America is on the cusp of the biggest power-hungry decade in a generation. Federal forecasters expect record electricity demand in 2025–26, even as oil producers are carefully managing output. Yet the main reason your bill is climbing isn’t just regional differences, it’s rules. A thicket of tariffs on essential hardware, marathon permitting timelines, and clogged grid interconnection queues layer a “policy premium” onto every kilowatt-hour. In a textbook supply-and-demand sense, demand is racing ahead, driven by AI data centers and electrification, while policy squeezes supply. Worse, Washington keeps “choosing” technologies, inviting regulatory capture and raising costs for everyone. If we want the AI era to benefit households instead of draining their wallets, we need neutral, pro-entry rules that let the cheapest reliable electrons win. 

Energy inflation tells the tale. Headline prices are running hotter than the 2 percent target, and the pressure points are concentrated in utilities. Electricity alone is up roughly 5 percent, and utility gas is up close to 12 percent versus a year ago. Average retail electricity prices have jumped nearly 9 percent already this year. The burden isn’t uniform. North Dakotans pay the lower energy costs, around 11.7¢ per kWh with a typical bill near $112, while Hawaiians pay roughly 42.3¢ and a $203 monthly tab. That gulf largely reflects system design and policy. North Dakota sits on abundant generation with strong ties to the Midcontinent grid; Hawaii is an islanded system importing fuel and equipment with layers of costs. When transmission is constrained and hardware is expensive, consumers pay.

Energy demand is taking off, mainly driven by data centers powering AI. Data-center electricity use tripled over the last decade to roughly 176 terawatt-hours, and federal analysts expect it to double, or even triple, again by 2028. That would take data centers from roughly 4.4 percent of US electricity consumption to something like 10–12 percent in just a few years. In markets with heavy data-center clustering, wholesale prices near key nodes have surged, reportedly doubling or tripling compared with five years ago, and those spikes bleed into retail rates. The largest campuses are now measured in hundreds of acres; Meta’s complex in Prineville, Oregon, covers well over a hundred. None of this is a problem if supply can scale. But supply is boxed in. 

Start with trade policy. The grid is steel, copper, aluminum, power electronics, batteries, transformers, inverters, and miles of conductor, exactly the things Washington keeps taxing. Many power-sector inputs now carry duties in the 25–50 percent range. The administration has announced additional tariffs on medium and heavy trucks, which will raise logistics costs across energy supply chains. For oil and gas, the exemption record is mixed: while crude itself may avoid new levies, drillers and midstream firms still buy tariffed inputs — steel casing, line pipe, valves — magnifying project costs. Steel and aluminum duties, recently doubled to 50 percent in some cases, raise the price of everything from rigs to transmission towers. On the “green” side, tariff policy is even more tangled. Grid-scale batteries face total duty stacks approaching the mid-sixties percent; aluminum and derivative products also pay 25 percent. Solar modules and cells still sit under safeguard tariffs near the mid-teens, layered atop other levies. The result is not a level playing field but a politicized one where lobbyists fight to be a “protected” winner and consumers lose either way. 

Then there’s the permitting time tax. Big energy projects routinely wait six or more years for approvals; some major transmission lines have been in limbo for more than a decade. Meanwhile, the grid clogs and aging infrastructure strains. Transmission congestion alone added roughly $11.5 billion to customer bills last year. Lengthy reviews no longer deliver meaningfully cleaner or safer outcomes; they mostly deliver uncertainty, which raises financing costs and deters entrants. In economics, that’s classic deadweight loss. 

Interconnection is the third vice. You can’t sell power until a grid operator studies how your plant will affect the system. For years, independent system operators and regional transmission organizations have been drowning in applications, many of them speculative. Backlogs keep new capacity — renewables, nuclear uprates, even industrial cogeneration — from plugging in. Regulators have introduced helpful reforms: “cluster studies” that analyze groups of projects at once and “readiness screens” to ensure entrants have site control and basic financing before staff spends months modeling them. But adoption is uneven. Markets like the California Independent System Operator are making headway with annual intake cycles and clearer milestones; others are still stuck, creating a patchwork of rules that adds friction and encourages forum shopping. Even good projects die on the vine if siting processes collide with wildlife, historic-preservation, and local zoning rules that were never designed for 21st-century energy density. 

Why do we tolerate all this? Because we keep trying to direct outcomes, favoring particular fuels, geographies, or industrial constituencies, rather than setting simple, technology-neutral rules. That invites regulatory capture. When agencies tilt toward the loudest incumbent or the trendiest technology, they’re not discovering the lowest-cost path to reliability; they’re rationing permits and tax credits. Households and factories pay the markup. 

What would a pro-consumer pro-technology agenda look like? 

First, stop playing favorites. Drop the tariff thicket on intermediate goods central to generation, storage, and transmission. Don’t carve out oil-and-gas inputs but punish solar or vice versa; end the game entirely. Let firms source the cheapest safe equipment globally and let market competition decide the mix of resources. When inputs get cheaper, so do bills. 

