John Steinbach opened his January electricity bill and did a double take: $281, against roughly $100 the month before. The retired Virginia homeowner has lived in the same house for nearly four decades and had never seen a jump like it. His county sits in the heart of America’s data-center corridor, and the machines running artificial intelligence are devouring power at a pace utility customers are only beginning to feel.
That household bill is the small end of a number that Jensen Huang put on the table this week. On Nvidia’s quarterly earnings call, the chief executive said annual spending on AI infrastructure will surge to $4 trillion — roughly four times what Wall Street expects. The money, he said, will eventually land on every ordinary person in the form of electricity bills, subscription fees and, he hopes, jobs.
The numbers behind the forecast
Nvidia’s own results give the projection its authority. The company reported revenue of $81.6 billion for its first fiscal quarter, up 85% from a year earlier, with the data-center business contributing $75.2 billion, a 92% jump that accounted for more than nine-tenths of total sales. Net income of $58.3 billion more than tripled. The second-quarter forecast of $91 billion came in more than $4 billion above analyst expectations, and the company added $80 billion to its buyback authorization.
The market’s response was muted by Nvidia’s standards — the stock has already climbed far enough that even a blowout quarter barely moved it — but the company’s market value, around $5.7 trillion, now exceeds Germany’s projected GDP for the entire year.
Four times the consensus
Huang’s $4 trillion figure is the striking part. Hyperscale cloud providers are currently spending about $1 trillion a year on AI infrastructure, the company’s executives said, and Nvidia expects that to climb to $3 trillion to $4 trillion before the end of the decade. The prevailing Wall Street estimate, compiled by Needham analyst Laura Martin, puts hyperscale capital spending at about $1.03 trillion by 2028. Huang’s number is four times that.
The gap shows how far apart the industry’s two camps are. Cloud providers describe AI investment in measured terms, tied to revenue they can actually see. Huang is describing a world in which AI capacity itself becomes the scarce resource, and companies buy it the way they buy electricity. Martin, who compiled the consensus, said in a research note that Huang’s vision differs from the scenarios the cloud providers themselves describe — and is more interesting.
The spending is already happening
The money side of the argument is visible in the quarterly reports. In the first quarter, Google spent $35.7 billion on capital expenditures, roughly double the prior year; Amazon spent $44.2 billion, the most of the four major clouds; Microsoft spent $30.9 billion, up 84%. Meta was the most aggressive, raising its full-year budget to a range of $125 billion to $145 billion — and the market punished it for that ambition, sending the stock down more than 9% the day after the announcement.
Combined, the four hyperscalers are expected to spend about $725 billion in 2026, and Bank of America predicts cloud providers will issue $175 billion in debt this year, six times the average of the past five years. The buildout is being financed with borrowed money, which means the bills come due whether or not AI revenue materializes.
The bills land on households
That is where Steinbach’s electricity bill comes in. Data centers already consume nearly 40% of Virginia’s electricity, and the grid operator covering the eastern U.S. has seen household bills rise on average about 15% since the AI buildout accelerated. Analysts project U.S. data centers will account for roughly 12% of the nation’s electricity by 2028, with residential bills up an average of 8% by 2030.
Huang’s case is that the spending is rational because the payoff is enormous — a bet that AGI, if it arrives, makes every dollar spent on the road to it look small. The cloud giants are placing the same bet with their balance sheets. Everyone else is paying for it on a monthly basis, whether through a utility bill, a subscription fee or the price of the goods whose makers are paying more for compute.
The scale of the number has also begun to attract skeptics. The largest cloud providers have repeatedly warned investors that AI capital spending is being measured against returns, and Meta’s stock fell after it raised its budget; some analysts argue that $4 trillion assumes a pace of data-center construction that the industry’s supply chain cannot physically support, from power transformers to land. Huang’s answer, characteristically, is that the constraint proves the opportunity: the harder it is to build, the more the capacity that exists is worth. For shareholders of every company funding the buildout, the question is which side of that argument the market ultimately believes.
Jensen Huang has put a number on the AI buildout that dwarfs every forecast the market was using: $4 trillion a year, four times consensus. Nvidia’s earnings back the confidence — revenue up 85%, data-center sales nearly doubled — and the hyperscalers are already spending at rates the industry has never recorded, financed increasingly with debt. The costs are no longer abstract: they are showing up in household electricity bills. The question the market must answer is whether Huang is describing the future or pricing it ahead of itself.
