Gas Turbine Shortage Sends AI Developers Back To Boilers And Steam

“The real test is whether turbine shortages persist long enough to justify a durable shift to on-site boiler-based steam generation, which would only be viable if efficiency losses and costs are offset by faster project delivery and policy support.”

Gas turbines are reportedly backlogged; the article argues that data centers will substitute boilers and steam turbines to keep power online, with SpaceX/Tesla pursuing in-house blade work and boiler suppliers like Babcock & Wilcox and Rentech poised to win. That could mean a near-term upside for on-site generation vendors and a faster path to AI campus builds (Applied Digital's 1.2 GW deal). But this rests on efficiency and cost assumptions: steam-turbine routes burn fuel on-site, increasing O&M and regulatory risk, and reducing overall plant efficiency. The piece omits whether turbine bottlenecks are temporary or structural, meaning the model could be a short-lived patch if supply normalizes.

“The shift toward steam-cycle on-site generation is a sign of structural desperation that will lead to higher operational costs and lower energy efficiency for AI data centers.”

The pivot to steam turbines and boilers (Babcock & Wilcox, Rentech) is a desperate supply-chain workaround, not a technological upgrade. While this bypasses the 2030+ gas turbine bottleneck for hyperscalers, it introduces significant thermal efficiency losses and increased operational complexity. Steam systems are inherently less responsive than gas turbines, which is a major liability for data centers requiring rapid load-following. Investors should be wary: this is a 'brute force' capital expenditure strategy that will compress margins for AI developers like Applied Digital. The long-term winners are not the boiler manufacturers, but the grid-scale nuclear and utility-scale providers who can offer reliable, baseload power without this inefficient on-site kludge.

“Steam turbines solve the timing problem but create a permanent efficiency tax and will themselves become supply-constrained if adoption scales beyond 50-100 GW.”

The article frames a supply crisis as solved by a workaround, but the workaround itself has severe constraints. Steam turbines + boilers are 15-20% less efficient than gas turbines at converting fuel to electricity, meaning AI operators pay permanently higher per-MW fuel costs. The 'older supply chain' advantage is real but temporary—if hyperscalers collectively pivot to steam, those three casting bottlenecks simply move upstream to boiler manufacturers. Babcock & Wilcox's $2.4B deal looks like a win, but it's a single contract; scaling to 100+ GW of AI demand will re-create the same shortage dynamics. The article also conflates 'can deliver faster' with 'can deliver at scale'—Rentech's one-year lead time assumes modest volumes, not thousands of units. This is a 2-3 year patch, not a solution.

“Tesla's in-house turbine casting gives it an 18-month execution edge over peers reliant on the three global casters.”

The gas turbine backlog through 2030 forces AI operators to pivot to steam-boiler systems with 1-year lead times, as Babcock & Wilcox's $2.4B contract for 1.2 GW shows. Musk's plan to cast blades in-house at Bastrop could shave 18 months off timelines for Tesla and SpaceX power needs. This highlights a near-term bridge solution while nuclear deals like Google's Finland pact mature. Efficiency losses from the extra combustion step and potential emissions permitting delays are understated risks. Steam supply chains may ramp faster but won't match gas turbine output density for hyperscale loads.