Second, rebuild permitting around shot-clocks, not calendars. Establish firm timelines and a single lead agency for major projects; if the clock runs out, move to a decision on the record. Focus intensive reviews on genuinely high-impact projects and allow routine, low-impact work, like reconductoring existing lines or swapping transformers, to proceed quickly under programmatic approvals. Provide judicial review, but time-limit it to curb endless litigation. 

Third, finish the interconnection fix. Make cluster studies and readiness requirements universal, but tune them so they screen out pure speculation without blocking smaller independent power producers. Publish transparent cost-allocation rules so developers can plan. Align interconnection with long-range transmission planning so we’re not endlessly studying plants for a grid that doesn’t exist yet. 

Fourth, open the door to firm, zero-carbon baseload. Small modular reactors and advanced designs, including molten-salt and thorium-based concepts, should compete on their merits. That means modern licensing that evaluates designs by efficiency and productivity, not lineage; standardized approvals for repeat builds; and clarity on waste handling. Clear the obstacles so private capital can try thorium-powered nuclear. If it can deliver reliable power at scale, the market will adopt it. If not, resources will flow elsewhere. Either way, consumers win. 

Finally, remember why we’re doing this. AI is a once-in-a-century general-purpose technology. It can raise productivity and living standards, but only if the power system scales without crippling costs.  

We don’t need another round of picking winners and losers on the energy front. We need to remove the artificial barriers that make supplying power slow and expensive. Cut the policy premium, tariffs that pad equipment costs, permits that drag on for years, interconnection rules that reward queue gaming, and America’s engineers, utilities, and entrepreneurs will do the rest. The cheapest reliable kilowatt-hour is the one that’s allowed to be built.

In recent decades, and especially since the release of OpenAI’s large language model ChatGPT in 2022, artificial intelligence use has rapidly spread into almost every industry. And as AI proliferates, so do fears that a wave of mass unemployment will follow. Concerns are widespread that AI will be deployed to accomplish an ever-greater share of the labor needed throughout the economy, leaving fewer and fewer jobs available for human workers.

This fear led Dario Amodei, one of the world’s leading AI technologists, to sound the alarm earlier this year about an impending “white-collar bloodbath.” The Anthropic CEO told Axios that one very possible scenario within the next one to five years is that, “Cancer is cured, the economy grows at 10 percent a year, the budget is balanced — and 20 percent of people don’t have jobs.”

This was predictably latched onto by economic interventionists, such as US Senator Bernie Sanders. “We must demand that increased worker productivity from AI benefits working people, not just wealthy stockholders on Wall St,” Sanders posted on X in response to Amodei’s statement. Sanders later posted that, “With the explosion of AI, new technology and increased worker productivity, we should demand a shorter work week, increased life expectancy and a decent standard of living for all.” Other politicians have gone even further since then. US Senator Josh Hawley, like Tucker Carlson started advocating years ago, supports banning self-driving cars to protect the jobs of car and truck drivers.

Indeed, AI has already created countless efficiencies and new capabilities throughout the economy that have made specific jobs redundant. Consider the global herbicide industry. A Bloomberg article titled “AI-Powered Weed-Killing Robots Threaten a $37 Billion Market” explains:

After almost a century of deploying a more-is-more approach to chemical herbicides, the global agricultural sector is rapidly rolling out advancements that promise to curb the use of weed-control sprays by as much as 90 percent. Using artificial-intelligence powered cameras, the new sprayers can identify and target invasive plants while avoiding the cash crops. If even a fraction of growers adopt the new tools, it could mean a big shift for crop-chemical majors like Bayer AG and BASF SE.

This potential tenfold improvement in herbicide use efficiency may significantly reduce the agricultural demand for herbicide production, and thus put many people out of their jobs in that industry. And this is just one example. From facilitating new scientific research methods, to transforming solar cell production, to improving healthcare safety, to proliferating self-driving car use, AI offers new ways of doing things throughout every industry, which devalues some skillsets in the job market and rewards others.

However, there is an opposite fear as well. Last month, during a press conference the day he won the 2025 Nobel Prize in economics, Joel Mokyr gave a speech about the grand sweep of economic history. Toward the end of his lecture he weighed in on the AI job loss debate: “As I see it, the main concern about the labor market is not technological unemployment. It’s labor scarcity.”

Mokyr is an economic historian who is renowned for looking at the big picture. And when you do that, it is clear that technological unemployment is nowhere near as big a concern as the set of problems that AI will help solve if politicians like Bernie Sanders and Josh Hawley don’t succeed in debilitating it.

Specific jobs becoming obsolete is only one side of the coin of technology’s effect on employment. The other side of the coin reveals why artificial intelligence will lead to no overall reduction in employment opportunities. Like all prior technological revolutions, AI will create at least as many jobs as it eliminates, and most importantly, the new jobs will tend to be preferable to and easier to find than the old jobs.

Unemployment rates have remained relatively constant and quite low in recent decades, currently standing at about 4.3 percent in the US and 4.9 percent across the globe, according to recent estimates. This is approximately as low as they have ever been.

