AI narrative faces ‘what-if’ risks as valuations stretch

Investing.com -- Barclays has cautioned that the artificial intelligence trade, which has driven record valuations in U.S. markets, could stumble if risks around data center spending materialize.

In a note dated Thursday, the brokerage said that while AI remains a powerful investment theme, the sector is frothy, but not a bubble (yet), and warned that valuations are more exposed than earnings if capital expenditures falter.

The backbone of the AI boom has been hyperscaler data center investment, which Barclays said is running at about 25% of sales.

That is elevated compared with historical norms but still below the 40% capex-to-sales ratio seen during the dotcom bubble.

The brokerage estimated that data center spending may have contributed about 1 percentage point to the U.S. GDP growth in the first half of 2025, compared with total growth of 1.4%.

A slowdown could carry significant consequences, with Barclays writing that data center disinvestment would likely compound an already bad situation if a recession emerges for other reasons.

Barclays modeled a scenario in which data center capex falls 20% over the next two years, rather than growing 30% annually as currently expected.

Under that assumption, SP 500 earnings per share would face a 3-4% headwind in fiscal 2026 and up to 1.5% in 2027.

However, the report warned the bigger threat is to valuations, projecting a 10-13% de-rating for the index, with hyperscalers and other AI beneficiaries facing multiple compression of 15-20%.

The AI trade remains both high flying and vulnerable to any material shifts in the narrative, Barclays said.

The brokerage identified three key risks that could stall the AI build-out: power constraints limiting new data center capacity, efficiencies that reduce the need for compute, and funding needs outpacing cash generation.

Electricity supply is already under strain, with Barclays noting that PJM power prices for 2026-27 delivery spiked 22%, and that without a cap, the increase would have been closer to 45%.

The U.S. Department of Energy has also warned of higher outage risks if new capacity fails to keep pace with demand.

The consequences extend far beyond technology. When the open-source DeepSeek-R1 model was released earlier this year, claiming performance comparable to leading models at lower cost, the selloff spread from semiconductor stocks such as Nvidia to natural gas operators and utilities that supply power to data centers.

The stocks most vulnerable to a disruption in the AI growth narrative could be predominantly from auxiliary industries, Barclays said, highlighting energy, industrials and networking firms.

Even as operating expenses rise amid a talent war, hyperscalers have so far been able to outpace costs with revenue growth, providing some cushion.

Yet Barclays pointed to a surge in speculative behavior. Its Equity Euphoria Indicator stands at 11.9%, nearly two standard deviations above its long-term average, underscoring elevated risk.