Babcock & Wilcox Backlog Jumps 470% on $2.4bn Gas Order for AI Data Centers

Babcock & Wilcox Backlog Jumps 470% on $2.4bn Gas Order for AI Data Centers A $2.4bn gas-fired build for AI loads, plus UBS's call for 25 GW of new Chinese data centers by 2029, marks power as the AI bottleneck. Babcock & Wilcox shares closed at $14.54, up 129% year to date, after the equipment maker booked a $2.4 billion design-build contract with Base Electron for 1.2 GW of natural gas-fired power, lifting its backlog 470% to $2.8 billion (2026-05-21)¹. Management guided 2026 core adjusted EBITDA to $70 million to $85 million, roughly 80% year-on-year growth, and said that figure excludes any further data center upside¹. The order puts a price and a fuel on a demand story that until now had traded mostly as a thesis. Babcock says Base Electron is weighing a second 1.2 GW option, and its global pipeline now exceeds $12 billion¹. Those are gas turbines being ordered to serve compute that does not yet exist, on the wager that AI will draw electricity the way earlier waves of industry did. BloombergNEF reached a blunter view. In a report covered by Dezeen on Tuesday (2026-05-19), the think tank found that the data center build-out behind AI will keep fossil fuels in use for longer, with AI-driven electricity use emerging as a key new source of demand through the coming decade². When a developer needs firm power quickly, the gas turbine gets built, not the solar panel. China is making the same bet at scale. Ken Liu, an analyst at UBS, expects the country to add another 25 GW of AI data centers by 2029, after building just 5 GW over the past two years, according to Economist reporting from Saturday (2026-05-17)⁴. That is a fivefold acceleration in two years⁴. China's edge is the power bill. Chinese AI operators pay around half the per-kilowatt-hour rate of many American ones, according to official figures cited in the same reporting, and because Beijing sets residential prices separately, there is little public backlash to building power-hungry infrastructure⁴. Cheap, plentiful electricity is becoming as much an input to AI competitiveness as advanced chips. The grid is the binding constraint. Energy experts warn that surging AI data center demand is pressuring aging grids and battery storage, forcing governments and technology firms to accelerate upgrades, parliamentnews.co.uk reported on Tuesday (2026-05-19)³. Power, not silicon, is shaping up as the limit on how fast AI can scale. That is what makes Babcock's order book worth scrutiny rather than applause. The headline backlog sits on a fragile balance sheet, with stockholders' equity of negative $131.5 million and a 6.50% note due for refinancing in 2026¹. A demand thesis is only as good as the developer's ability to finance the steel. Whether Base Electron commits to the second 1.2 GW block will say more than the contract already signed¹. Order a gas turbine and you commit to burning gas for a decade, with knock-on demand for the molecules that feed it². So far the orders are running well ahead of the megawatt-hours they are built to serve¹.