Meanwhile, ever since the industrial revolution, there has been an explosion of technological advancement and proliferation that has almost completely transformed all economic industries and everyday life across the globe.

Virtually every instance of technological progress reduces the need to employ some specific form of labor. Dish washing machines have meant that homes and restaurants need only employ a small fraction of the cleaning staff that was once required per meal served. Refrigeration has conserved countless labor hours preserving foods. Electric power has exponentiated the manufacturing capability of each factory worker by enabling a new class of machinery. And the list goes on and on.

The constantly low unemployment rate, in light of recent history’s technological progress, would amaze the Industrial-Revolution-era activist group known as the Luddites, a radical labor movement of anti-technologists who rioted and destroyed factory machinery to protect their jobs from automation in the early 19th-century. Their initial concern was for the jobs of craftsmen competing with mechanized looms and knitting equipment that allowed a single worker to produce the output of a hundred craftsmen. But the scope of Luddite animosity quickly widened to oppose industrialization generally.

The nineteenth-century Luddites made the same mistake, known to economists as the “lump of labor fallacy,” that antagonists of AI are making today. In a New York Times article titled “Trump, Immigration and the Lump of Labor Fallacy,” Nobel Prize winning economist Paul Krugman explains the fallacy: “This is the view that there is a fixed amount of work to be done and that if someone or something — some group of workers or some kind of machine — is doing some of that work, that means fewer jobs for everyone else.”

Krugman also explains why this “lump of labor” view is false:

When incomes rise, people will find something to spend their money on, creating jobs for workers displaced by technology or newcomers to the work force. Machines do, in fact, perform many tasks that used to require people; output per worker is more than four times what it was [in 1952], so we could produce 1952’s level of output with only a quarter as many workers. In fact, however, employment has tripled. … No, AI and automation, for all the changes they may bring, won’t ultimately take away jobs, and neither will immigrants.

The point about “incomes rising” is key, because it is a predictable consequence of technological automation and it is also the reason, along with new forms of work created by new technologies, why the newly created jobs will tend to be preferable to the old jobs.

Workers will only be displaced by technology if the technology is capable of producing more valuable output, meaning some net-positive combination of cheaper and higher quality, than human labor. Otherwise, mechanization won’t be profitable and employers will stick to employing humans. Therefore, we can safely bet that wherever AI is displacing human workers, consumers are benefitting from some net-positive combination of falling prices and rising quality. This means that consumers will have more income left to spend, and will therefore consume an ever-wider range of goods and services, thus funding more and more niche forms of employment that replace the jobs lost to technological improvements in the production process.

This is crystal clear if you look at the history of labor specialization. Since the mid 19th century, agriculture has gone from employing about 70 percent of US workers to employing only 2 percent. Meanwhile, the productivity of US agriculture has massively increased due to improved farming technology and food has become about 10 times more affordable for blue-collar workers in the last century.

If no new jobs had been created in America since 1840, this process would have left 68 percent of the population permanently out of work. But to the contrary, the modern economy is composed largely of jobs that people in any prior century probably never even imagined. Web designers, pet therapists, dietitians, travel agents, nail technicians, cosmetic surgeons, neuroscientists, and countless others are employed in now-common forms of labor that would have seemed either laughably frivolous or entirely unimaginable just a century ago. “Really, therapy for my dog while my children are starving?”

For a detailed description of one of the most recent and frivolous new industries imaginable, read the article titled “Does Your Plant Need a Nanny? A new crop of caretakers will spritz, polish and prune your houseplants — and even send photos while you’re away.” published by the New York Times in April. These new “plant nannies” offer virtual plant visits, and services such as speaking to, lighting, and photographing plants while their owners are away, in addition to traditional plant maintenance practices.

New and ever-more-niche and personalized forms of work are constantly made possible by the vast abundance of labor and consumption freed up by labor-saving technology. Monotonous or backbreaking tasks are left to machines while humans are freed to invent new tasks to pay each other for with all the wealth that technological efficiencies have saved or created for them.

As amazing as the present is compared to the past, the future can be an even more extreme improvement. If AI proves powerful enough to continuously transform the macroeconomy as the fearmongers predict, and so do I, this will mean such massive productivity gains that machines will eventually be running the farms and factories at almost no cost. This will render necessities more and more affordable until almost every worker can engage in more niche artforms, services, and research areas than employ almost anyone full-time in today’s economy. In addition, whole new realms of technological possibility can be unlocked, in outer space, the Metaverse, and elsewhere.

It is hard to imagine ever reaching a logical conclusion of this process, at which artificial intelligence is better than human labor for literally every purpose. That would even involve convincing those who prefer some services to have a “human touch” for sentimental, aesthetic, philosophical, or spiritual reasons that human labor is unhelpful even to them. If AI does get that powerful, that will be an end of material scarcity in which everything is free and worrying about “unemployment” will make no sense at all.

Instead of worrying about that hypothetical utopia of widespread godlike agency and infinite abundance, let’s focus on reducing human drudgery and increasing material wealth by proliferating artificial intelligence and other technological progress in the here and now.