Data center demand drives 66% surge in natural gas power plant costs

BloombergNEF says the cost to build a new combined-cycle gas turbine plant has risen 66% in two years, from under $1,500 per kW in 2023 to $2,157 last year, while completion times are now 23% longer. The report highlights a gas turbine shortage, with equipment prices expected to be 195% above 2019 levels by year-end and waitlists extending into the early 2030s. For data centers and AI infrastructure, the article points to rising natural gas costs as a headwind and notes Google is pushing a cheaper alternative using renewables plus long-duration storage.

Analysis

The market is treating gas-fired power as the fastest path to AI capacity, but the bottleneck is shifting from fuel economics to industrial capacity. That changes the winner set: the most exposed beneficiaries are not upstream gas producers, but turbine OEMs, EPC contractors, and grid-equipment suppliers with pricing power and long backlogs; the losers are hyperscalers that assumed they could self-build generation on a software-like timeline. The second-order effect is that a higher share of data-center capex will be tied up in fixed assets with multi-year completion risk, which lowers the IRR of AI infrastructure and pushes more projects toward leased/utility-backed capacity.

The key risk is that this is not just a cost inflation story, it is a time-to-power story. Delays into the early 2030s imply a widening gap between announced AI demand and delivered compute, which can become a constraint on revenue growth, model rollout, and cloud utilization if grid interconnects and generation fail to keep pace. In the near term, any announcement of large hyperscaler PPAs or utility partnerships can still move the stocks, but over the next 6-18 months the market should start discounting schedule slippage and margin compression rather than just headline capacity additions.

Google’s renewable-plus-storage angle is the more interesting strategic hedge because it attacks both scarcity and political backlash. If long-duration storage can be contracted at scale, it creates an option value for firms that can tolerate a slightly slower ramp in exchange for lower execution risk and better public optics; that should incrementally favor GOOGL versus peers. The consensus is underappreciating that gas scarcity may end up being bullish for non-gas dispatchable solutions faster than it is bullish for gas utility economics, because the marginal buyer is tech, not regulated load.

Key Decisions for Investors

• Long GOOGL / short MSFT-META basket for 3-6 months: express the view that renewable-plus-storage execution will be rewarded while gas-heavy self-build strategies face rising capex and delay risk. Best risk/reward if long leg is sized smaller and financed with an equal-dollar short basket.
• Buy call spreads on GOOGL 6-12 months out: the catalyst is a broader market re-rating of AI power strategy as a competitive advantage, not just compute spend. Structure for moderate upside, since the thesis is incremental rather than binary.
• Avoid chasing MSFT and META on data-center expansion headlines; consider using rallies to sell covered calls or trim exposure over 1-3 months. Execution risk is now a hidden tax on AI capex, and the market may not fully price schedule slippage until project milestones miss.
• Long industrial infrastructure names with turbine/grid bottleneck exposure versus broader tech capex proxies for a 6-12 month trade. The cleanest expression is through suppliers with existing backlogs and pricing power, since they benefit from the scarcity premium rather than the absolute level of power demand.
• Set a trigger to fade gas-generation enthusiasm if turbine lead times extend further or new project costs rise again over the next quarter. At that point, the market should rotate from 'gas solves AI power' to 'power scarcity slows AI monetization,' which is a materially different equity regime.