China's Ministry of Industry and Information Technology held a press conference this week and laid out what the next five years look like. Nearly 200 national AI standards already written. A plan to build 500 zero-carbon factories. Accelerated work on 6G. Intelligent computing capacity of 2,185 EFLOPS as of the end of June, organized around more than 70 computing corridors. Core AI industries valued above 1.2 trillion yuan, roughly 176 billion dollars, with more than 6,200 AI companies operating nationwide.
Take those numbers with the appropriate salt. They come from a state press conference reported by a state news agency, and nobody outside Beijing audits them. But the numbers are not the point. The structure is.
One ministry stood up, described a five-year plan, and named the industries it intends to build: integrated circuits, aerospace, biomedicine, low-altitude economy, energy storage, robotics. Then it named the ones after that: quantum, biomanufacturing, hydrogen and fusion, brain-computer interfaces, embodied intelligence, 6G. Whether every target gets hit matters less than the fact that a single answer exists to the question "what is the plan."
Ask that question in the United States and you get 53 answers.
I track every AI bill in the country. As of this morning: 2,431 bills across 53 jurisdictions. Congress accounts for 574 of them. New York has filed 185, California 137, New Jersey 113, Illinois 108, Hawaii 97. Twelve jurisdictions have introduced fifty or more. Eleven have introduced ten or fewer. Puerto Rico, Montana, Nevada, and Arkansas have introduced none at all.
That is not a policy. That is 2,431 opinions.
And the bills that pass do not settle anything either. Colorado wrote a comprehensive AI act, then repealed it and replaced it with a different framework that businesses now have to comply with starting in January 2027. California's transparency requirements for frontier models became operative this month. Texas has its own statute in force. New York finalized the RAISE Act. Every one of these was written by serious people trying to solve a real problem, and every one of them is a different problem definition.
Washington's response has been to try to switch the states off rather than to lead them. A provision in the 2025 budget bill would have barred state AI enforcement for ten years; the Senate stripped it 99 to 1. A version resurfaced for the defense bill and was left out of the final text. In December, an executive order stood up a Justice Department litigation task force to challenge state AI laws. Thirty-six state attorneys general, from both parties, publicly opposed broad preemption in March. As of today no federal statute and no court has preempted a single state AI law. The fight is live and unresolved, which means a company building an AI product has to comply with everything while the government argues about who is in charge.
Here is the part worth being precise about, because it is where most commentary goes wrong.
The United States is not behind on artificial intelligence. American labs still ship the frontier models. American companies still raise the capital, and the capital expenditure figures I pull straight from SEC filings each quarter are not close: Microsoft, Amazon, Alphabet, Meta, and NVIDIA are spending on infrastructure at a scale no other country's private sector approaches. On the technology itself, we are fine.
What we do not have is a national focus. We have fifty state focuses, a Congress that has introduced 574 bills and passed no comprehensive framework, an executive branch trying to clear the field by litigation, and an industry that funds all sides and waits. China does not have better engineers. China has a decision-making structure that produces one plan instead of 53, and that is starting to show up in the places where coordination is the whole game: standards bodies, industrial deployment, grid and compute buildout, and the timeline from a research result to a factory floor.
Standards are the quiet one. Nearly 200 national AI standards is not a headline anyone will read twice, and it is the most consequential item on that list. Standards decide whose equipment interoperates, whose certification a buyer in Jakarta or Lagos or São Paulo accepts, and whose assumptions get baked into the next decade of infrastructure. That contest is not won by having the best model. It is won by showing up to the committee, every time, with one position.
We show up with 53.
The fix is not a five-year plan; that is not how this country works, and pretending otherwise wastes everyone's time. The fix is narrower and duller. Pick the handful of things that genuinely require one national answer, which is roughly: a single federal disclosure standard so companies stop building fifty compliance regimes, procurement rules that let the government buy AI at the speed it is produced, permitting and grid capacity for compute, and an actual presence in international standards bodies. Leave everything else to the states, where experimentation is a feature.
That list would fit on one page. We have 2,431 bills and no page.
The uncomfortable question is not whether we can out-build China. It is whether a system this fragmented can decide anything at all before the decisions get made for it, somewhere else, by people who did their committee work.