That is according to new analysis from Wood Mackenzie reported by Bloomberg.
Of the 1,066 GW currently requested, roughly 12% of projects have secured firm service commitments and a further 17% look likely to proceed, together accounting for the 28% figure. The remainder, an estimated 768 GW, is what Wood Mackenzie characterises as largely “phantom load”: speculative or duplicate requests unlikely ever to reach construction.
The mismatch stems from how developers are approaching interconnection queues. Ben Hertz-Shargel, Wood Mackenzie’s global head of grid edge, has said the pipeline is weighted toward a small number of massive, speculative projects concentrated in the South and Southwest, and developers are submitting the same project to multiple utilities simultaneously while they shop for the best terms and site. The practice inflates headline demand figures without adding genuine buildable capacity.
Texas illustrates the scale of the distortion. ERCOT is currently tracking 474 gigawatts of connection requests, more than five times the grid’s record peak demand, a mismatch significant enough that it fed directly into the state’s audit of some 300 data centre projects ordered by governor Greg Abbott in August. Utilities elsewhere are already responding: Exelon has cut its own data centre pipeline by nearly 40%, down to an 11 GW queue it considers credible, and grid operators more broadly are introducing application fees, collateral requirements and credit-rating thresholds to filter speculative bids from serious ones.
The pattern is not confined to the US. Capacity has previously reported on “phantom investments” in the UK data centre sector, where a Guardian investigation raised similar doubts about the credibility of pledged capacity, and on Ofgem’s own crackdown after UK demand connection applications rose from 41 GW to 125 GW in under a year, driven largely by speculative data centre projects.
For grid operators, the practical consequence is a widening gap between what is announced and what gets built, one that complicates both regulatory planning and site-selection decisions for developers.
Wood Mackenzie’s analysts do not expect the visibility problem, distinguishing real commitments from speculative bids, to resolve before 2028, meaning interconnection queues are likely to remain a poor guide to actual future demand for some time yet. That uncertainty also underlines the stakes behind FERC’s recent push for grid operators to reform large-load interconnection rules, five of which opted to delay rather than file reforms this month.
RELATED STORIES
Texas hits pause on its own AI boom: Inside the 300-project grid audit
UK AI data centre investments: Empty promises or slow growth?
UK energy regulator promises power crackdown for data centres
