A typical AI-optimized hyperscale data center uses as much electricity in a year as 100,000 households. The largest now being built are expected to use 20 times that much, the International Energy Agency estimates. That level of energy use underlines why Wall Street suddenly is nervous about the economics of artificial intelligence – and why the Dow Jones Industrial Average fell 498 points Tuesday in its fourth straight day of declines.
The sell-off, which also saw the S&P 500 fall 0.8% and the Nasdaq tumble 1.2%, is being driven by fears the AI boom has inflated valuations far beyond sustainable levels. Investors have watched tech giants pour hundreds of billions into data centers and AI chips without yet delivering proportional profit gains. “Tech companies have to spend to keep up with surging demand, but that demand largely hasn’t turned into profits or productivity,” said Callie Cox, chief market strategist at Ritholtz Wealth Management.
However, the 2025 gains had been concentrated in a narrow set of mega-cap names, the so-called “Magnificent Seven” of Alphabet, Amazon, Apple, Meta, Microsoft, Tesla, and Nvidia. Their dominance made indexes more vulnerable to volatility. The SPDR S&P 500 ETF Trust is now 41% concentrated in just 10 companies, with Nvidia, Apple, and Microsoft accounting for 21.5% of the weighting. When those stocks falter, index performance deteriorates fast. Nvidia shares have slipped nearly 9% since late October, while Meta has fallen 17% over the past month.
Behind the valuation concerns linked to the physical reality of AI infrastructure lie US data centers, which in 2024 consumed 183 terawatt-hours of electricity-more than 4% of national consumption and annual demand for Pakistan. That will rise to 426 TWh by 2030. Energy intensity is particularly acute for workloads like AI; there, advanced GPUs draw two to four times more power than traditional chips. The cooling systems add further load and account for as much as 30% of a facility’s total energy use, while hyperscale sites use trillions of gallons of water annually.
These costs ripple into local economies. In 2023, data centres consumed 26% of all the electricity used in Virginia, placing grids under strain and forcing utilities to invest in expensive upgrades. Data centre demand drove a $9.3bn capacity price increase for 2025-26 in the PJM electricity market, serving parts of Illinois to North Carolina, adding $16-$18 a month to residential bills in some areas. Researchers at Carnegie Mellon University estimate that by 2030, data centres and cryptocurrency mining could push up the average US electricity bill 8%, with spikes above 25% in high-demand regions.
Another fault line is debt financing. Bank of America has warned that the large technology firms are borrowing heavily to finance AI infrastructure – sometimes for capacity well beyond current demand. Amazon recently conducted its first bond sale in three years, partly to help finance AI expansion. Alphabet and Meta too have tapped credit markets for that same reason. This leverage raises risks if revenue growth disappoints expectations.
Sundar Pichai, chief executive officer of Alphabet, acknowledged both the promise and the excess in the current cycle. “I think no company is going to be immune, including us,” he told the BBC, noting “immense” energy needs and slippage on climate targets. AI accounted for 1.5% of global electricity consumption last year and Pichai stressed the need to develop new energy sources if economic growth is not to be constrained. Strategists are divided on whether the current pullback is a healthy correction or the early stages of a bubble deflation. Some, such as James Denmert at Main Street Research, argue valuations still trade at discounts to earnings growth rates, with feeling “mildly bullish and not euphoric.”
Others see parallels to the dotcom era, when concentrated bets on transformative technology led to sharp drawdowns. The difference now is that AI infrastructure is tangible – hyperscale campuses, chip foundries and power contracts – yet the monetization timeline remains uncertain.
To investors, the concentration risk couldn’t be clearer: more than a third of index weight is tied up in a handful of AI leaders. When those stumble, the rest of the market catches cold. The question is now whether the capital-intensive buildout of AI can translate into productivity gains and profits before energy costs, debt loads, and market skepticism force a more painful reset.
