The pattern is most visible in a six-week whipsaw on frontier-model security reviews. On May 4, the administration signaled it would pursue mandatory federal reviews of advanced AI models before public release, an FDA-style gate ^[4]. On May 20, Trump cancelled the executive order hours before signing, after lobbying from Elon Musk, Mark Zuckerberg, and White House AI advisor David Sacks. Trump was explicit about why.

I really thought that could have been a blocker. — Donald Trump

On June 2, he signed a voluntary version instead, with the review window cut from 90 days to 30 and mandatory licensing explicitly prohibited. Sacks called it a "game changer" for not delaying releases ^[5]. On June 24, after Anthropic's Mythos model was found to have penetrated classified U.S. government systems "within hours," the reviews were reinstated. They remained voluntary ^[6]. In six weeks the policy lurched from mandatory to cancelled to voluntary to reinstated-voluntary. Each time security demanded control and economics demanded speed, the compromise landed on a pledge with no binding force. The Mythos incident is the moment the two goals stopped being separate tracks and became a single event. When Anthropic withheld Mythos from public release because of catastrophic cybersecurity risks, Treasury Secretary Scott Bessent and Fed Chair Jerome Powell convened emergency meetings with Wall Street bank CEOs about potential financial-system destabilization ^[7]. A security-driven model withholding immediately triggered financial-stability alarms. The administration could not pursue one goal without tripping the other. On the economic side, the same voluntary pattern holds. In February, Trump announced the Ratepayer Protection Pledge, asking AI and tech companies to fund their own power plants and grid upgrades so consumer electricity bills would not rise to subsidize the data-center boom. The pledge now covers 80 percent of U.S. electricity delivery with more than 220 signatories including Google, Microsoft, Meta, Amazon, OpenAI, Oracle, and xAI ^[8]. ICF has warned that AI-driven demand could increase monthly utility bills by 15 to 40 percent by 2030 ^[9]. Nine major trade groups told Bessent and Commerce Secretary Howard Lutnick that the AI data-center buildout is causing a critical memory-chip shortage that will produce "significant and sustained near-term price increases for American households" on consumer electronics, vehicles, and medical devices, with SK hynix estimating shortages could persist until 2030 ^[10]. The Federal Reserve identified AI infrastructure as a primary new inflation threat on July 19, with Governor Lisa Cook and New York Fed President John Williams citing more than $1.5 trillion in data-center plans driving up costs that have already forced Apple and Microsoft to raise laptop, iPad, and Xbox prices ^[11]. The Pledge is voluntary. Critics call it a "handshake agreement." Senators Richard Blumenthal and Josh Hawley introduced the GRID Act to legally mandate that data centers source power outside the grid, and Senator Bernie Sanders proposed a nationwide moratorium on new data-center construction. The House Energy and Commerce Committee approved a bipartisan bill to formalize protections. None have passed ^[8]. The hawks on both flanks agree on one thing: the voluntary compromise is insufficient. On security, Anthropic CEO Dario Amodei has proposed binding FAA-style regulations with government authority to block model releases, arguing that security and economic protection are inseparable ^[12]. Google DeepMind CEO Demis Hassabis proposed a FINRA-style federally overseen AI watchdog with mandatory safety reviews. White House AI advisor Sriram Krishnan rejected the idea outright.

there will not be an FDA for AI. — Sriram Krishnan

On economics, Hawley, Blumenthal, and Sanders have each pushed binding consumer protections that went nowhere ^[9]. The administration's answer to both camps is the same: a voluntary framework that asks for cooperation and cannot compel it. Energy Secretary Alfie Moon made the administration's position plain at the July expansion of the Ratepayer Pledge.

We’re going to win in AI and we’re going to stop the rise of electricity prices because of the seven leaders that are sitting around this table… The President recognized from the start [that] we have to lead in AI and we have to stop the electricity price rise, and those are not in contradiction. — Alfie Moon

The record says otherwise. The one binding measure produced a Chinese competitor that now leads the world on price and usage. The voluntary security framework was cancelled the moment it threatened economic growth, then reinstated only after a model breached classified systems, and stayed voluntary. The voluntary economic pledge asks companies to absorb costs that the same administration's export controls, by fueling cheaper AI abroad, are helping to inflate. The Treasury's own career analysts have drafted a report warning of a systemic AI bubble, comparing it to the dotcom era but noting that AI firms are more deeply entrenched across stocks, chips, utilities, and private credit ^[13]. The Fed warns that persistent AI-driven inflation may force rate hikes that would "negatively affect AI-driven market valuations" ^[11]. The two goals are not in contradiction, the administration insists. The policy they built to manage them is.