AI boom means US is now ‘investing more’ in fossil-fuel power than China

The IEA says US gas-power investment surged threefold in 2025 and is expected to overtake China’s fossil-fuel power spending in 2026, driven by rapid data-centre growth and AI demand. Global orders for new gas-power plants hit 130 GW in 2025, a 25-year high, while data-centre-related grid, power equipment and generation investment reached $105bn. The report also warns that US demand is constraining turbine availability elsewhere, even as it supports emerging clean technologies such as small modular reactors and advanced geothermal.

Analysis

This is less a generic power-demand story than a capital-allocation shock that shifts scarcity rents toward equipment, balance-sheet capacity, and project sponsors who can self-build generation. The bottleneck is not gas supply; it is turbine availability, interconnection queues, permitting, and EPC execution, which means the first-order beneficiaries are likely to be industrials and infrastructure names with near-term manufacturing slots, not the eventual generators themselves. Over the next 12-24 months, the market may still underprice how much AI-driven load growth is being solved off-grid, which effectively privatizes power demand and shortens the path to new fossil capacity.

The second-order winner set likely extends into gas midstream and gas turbines, while the loser set is broader than clean-power developers: merchant power pricing gets more volatile, renewables attached to congested grids face longer monetization timelines, and utilities with weak rate-base flexibility risk being bypassed by captive builds. There is also a geopolitical angle: if US captive demand absorbs turbine supply, deployment schedules outside the US slip, creating a delayed-revenue problem for emerging-market grid buildouts and for any strategy predicated on rapid global electrification. That said, the concentration of AI capex in a handful of hyperscalers makes this demand more cyclical than secular-investor consensus implies; a pause in AI capex or a compute-efficiency breakthrough could quickly unwind the urgency premium.

The contrarian view is that the current reaction may be too linear: more data centers do not automatically translate into more fossil generation if utilities, regulators, and hyperscalers lean harder into behind-the-meter nuclear, geothermal, storage, and demand-response over the next 2-5 years. The risk window is bifurcated: near term, the trade is about equipment scarcity and project backlogs; medium term, it is about whether capital markets finance captive power at acceptable IRRs once turbine inflation and gas price volatility are fully incorporated. If power-price spreads compress or permitting accelerates for clean firm power, the current gas-heavy buildout narrative could flip from scarcity premium to stranded-asset concern.

Key Decisions for Investors

• Long industrials with turbine exposure (GEV, ETN) vs short renewable developers with grid-connection dependence (NEE, FSLR) for 6-12 months; thesis is that bottlenecks monetize equipment makers before they monetize kilowatt-hour producers.
• Long gas infrastructure / midstream names with volume leverage (KMI, WMB) on a 3-9 month horizon; risk/reward is favorable because captive data-center builds can create sticky throughput even if power-generation economics stay noisy.
• Pair trade: long utility-scale power equipment / grid capex beneficiaries (PWR, EMR) vs short merchant power names exposed to delayed supply response; catalyst is that AI load growth raises capex faster than it raises fully contracted earnings.
• Buy upside in GE Vernova (GEV) via 6-12 month call spreads; objective is to capture turbine scarcity pricing, with defined downside if order timing slips or hyperscaler capex pauses.
• For cleaner convexity, consider a small long-duration basket in SMR/geothermal enablers (OKLO, VST or a diversified clean-firm proxy) as a hedge against the consensus over-owning gas; this works only on a 2-5 year horizon and should be sized modestly.