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

IEA data point to a threefold increase in US gas-power investment in 2025, with orders for new gas-power plants reaching 130 GW globally, a 25-year high. US electricity demand is projected to rise 2% annually from 2026 to 2030, with data centres driving half of the increase, while US investment in fossil-fuelled power plants is expected to overtake China’s in 2026. The report also notes a $105bn global spend on grid upgrades, power equipment and electricity generation for data-centre infrastructure, alongside weaker US wind and solar forecasts after renewable tax-credit phase-outs.

Analysis

The first-order winner is not just US gas supply; it is the entire domestic thermal-power value chain. The bottleneck is now turbine availability and project execution, which means pricing power accrues to equipment makers, balance-of-plant contractors, and utility-scale developers with merchant exposure. More importantly, this is a capacity-scarcity cycle, so the best economics likely sit with firms that already have interconnection rights, pipeline access, and equipment under contract rather than “AI story” names that still need permits and metal.

This also creates a second-order wedge in power markets: captive generation for data centers can suppress local grid demand growth while increasing regional gas basis volatility and congestion rents. That favors midstream assets with exposure to constrained basins and takeaway bottlenecks, but hurts renewables developers that rely on quick interconnection and policy subsidies rather than behind-the-meter economics. The clean-tech implication is not extinction; it is bifurcation—SMR, geothermal, and storage win where they can sell reliability and speed, while standalone wind/solar on weak grid nodes lose relative attractiveness over the next 12-24 months.

The market may be underestimating how cyclical this looks: if AI capex pauses or hyperscalers internalize the power buildout too aggressively, turbine orders can cliff after 2026 because this is mostly a front-loaded capacity build, not a perpetual demand curve. A key reversal trigger is a meaningful step-up in grid interconnection reform or a cheaper-than-expected battery+storage stack that restores renewables competitiveness; both would pressure gas-power expectations and compress “AI power” multiples. On the other hand, if gas turbine lead times keep stretching, the scarcity premium can persist longer than consensus expects, especially for companies with booked-out delivery slots.

The contrarian angle is that the real trade may be against the obvious hyperscalers and into picks-and-shovels industrials. Investors are likely overpaying for “AI power demand” beta in software/platform names while underpricing suppliers with near-term backlog conversion and contractual pricing. If the buildout remains supply-constrained, earnings revisions should show up first in the equipment, transmission, and engineering layer rather than in the data-center operators themselves.

Key Decisions for Investors

• Long CEG / NRG / VST on a 6-12 month horizon: these names gain from scarce dispatchable power and tighter regional capacity pricing; prefer pullbacks because the upside comes from contract repricing and merchant tailwinds, not immediate macro beta.
• Long GE Vernova (GEV) vs short Invesco Solar ETF (TAN) for a 3-6 month pair: the former has direct exposure to turbine scarcity and grid capex, while the latter is more exposed to weaker near-term utility-scale demand and policy uncertainty.
• Long KMI or WMB on a 6-18 month basis: captive gas generation should support basin throughput and transport demand; use as a lower-volatility way to own the gas-infrastructure bottleneck with attractive downside if power demand slows.
• Buy GEV Jan-2027 calls funded by selling out-of-the-money TAN calls: favorable if turbine scarcity extends and power-equipment backlog converts faster than renewables can re-rate.
• Avoid or underweight pure-play utility-scale solar developers for now; if seeking clean-energy exposure, rotate into storage or advanced geothermal names with reliability value and shorter interconnection sensitivity.