IREN's Energized Megawatts Still Have to Clear the Gate Before They Become Revenue

To an engineer, a switched-on data center is done. To an accountant, it is barely started. IREN has already energized roughly 650 megawatts at its Childress and Sweetwater sites in Texas, pulling the load out of bitcoin mining — yet none of that power earns AI revenue yet. The halls underneath it still need GPU installation, cooling validation, and a formal acceptance that a customer like Microsoft has to sign off on. Energized, in other words, is not the same as monetized. That physical distinction is the real explanation for the number gap that keeps surfacing in IREN's story: an operating annualized run-rate near $1 billion and more than $4 billion of contracted capacity targeted as operational by December 31, set against just $70.5 million of AI cloud revenue actually recognized for the entire June quarter. The company's own co-CEO said this week that investors have gone numb to billion-dollar contract headlines and will credit the run-rate only in delivered, customer-accepted capacity. The cheap, grid-secured power is what makes the whole conversion possible — but a megawatt becomes a revenue-bearing asset only after it survives a fixed delivery sequence. On this company, that sequence, not customer demand, is the binding constraint.

Management considers the power itself largely secured. The path runs from grid interconnection, whose operator queues and studies can stretch past 24 months, to substations and transformers, whose 12-to-24-month manufacturing lead times are described as the industry's most persistent global choke point. Then come data-hall commissioning, validation of the liquid cooling that AI hardware demands, and GPU installation and networking. Only the final gate — formal customer acceptance — triggers revenue collection. Each stage presses on the next, so a slip in any one leaves the capacity that came before it waiting. The per-megawatt economics only accrue to the accepted portion. Recent three-year contracts run above $20 million of annual revenue per IT megawatt, with active discussions around $25 million and a roughly two-year payback on the compute investment. But that figure is full-stack compute revenue — GPUs plus services — not bare-shell rent for an empty shell, and it applies only after acceptance. Cooling is the sleeper variable: it sets the achievable compute density, which in turn dictates the revenue a given megawatt can command.

IREN AI cloud capacity buildoutPlanned AI cloud capacity in megawatts, by delivery year

Planned AI cloud capacity steps from 480 MW delivered in 2025 to a target of 1,200 MW in 2027, and it is across this base that the ~$20-25M per-MW AI revenue economics must operate.

Period	AI cloud capacity (MW)
2025 delivered	480
2026 target (basis of >$4B ARR)	480
2027 target	1200

The buildout shows how concentrated the payoff is. IREN delivered 480 megawatts in 2025, targets another 480 megawatts by the end of 2026 — the capacity that the more-than-$4-billion run-rate rests on and that is largely sold out — and reaches for 1,200 megawatts in 2027. Nothing in that schedule matters until the enrolled megawatts convert from energized to accepted on time. So the number to watch is neither contracted capacity nor energized load. It is the acceptance cadence: how many megawatts pass that final gate each quarter. Microsoft accepted the first block, Horizon 1, a 50-megawatt site, in August; three more blocks, plus the other sites, must land before year-end for the December target to translate into actual reported revenue. Each accepted megawatt moves a slice of the contracted run-rate from a marketing number into revenue. Until it clears the gate, energized capacity is a promise about intent — not proof of progress.