Keeping an Eye on the AI Prize

You’re reading Dispatch Markets, a weekly newsletter on economics featuring Scott Lincicome, Kyla Scanlon, Karl Smith, Marian Tupy, and Adam Ozimek. To access more Dispatch reporting and analysis, become a member today. Welcome to Dispatch Markets! The public debate over AI here in America is awash with concerns about the local impact of data centers and the employment consequences for the next round of college graduates. These serious issues should not be ignored. They shouldn’t, however, be the central question. When AI was an idea on the horizon, the tone of conversations about AI was wondrous and hopeful. We imagined all the ways that technology could change the world and the future of humanity. As we have rounded the corner on the reality of usable AI, the tone has changed. Rarely is wonder or even optimism part of the discussion. When we focus only on the constraints or the negative consequences associated with AI, we miss the forest for the trees. Meanwhile, China seems to grasp the scope and gravity of the technological revolution we are facing. We need to understand, as the Chinese seem to, that whoever wins the AI race will set the tone for the future. What ultimately matters is American AI domination, with the greater goal of setting humanity’s future on a liberal trajectory, instead of an authoritarian one. AI Gung Ho* Illustration by Noah Hickey/The Dispatch (Photos via Getty Images). The usual policy framing around AI begins with considering palpable harm and asking how to reduce it. Data centers create burdens for local communities. Advances in AI model sophistication increase cyber risk. Automation of clerical “laptop” work threatens to displace millions of white-collar workers. The underlying concerns in each example are valid, yet lead to a suboptimal approach. Our old practices are designed to minimize harm and treat forgone capability as the price of mitigation. Sufficient capability, however, is not a foregone conclusion. We need to operate under a national imperative that focuses on maximizing capability growth and letting harms enter as constraints to be measured, mitigated, engineered against, or compensated for, rather than allowing them to stall or prevent growth. The central question is whether AI will empower free and open societies—ones grounded in liberalism (broadly defined)—or support a censored and surveilled totalitarian order grounded in a byzantine adherence to ideological dogma. This may sound like an alarmist exaggeration. It is not. The world is increasingly split between two superpowers, the U.S. and China, each leaning heavily in its own direction. The advance of artificial intelligence raises the stakes in this preexisting split, not because of the myriad things that might go wrong with AI, but because of the breadth, scope, and depth of all that is likely to go right. All other concerns, as real and valid as they are, ought to be subordinate to this one. It means our primary orientation toward AI should be around its potential successes rather than its failures. AI lowers the cost of cognition. When it works well, AI shortens every segment of the gap between suspecting a problem and effectively acting on it. Or between that first spark of inspiration and its full-throated actuality. That gives AI leverage, as a force multiplier, greater than any innovation since at least the advent of the printing press. Ukraine, for nearly two years now, has been using AI-assisted guidance to keep its strike drones locked in on targets despite Russian jamming attempts. The U.S. Defense Department’s Project Maven uses artificial intelligence to compile massive streams of data into a single unified picture of the battlefield. In doing so, it allows commanders to detect threats faster, thereby improving the speed, precision, and ultimately the lethality of the response. In cyber operations, the Cybersecurity and Infrastructure Security Agency is using a frontier model to hunt down bugs that hostile foreign services would otherwise find. The status quo is untenable. Let’s take a look at the most tangible debate about how AI might go awry: the location or placement of data centers. In the fight for AI dominance, these debates have become chokepoints. They will determine whether the industrial base of American AI development expands at a competitive pace or risks being overtaken as delays and cancellations snowball. Any deliberation over the harms that might arise from data center development is incomplete without a complementary account of how to mitigate those harms while staying as close as possible to the maximum practicable rate of compute build-out. The current process is practically the opposite of that, and policymakers at all levels lack both the roadmap and the will to update it. Below are three examples of local, state, and federal rules that have impeded the data center build-out. Northern Virginia, the epicenter of the U.S. data center boom with roughly 250 facilities, has been unable to escape local backlash. Early last year, Loudoun County terminated its longstanding policy under which, if county administrators certified that a project satisfied the local zoning ordinance and building codes, construction could commence. Now every single project will have to undergo a legislative review that could add months, if not years, to the approval process. A process that once ensured compliance with zoning and building rules now triggers public hearings, discretionary votes, and all the vicissitudes of local politics. In San Jose, California, a proposed Microsoft data center campus was stalled for more than five years amid repeated hearings before the local planning commission and city council, and amid demands to revise the environmental impact study required by state law. At the Susquehanna nuclear plant in Pennsylvania, Amazon tried to solve the AI power problem by building the data center next to the generator. Nonetheless, statutory requirements meant the Federal Energy Regulatory Commission still had to analyze whether and how that power might be diverted from the regional grid, who would bear the loss of that power, and whether the arrangement threatened reliability or consumer costs. In November 2024, the agency rejected the amended interconnection agreement, leaving the project in limbo. These three cases together elucidate how reasonable efforts to forestall harm can make the path to data center development a tortuous one for developers. The United States has to treat each of these as a compounding constraint on the rate at which it can turn capital, energy, labor, and land into national