AI stocks weigh bubble risk as power limits cloud capex

Investors may overestimate AI growth amid hyperscaler capex, data center power constraints

Investors may be overestimating the near‑term trajectory of AI investment as hyperscalers confront physical and regulatory limits. Power availability, grid interconnects, and specialized supply chains could slow build‑outs.

Economists are also debating whether 2025 AI spending truly boosted U.S. growth, given import leakages and data center costs, as reported by Finance & Commerce (https://finance-commerce.com/2026/02/ai-spending-us-economic-growth-2025-debate/).

Why it matters: exposure across semis, data centers, and S&P earnings

AI infrastructure capex flows from hyperscalers to semiconductors, networking, power equipment, and data center operators. Revenue realization then depends on monetization in software and services, with knock‑on effects for index‑level earnings.

If large AI players reduce capital intensity, sector earnings could recalibrate. According to Goldman Sachs, a reversion of spending to 2022 levels could cut S&P 500 sales growth tied to AI by roughly 30% (https://www.tradingview.com/news/invezz%3A9d55c57cb094b%3A0-goldman-sachs-sounds-alarm-on-ai-bubble-risk-why-analysts-are-worried/).

At the time of this writing, NVIDIA traded near $190.10, up about 0.14% intraday, based on data from NasdaqGS via Yahoo Finance.

Institutional positioning has turned more cautious. A bank of America fund manager survey identified an “AI bubble” as a top tail risk and flagged potential overinvestment in AI‑related capex (https://fortune.com/2025/11/18/fund-managers-fear-ai-bubble-say-companies-overinvested-bofa-survey/).

central bank commentary underscores that productivity gains may take time to register in macro data. Austan Goolsbee, President of the federal reserve Bank of Chicago, said AI “has not been as big a driver…” of growth as some suggest (https://breakingthenews.net/Article/Fed’s-Goolsbee:-AI-role-in-growth-overstated/65733810).

Sustainability of current spending is not guaranteed. Wolfe Research cautioned that maintaining 2025’s AI capex pace into 2026 is uncertain amid power limits, materials bottlenecks, and regulation (https://uk.investing.com/news/stock-market-news/are-ai-spending-expectations-simply-too-high-4522686).

Scenarios and indicators: capex paths, supply limits, regulation to watch

Base, slower capex, and constraint-driven pause scenarios

Base case: capex grows from elevated levels but normalizes as efficiency improves and deployment scales. Earnings contribution broadens from infrastructure to software, with lags between spend and monetization.

Slower capex: procurement cycles lengthen, ROI hurdles tighten, and some projects slip. Suppliers see order pushouts, while utilities and landlords pace expansions to interconnect timelines.

Constraint‑driven pause: grid interconnect delays, GPU availability, and permitting slowdowns trigger temporary pauses. Revenue recognition shifts right, raising volatility for semis, equipment, and data center operators.

Indicators: utility interconnect queues, capex guidance, GPU supply, regulation

Watch utility interconnection queues, substation timelines, transformer lead times, and regional power pricing. These signal whether capacity is aligning with hyperscaler roadmaps.

Track capex guidance from major AI infrastructure buyers, reported GPU and networking lead times, and regulatory developments on siting, power markets, and data governance.

FAQ about AI bubble risk

Can hyperscalers maintain current levels of AI capex into 2026?

Sustaining 2025’s pace is uncertain amid grid constraints, component lead times, and policy hurdles. Budget sequencing and efficiency gains may force moderation.

How will power and supply constraints affect AI data center build-outs?

Interconnect backlogs and limited high‑end chip supply can delay commissioning, pushing revenue recognition later across semiconductors, power equipment, data centers, and software.