Cloverleaf Infrastructure crossed the wire at 09:30 Eastern — 13:30 UTC — on 21 August, seven minutes after this window opened, announcing a strategic partnership with Nvidia covering design tooling and go-to-market work, and disclosing that Nvidia "made a minority investment in Cloverleaf to further support the build out of AI factories."

What the common framing gets wrong

Every headline attached a figure of several hundred million dollars. That number is not in the release — it comes from a Wall Street Journal report published about an hour earlier, describing talks. Two separate things have been fused: a press release announcing a partnership with an undisclosed minority stake, and a scoop carrying a number that neither company will stand behind. Read the primary document and there is no amount, no percentage, no valuation, no megawatt total, no project count and no named customer. The absence is verifiable on the text; the figure is not.

"Multiple GW-scale projects" has no denominator

The release describes a pipeline of "multiple GW-scale projects across North America." Cloverleaf's product is powered land — parcels with utility interconnection agreements attached. A gigawatt-scale project in that business is a grid position, not a building, not a tenant and not an installed accelerator. The unit being counted is permission to draw power, which is a necessary condition for compute and nowhere near a sufficient one.

Nvidia is financing its own demand, again

The company was founded in 2024 and capitalised with $300m from Sandbrook Capital and NGP Energy Capital. Nvidia's investment follows $1.5bn into SB Energy, an OpenAI-linked Ohio data-centre project, announced earlier in the same week. The pattern is consistent: capital moving down the stack toward whatever is scarcest.

Why interconnection is the real asset

Accelerator supply is a manufacturing problem and manufacturing problems get solved. Interconnection queue position is an administrative and physical one — transformers, substations, transmission upgrades and utility approval on multi-year timelines. It is the constraint that no volume of GPU output relieves, which is why a company that owns queue positions is worth an equity cheque from the company that sells the chips those positions will eventually power.