The media says it’s an emergency, and we must sacrifice so much to “beat China” — but it’s the oldest trick in the book. Image from Gemini
There has never been a ballot question on artificial intelligence. No presidential debate has asked whether the country should build it, and no referendum has asked how fast. Still, Americans are being told every week that the matter is urgent, that the future depends on it, and that anyone who hesitates is either naive or working for the other side. The two sides disagree on almost everything else, but both describe AI as an emergency. That shared framing deserves as much scrutiny as the technology.
The oldest move in the playbook
The United States has a long history of treating technological competition as a crisis, and of building permanent institutions on the strength of that crisis.
The clearest early example is the missile gap. After Sputnik, fear that the Soviet Union was pulling ahead in nuclear missiles became a central issue in the 1960 presidential race. Eisenhower eventually learned through his military intelligence officials that the gap did not exist, and that if there was any imbalance, it favored the United States. A National Intelligence Estimate briefing in September 1961 found the Soviet Union had only 10 to 25 launchers, far fewer than the more than 100 American land and sea based missiles deployed forward. Historian Christopher Preble has documented Kennedy’s awkward admission once in office that the gap was a myth, an admission that did not stop the military buildup he had used the gap to justify. The emergency disappeared, but the spending stayed.
Four years later, a second crisis produced a war. An internal National Security Agency history declassified in 2005 concluded that the destroyer Maddox did engage North Vietnamese forces on August 2, 1964, but that the reported incident of August 4 rested on bad naval intelligence and misreadings of North Vietnamese communications. Congress passed the Gulf of Tonkin Resolution within days, and the authority it granted outlasted the facts behind it.
The pattern repeated after September 11, 2001. The Patriot Act, the Department of Homeland Security, and Immigration and Customs Enforcement were all born in the emergency period that followed, and the Iraq War was sold on weapons of mass destruction that inspectors never found. The emergency passed, but the agencies, budgets, and powers became permanent. The lesson from these episodes is not that every threat is invented. The Soviet Union was real, and so was al Qaeda. The lesson is that when a threat is framed as a race, deliberation starts to look like a luxury the country can’t afford. Programs and powers get built quickly, and they rarely get unbuilt.
The race as organizing story
The current federal approach to AI is built on the language of the race. The White House’s 28-page AI Action Plan, released on July 23, 2025, was explicitly framed as an effort to win the AI race against global competitors. Its companion documents lean even harder on history. The executive order launching the Genesis Mission says America is in a race for global technology dominance, and calls for a national effort comparable in urgency and ambition to the Manhattan Project. Law firms advising clients on that order read the language the same way; one described it as placing AI where nuclear technology sat during the Cold War, at the center of long-term strategic competition.
Emergency framing tends to come with a request for centralized authority, and that followed. In December 2025, President Trump signed an executive order setting a “minimally burdensome” national AI framework, creating a Justice Department AI Litigation Task Force, and tying some broadband funding to whether states pause enforcement of AI laws that conflict with the order. The administration’s reasoning is that state-level safety, bias, and transparency rules could impose ideological constraints and slow the innovation needed to win the AI race. Supporters see a sensible national standard that spares companies fifty sets of rules. Critics see a race narrative being used to override local control. Congress had already weighed in: the Senate voted 99 to 1 against an earlier attempt to block state AI laws, and state lawmakers and attorneys general from both parties opposed it, saying it would weaken their ability to protect residents from fraud, deepfakes, and child abuse imagery.
The race framing is also flexible when business interests are involved. In December 2025, the president announced that Nvidia could ship its H200 AI chips to approved customers in China, provided the U.S. government received a 25 percent share of the sales. Oren Cass of the conservative think tank American Compass testified that the deal showed policy being shaped by executives who treat China as a partner and put short-term profit ahead of the national interest. Defenders of the decision make a different argument, that overly strict export controls can speed up China’s domestic chip industry by guaranteeing demand for local alternatives. Whichever side is right, the episode shows that the race can be relaxed when a sale is at stake. An emergency that can be priced is not a pure emergency. It is also a negotiating position.
The counter-emergency
Opposition to the race has taken on its own emergency language. The most visible tool is the moratorium, which one policy tracker notes is a tool usually reserved for crisis moments: a formal pause on new data center projects while states assess the effects on electric grids, water, land use, and public budgets. On March 25, 2026, Senator Bernie Sanders introduced the Artificial Intelligence Data Center Moratorium Act, and Representative Alexandria Ocasio-Cortez introduced an identical House bill on June 24. The bill would bar building or upgrading AI data centers until Congress passes laws requiring federal safety review before AI products are released and ensuring that the economic gains benefit workers.
States and towns have acted faster than Congress. On July 14, 2026, New York Governor Kathy Hochul issued an executive order that created the first statewide moratorium on new data centers, pausing discretionary environmental permits for facilities expected to draw more than 50 megawatts. One tracker counts 321 moratoriums and restrictions across 32 states, 261 of which are currently in force.
Many of these concerns are practical, not philosophical. Residents worry about power bills, water tables, noise, and farmland. But at the national level, the counter-emergency can mirror the emergency it opposes. One version says AI is so important that normal checks must be removed. The other says AI is so dangerous, or so useless, that building must stop completely. Both skip the ordinary questions a community would ask about any new industrial neighbor: who pays, who benefits, and on what terms.
