- AI is expected to increase demand by 25% or more by 2030
- The solution to the U.S. AI energy crisis is a strategy almost as old as the country’s grid
- America urgently needs a wartime-style Energy Mobilization Plan (EMP) to avoid a power calamity
I hate to be the (circuit) breaker of bad news, but the United States has been blindsided by the energy requirement to fuel the AI revolution, and its current attempt to triage the problem with a slate of emergency measures won’t be enough to avert an energy catastrophe.
Expect higher energy bills, brownouts and blackouts — like the ones that recently took out the grid in Portugal and Spain — within three to five years and the delay or even derailment of America’s leadership in the emerging global AI economy.
AI in data centers is expected to drive nearly half of all new U.S. power demand by 2030. Goldman Sachs sees global energy use by AI data centers surging 165%. Arm’s CEO predicts AI could consume 25% of US electricity by 2030.
The exact number doesn’t matter. The shocking truth does: AI isn’t just a software revolution, it’s a megawatt arms race and America’s grid isn’t ready.
That leaves the tech industry facing a new type of bottleneck — not chip speed or network capacity, but the limitations of running AI over a grid containing transformers installed well before the invention of television.
The U.S. is taking various urgent steps in response to the AI power crisis, including incentives and regulatory measures at both the federal and state levels. In April, President Trump signed executive orders to expedite the use of coal energy to address the energy needs of the AI industry.
Mission accomplished? Hardly. These moves are reactive. It’s all too little, too late.
The fastest way to boost the U.S. grid is by building facilities powered by fossil fuels like natural gas or coal, provided the government expedites the process, but even that takes three to five years. New nuclear plants? Seven to 15 years. Renewable energy sources like solar and wind are options but require additional expensive battery storage facilities to ensure a constant power flow to the AI data centers (in the aftermath of the Iberian power outages, some analysts are pointing the finger at new, cheaper solar as one of the causes).
Power generation is only half the problem. The grid that delivers electricity also needs to be upgraded, an essential task that America has neglected for the last 50 years, with major transmission projects taking a decade or more due to red tape and nimbyism.
Many elements of the U.S. grid – particularly transmission lines and substations – are over 50 years old. Some parts of the electricity infrastructure in cities like Boston, Philadelphia, New York and Chicago date back to the 19th century.
It’s a recipe for disaster, especially in the next three or four years, when AI-driven electricity demand will outstrip supply in many parts of the country. You can’t run the information superhighway on a narrowband country lane and you can’t run Skynet over a Type S rubber-insulated cable installed under cobblestones by someone who was alive when Lincoln was shot.
Is the rest of the world facing the same power problem as the U.S.? Yes. And no.
Europe’s AI power issues are even worse. Goldman Sachs thinks Europe needs to spend $1 trillion to upgrade its grid for AI. (Germany’s right-on decision to ban all nuclear power looks especially idiotisch in hindsight). But China, as usual, is already decades ahead of the U.S. It significantly upgraded its grid in the noughties and now deploys more solar panels per year than the rest of the world combined.
America is not facing an energy shortage; it’s in a slow-motion blackout. And AI is about to fry the grid.
Something borrowed
I’ve had a “lightbulb” moment on this one. To reverse course, America must stop trying to fix problems after they happen and start acting proactively. It can do that by channelling one of the most significant periods of its history and one of its most outstanding leaders.
That’s right — let’s get this baby on a war footing! To that end, I’m calling for the U.S. to tackle the AI-era energy crisis via a Rooseveltian, wartime-style Energy Mobilization Plan (EMP).
EMP Step 1 – Triage and get dirty. Burn (baby burn) gas, diesel and coal, and extend the lifespan of old plants.
EMP Step 2 - Clean it up. Spend the vast sums required to develop clean power — solar, wind, nuclear, running over a shiny new grid — for 2030 and beyond.
America’s EMP mobilization would replace today’s regional grid politicking with a national energy strategy, and incremental upgrades with a massive overbuild. It also might be just the inspiring national mission to get the country to pull together for once, just as Roosevelt’s New Deal did after the Great Depression.
A new new deal
EMP will require significant investment, but Roosevelt proved that where the government leads, the debt market will follow. About 70% of the New Deal was financed through bond issuances, not taxes. America could supplement those monies by re-allocating a wee portion of the $1 trillion a year it spends on defense to safeguarding its economic future. And what about getting the companies causing the AI energy problem, the hyperscalers, to chip in a few bucks?
Let’s break this down:
Redirecting 10% of the U.S. defense and intelligence budget to clean energy mobilization would provide $100B annually.
Levying a 10% AI energy tax on the revenues of America’s five largest hyperscalers adds another $110 billion (Note: I used 2023 numbers; they’re a lot higher now)
Now, add in a government contribution of $50 billion (in year one, it can just use the BEAD money it set aside for broadband upgrades but never actually spent)
Top that up with another $150 billion by selling EMP bonds to the patriotic citizenry
[Sound of tap-tapping on calculator]
Whoa. Looks like treating the coming AI energy crisis like WWII shipbuilding would generate over $400 billion annually ($4 trillion by 2035) more than enough to avert a national power crisis, maintain the United States’ dominance in AI, and build an economy that lasts, if not 1,000 years, certainly more than 10. And all this just by doing what America does best: treating everything as a big fight.
AI pigs in flight
So, will the U.S. do any of this?
In a word: no.
The problem with a country run by a newly re-elected ur-fascist, pre-totalitarian demagogue is that it tends not to think bigly picture any more.
President Trump’s strategy of overwhelming his political opponents, the media and the judiciary by doing everything bad, everywhere, all at once has proven massively effective.
It’s just not realistic to expect the U.S. administration or the country it rules over to deal with a pending AI energy crisis when it’s pre-occupied with tariff terrorism, screwing over Ukraine, helping genocide Gaza and deporting infant US citizens with stage four cancer without due process.
That idea I had about taking money from the defense budget to fund infrastructure? Never going to happen. America is one big war tribe; about the only thing most of its citizens can agree on is how important it is to keep killing people in countries it doesn’t like or hasn’t heard of.
And tax the hyperscalers? Ludicrous suggestion! Today’s hyperscalers are over 100 times larger than the biggest companies in Roosevelt’s time and exert a level of control over the US government to match. Getting them to pay taxes to fix an AI energy problem that they are responsible for causing is like asking Gayle “have you been to space?” King if she understands why people don’t like her anymore (pointless).
The bottom line is that the U.S. is no longer in a position — either politically, morally or culturally — where it can do what it needs to do to avert this particular crisis. My advice? Start stocking up on candles.
Lights out, now, kids!
Steve Saunders is a British-born communications analyst, investor and digital media entrepreneur with a career spanning decades.
Op-eds from industry experts, analysts or our editorial staff are opinion pieces that do not represent the opinions of Fierce Network.