computing power. AI’s double-edged sword. Just about every capability of AI can be used for offense or defense. The very same models that let a crime syndicate run cyberattacks at an industrial scale also let a corporation sift through its code to find bugs before they become breaches. Both sides get the multiplier. The question is how to turn that multiplier into an enduring strategic advantage before the other side does. Trying to hold back the future by imposing regulations on models’ capabilities will only make our defenses weaker, while attacks by China or common criminals become more effective. Cyber policy, therefore, needs to ensure our society’s defensive adoption compounds faster than offensive exploitation. Advertisement Stay ahead of the policies shaping free enterprise and impacting American business. Get the U.S. Chamber’s free newsletter for insights on the economic policy, workforce trends, and regulatory landscape that affect businesses and markets. By subscribing you agree to receive communications from the U.S. Chamber of Commerce. Likewise, worries about job displacement should be addressed, in part, by solutions grounded in the proportional productivity gains that inevitably accompany workforce reductions. If a task that took 10 workers now takes seven, that is a displacement problem and also a 43 percent gain in output per worker. Higher productivity has in the past led to higher wages. But let us suppose the coming displacement is so rapid and massive that workers cannot adjust fast enough to take advantage of those gains. Even in that case, higher productivity still leads to higher GDP, and higher GDP leads to increased tax revenue. A nation of displaced workers will require tax revenue. Retraining costs money. So does the health and income support that workers may need during transition. Returning to work may also require transportation, broadband, relocation assistance, or some combination of the three. The money has to come from somewhere. A policy that tries to regulate displacement away will starve itself and find that it is unable to address the labor market stresses that will inevitably leak through. Even delaying the advancement of American AI has a strategic cost. The AI models and industrial base that reach scale first will shape the infrastructure and standards that the rest of the world relies upon. If the U.S. cannot overcome its domestic obstacles to AI advancement—or, to be frank, AI dominance—then we will cede the future to someone who can. The way forward. We need a framework that aims for U.S. AI dominance, and the objective must be to grow our AI faster than China’s. Metrics like data center capacity and frontier-model lead time don’t capture everything, but they indicate whether policy is moving in the right direction. The potential harms ought to remain in the analysis. They provide discipline and serve as constraints on the rate of progress the U.S. can sustain. That’s a lot. What they are not, though, is a set of independent targets we implicitly pay for with deceleration in our AI capability growth. Even economists, who are invariably trained to consider good and bad outcomes, are not used to thinking in terms of national imperatives. For example, an environmental economist researching climate change policy habitually frames the objective as reducing carbon emissions as efficiently as possible, even if the resulting reduction falls short of what many policymakers might prefer. Such an economist has never been trained to focus on minimizing the emissions that result from the maximum possible increase in available energy. Yet this is exactly the kind of imperative we need to secure American dominance in AI. Sixty-five years ago, Americans watched in shock as Yuri Gagarin became the first man to travel into outer space. Our response to that event remains the stuff of legends. We, of course, ended up beating the Soviets to the moon. But winning the space race demanded an unprecedented marshaling of resources and unity of focus. We are now engaged in a similar race against China. Only the stakes today are much higher. If America is to win the AI race, then AI policy needs a full-scale reprioritization. Myriad things might go wrong as AI advances. That was true with every technological revolution from the steam engine to the internet. There are moments, however, when the fate of liberalism itself hangs in the balance. Arguably, this was true when we faced a hot war with Japan and Nazi Germany. Unquestionably, it was the case when we faced a Cold War with the Soviet Union. And it is the case now. A China that becomes AI-dominant will soon become militarily and economically dominant. It will eventually come to control the infrastructure of modern life. With that, it will be able to impose censorship and surveillance worldwide. We cannot risk that possibility. *Author’s note: Today we identify the phrase “gung ho” as a U.S. Marine Corps battle cry, used in popular conversation to indicate enthusiastic support. Historically, it is derived from the Chinese phrase gōngyè hézuòshè, meaning “industrial cooperation.” During World War II, its contracted form, gōnghé, which simply means “work together,” was overheard by Maj. Evans Carlson, who turned it into his battalion motto, and from there it spread to the rest of the Marine Corps and American society generally. Markets FTW America’s major airline hubs are maxed out, and the incumbent airlines are pushing Congress to double down on the $12.5 billion it already has dedicated to upgrade the nation’s outdated air traffic control system in the wake of the deadly collision over Reagan National Airport involving an Army Black Hawk helicopter and an American Airlines jet last year. Aircraft manufacturer Electra.aero isn’t waiting on Congress or the FAA to get its act together. Electra has developed and successfully tested a 17-passenger point-to-point airplane that utilizes blown-lift technology to take off on runways as short as 150 feet. Its hybrid-electric propellers are quieter than conventional aircraft, meaning it could realistically take off and land from airfields distributed within an urban area. Such point-to-point travel would eliminate the massive air traffic control problems associated with large regional hubs. Chart of the Week The 2022 CHIPS and Science Act authorized $280 billion in spending to briefly triple total spending on manufacturing facilities. Yet manufacturing employment has barely budged. It’s another lesson that industrial policy is rarely good growth policy. Disclaimer: The opinions expressed above do not necessarily reflect those of the presenting sponsor.