The money in the middle
Both emergencies are well funded. Leading the Future, a $140 million pro-AI super PAC network whose major backers include Andreessen Horowitz and OpenAI president Greg Brockman, spent $8 million trying to defeat New York Assemblymember Alex Bores in a congressional primary after he sponsored the RAISE Act, an AI safety law signed last December. The group’s stated goal is to identify, keep, and grow pro-AI candidates who will support an AI innovation agenda at the state and federal level. On the other side, former members of Congress Brad Carson and Chris Stewart founded the Public First super PAC network to counter that influence, and its advocacy arm, which pushes for stricter AI requirements, has strong backing from Anthropic.
Across the whole field, AI-focused political committees have raised $107 million and spent $55.5 million in federal races this cycle, with some of the heaviest spending in congressional contests in Alabama, Kentucky, New York, Oklahoma, and Texas. That makes AI one of the best-funded issues in the 2026 midterms, even though it appears on no ballot by name.
A fight that doesn’t follow party lines
It would be simple if this were a clean partisan split, with Republicans for acceleration and Democrats for restraint. It isn’t. The strongest resistance to the president’s AI agenda has come from his own coalition. Senator Josh Hawley has called for defending working people against unchecked AI development, and Florida Governor Ron DeSantis has attacked federal preemption of state rules as bad policy and worse politics, days after his attorney general sued OpenAI. Steve Bannon told CNN that the path to power in America now runs through opposition to tech oligarchs. Conservative critics worry about displaced workers, an economy rebuilt around Big Tech, power concentrated in a small unaccountable elite, and the effects on families, children, and local communities.
Democrats are split too, between members who want a national industrial strategy to compete with China and members who want to pause construction altogether. The alignment that actually exists is less about left and right than about proximity. People who are close to the capital, whether in Washington or Silicon Valley, tend to speak about the race. People who live next to the proposed site tend to speak about the power bill.
Infrastructure, not invasion
Much of the heat comes from how the technology is described. In the race version, AI is a weapon, and the country that builds it first wins. In the alarm version, AI is an invading force that threatens jobs, culture, and possibly survival. A third view gets less airtime: that AI is infrastructure, closer to electricity or the printing press than to a missile or a monster. Infrastructure is not something a country wins or loses. It gets built, priced, regulated, and argued over at utility commissions and zoning boards. Rural electrification in the 1930s involved decades of fights over rates, cooperatives, and public versus private power, but it was handled through ordinary democratic institutions rather than emergency powers.
The infrastructure view has critics. Many researchers argue that general-purpose AI differs from past technologies because it can act and not just transmit, and that some risks, such as misuse in biology or cybersecurity, call for more than zoning rules. Others argue that the competition with China is real even if the rhetoric is inflated, since export controls and military applications are concrete policy choices. And people who see AI as a clear net benefit, pointing to gains in medicine, science, and productivity, are sometimes frustrated that the benefits get lost in the fight. These positions can all be held seriously. What the infrastructure frame offers is a way to hold them without accepting that the only choices are to sprint or to stop.
What a normal person actually thinks
Recent polling gives a clearer picture of public opinion than campaign messaging does, and it doesn’t fit neatly into either emergency.
Americans are uneasy. A Pew survey from June 2026 found that 52 percent of U.S. adults feel more concerned than excited about AI. For the first time, a majority of adults under 30, 55 percent, said the same. About two-thirds of Americans say AI is advancing too fast, and only 2 percent say it is moving too slowly. Seventy-one percent expect AI to mean fewer jobs over the next two decades, up from 64 percent in 2024.
Yet Americans aren’t rejecting it either. Six in ten say they read the AI summaries at the top of search results. A Reuters and Ipsos poll found that only one in three Americans approve of the current pace of data center construction, and 57 percent, including two-thirds of Democrats and half of Republicans, would oppose one in their own community. In the same series of polls, 62 percent agreed that building data centers in the U.S. is necessary to stay competitive in AI.
Taken together, these numbers describe a public that uses the tools, doubts the pace, accepts the national argument in the abstract, and objects to the local cost in practice. That is not confusion. It is the reasonable position of someone who has been told there is a race, has not been asked whether to enter it, and is being asked to pay the entry fee through their electric bill.
The absence of a real vote shows most clearly at the local level, where people have tried to get one. In the village of Wrightstown, Wisconsin, 87 percent of voters said no to an advisory question on whether the village should authorize or take part in public improvements for large data center projects. The village administrator said residents had been insistent at meetings about wanting a direct vote. Other attempts didn’t make it to the ballot. In Box Elder County, Utah, officials rejected three applications for data center referendums, saying the questions could not legally be referred to voters. In Frederick County, Maryland, the state Supreme Court struck down a data center referendum petition signed by 21,000 residents, ruling that its wording was too unclear for the signatures to count.
The federal government has acknowledged the unease. The administration’s March 2026 legislative framework says that some Americans feel uncertain about how AI will affect their children’s wellbeing and their monthly electricity bills. That is an accurate description of the public mood. It is also a description of exactly the questions that emergency framing tends to set aside.
The questions that were skipped
The Cold War taught a lasting lesson about races: whoever defines the finish line gets to set the terms of the contest. A missile gap justified a buildup that outlived the gap. A phantom naval attack authorized a war. A real attack in 2001 produced a security apparatus that is still growing. In each case, the emergency provided the permission, and the permanence came later.
AI may turn out to be one of the most useful tools people have ever built. It may also create real risks that need real rules. Neither possibility requires treating the public as spectators. Somewhere between the call to win and the call to halt is a set of ordinary democratic questions: who pays for the power plants, who keeps the gains, who decides where the buildings go, and who gets to say no. Those questions don’t need an emergency to be answered. They need a vote, and so far no one has been asked to cast one.