Melius Research Says AI's Real Bottleneck Isn't Turbines or Capital Anymore. It's Permission — With 476 Gigawatts Stuck in One State's Queue.

The lines can carry the power. Whether they're allowed to connect a new customer is now the open question. Photo: Robert So / Pexels

Melius Research managing director James West told clients in a note on August 24 that "the binding constraint on growth has shifted" for AI infrastructure. "It is no longer turbines, capital, or customer intent," West wrote. "Rolling into midterms, it is permission." The scale already waiting on that permission is visible in Texas alone: as of August, the Electric Reliability Council of Texas was reviewing 476 gigawatts of proposed new electricity demand, most of it tied to data centers — roughly six times the state's own current peak demand.

Texas Governor Greg Abbott has already halted new data-center connections to the grid pending an assessment of energy and water impacts — the same audit referenced in a Bank of America note earlier this month that named the state's governor's race a market bellwether. West's note adds a structural detail: deal terms themselves are changing shape in response, "migrating toward customer-funded generation and contractual curtailability" — meaning developers increasingly pay to build their own power supply and accept being cut off first during grid stress, concessions extracted directly by the political environment rather than by engineering limits.

None of this means capital and hardware constraints have disappeared — memory shortages and financing structures remain very real limits, as this week's other headlines show. But Melius Research is a firm institutional investors read closely, and its explicit reframing of the binding constraint, backed by a specific queue number rather than a vague sense of "backlash," is itself evidence the calculus has shifted. Two months out from the midterms, the industry's growth rate may depend less on how fast turbines ship than on how a handful of state races turn out.