Grid queues are years long. Crusoe isn’t waiting.
Crusoe’s announcement with Bergen Engines describes a commercial agreement covering a reported 750 megawatts of on-site power generation capacity for Crusoe’s AI data center portfolio in the United States. The structure is reportedly a 438-megawatt equipment contract paired with a 310-megawatt letter of intent, a staging approach that locks in near-term capacity while preserving optionality at scale. Both vendor sources were inaccessible ; figures carry “reportedly” throughout.
The generator specs reported in the joint release, 27 units of the B36:45V20AG2 model at 12.5 megawatts electric each, plus 20 units of the B36:45L9AG model at 5 megawatts electric, are the kind of detail that appears in manufacturer press releases, not marketing summaries. The specificity is internally consistent. Financial terms of the agreement weren’t disclosed, which is standard for B2B industrial contracts of this type.
Verification
Qualified Crusoe Energy + Bergen Engines joint release (named parties, named executive, John Adams, SVP of Power, Crusoe) Bergen Engines source DEAD; Crusoe source returned homepage only (SOURCE-MISMATCH). All capacity figures, generator specs, and deal structure require 'reportedly' framing. Financial terms not disclosed.The “why behind-the-meter” rationale came directly from Crusoe’s SVP of Power, John Adams, per the companies’ joint announcement: on-site generation bypasses grid queue times. That’s a vendor claim. But the underlying market dynamic it describes is documented. New large-load electricity customers in most US regions face interconnection queues measured in years, not months. For AI data center operators whose economics depend on deployment speed, waiting for a grid connection is a competitive liability. On-site generation, particularly natural gas baseload, trades that wait time for direct fuel supply risk and emissions exposure.
That trade is worth naming. Natural gas generation at scale creates its own regulatory surface. State-level scrutiny over carbon and methane emissions from behind-the-meter industrial generation is growing, a regulatory gap that the hub’s regulation pillar is flagging for dedicated coverage. EU Methane Regulation compliance implications for US natural gas equipment exports are also in the mix, though that’s a forward-looking issue rather than a current enforcement action. Compliance teams with clients in AI infrastructure should be watching this space.
The Crusoe/Bergen deal is the third AI power supply agreement TJS has tracked at 750MW or above in recent weeks. The gigawatt race piece from May 27 documented how SoftBank, TeraWulf, and NV Energy were locking in compute power capacity; the Oracle/OpenAI Barn groundbreaking (, F02) adds another 1GW to that picture. Crusoe’s behind-the-meter approach is structurally different from those grid-connected deals, it’s solving a deployment timeline problem, not a grid capacity problem. That difference has regulatory and environmental implications that the grid-connected deals don’t carry in the same way.
Warning
Behind-the-meter natural gas generation solves the grid queue problem but creates a new one: regulatory exposure. State-level carbon and methane emissions requirements for industrial generators don't disappear because the load is labeled 'AI infrastructure.' This is the emerging compliance surface that the hub's regulation pillar is flagging for dedicated coverage. AI infrastructure developers using this model should be tracking state environmental permitting requirements now, not after the first enforcement action.
This is the third infrastructure deal led by named parties bypassing or supplementing traditional utility delivery. The pattern isn’t coincidental, it’s a response to interconnection queue pressure that multiple AI infrastructure developers are hitting simultaneously.
Don’t bet on behind-the-meter gas generation remaining a regulatory grey area indefinitely. State environmental regulators are already engaged. Watch for the first state-level permit challenge or emissions reporting requirement applied specifically to behind-the-meter AI generation as the trigger that changes the calculus on this deployment model.