Barclays has cautioned that the artificial intelligence sector, which has driven U.S. market valuations to record levels, could face headwinds if spending on data centers slows.
In a note released Thursday, the bank said that while AI continues to be a strong investment theme, the sector is “exuberant, but not yet a bubble,” and highlighted that valuations are more vulnerable than profits if capital expenditures weaken.
Hyperscalers’ investment in data centers has underpinned the AI boom, with Barclays estimating that such spending accounts for roughly 25% of total sales. While high by historical measures, this remains below the 40% capex-to-sales ratio seen during the internet bubble.
Barclays calculated that data center spending may have contributed around 1 percentage point to U.S. GDP growth in H1 2025, relative to overall growth of 1.4%. The bank warned that a slowdown could have material consequences: “data center divestment would likely worsen an already bad situation” if a recession arises for other reasons.
In a scenario modeled by Barclays, a 20% decline in data center capital expenditure over the next two years—rather than the projected 30% annual growth—would reduce S&P 500 earnings per share by 3-4% in fiscal 2026 and up to 1.5% in 2027.
The report emphasized that valuations face the greatest risk, projecting a 10-13% decline for the S&P 500, with hyperscalers and other AI beneficiaries potentially seeing a 15-20% compression in multiples. “AI commerce remains both highly capable and vulnerable to any material shift in the narrative,” Barclays said.
The broker identified three primary risks to AI development: constraints on energy limiting new data center capacity, efficiency improvements reducing the need for computing power, and financing requirements exceeding cash generation.
Electricity supply is already under strain. Barclays noted that PJM electricity prices for 2026-27 delivery have risen 22%, and without a cap, “the increase would have been closer to 45%.” The U.S. Department of Energy has also warned that blackout risks could rise if new capacity does not keep pace with demand.
The impact could extend beyond tech companies. When the open-source DeepSeek-R1 model launched earlier this year, claiming performance on par with lower-cost advanced models, the market reaction spread from semiconductor stocks like Nvidia to natural gas providers and utilities supplying data centers. “Stocks most vulnerable to disruption to the AI growth narrative may primarily be in ancillary industries,” Barclays said, highlighting energy, industrials, and networking companies.
Despite rising operating expenses amid a “war for talent,” hyperscalers have largely offset costs through revenue growth, providing some buffer. However, Barclays noted an increase in speculative activity, with its stock market euphoria indicator at 11.9%, nearly two standard deviations above the long-term average, signaling heightened risk.
This content is for informational purposes only and does not constitute financial, investment, or other professional advice. It should not be considered a recommendation to buy or sell any securities or financial instruments. All investments involve risk, including the potential loss of principal. Past performance is not indicative of future results. You should conduct your own research and consult with a qualified financial advisor before making any investment decisions.