IREN announced $2.8 billion in new multi-year AI cloud contracts on July 20 and raised its year-end annualized revenue target from $3.7 billion to more than $4 billion. The customer list reads like a map of the AI economy: Microsoft, NVIDIA, Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI, and one undisclosed AI developer.
A company that mined bitcoin three years ago now has 85 percent of a $4 billion AI revenue target under contract, with customers prepaying roughly 45 percent of the GPU capex. The neocloud model is maturing from speculative buildout into contracted, prepaid infrastructure.
The deal mechanics
The $2.8 billion is total contract value across bare-metal and managed cloud agreements with a weighted average term of about four years. It stacks on top of IREN’s roughly $9.7 billion Microsoft agreement from November 2025 and a $3.4 billion NVIDIA deal from May. The revenue target is backed by 480 megawatts of gross AI cloud capacity planned by year end. The prepayment structure is the detail worth dwelling on: customers funding nearly half the GPU purchases up front shifts financing risk from IREN’s balance sheet onto the demand side.
The signal
Compute scarcity is doing the underwriting. When Together AI, itself a GPU cloud, shows up as a customer of another GPU cloud, and when hyperscalers like Microsoft rent from a former miner, the message is that nobody can build fast enough in-house, the same pressure behind Google rationing Meta’s Gemini access. The financing pattern matches what we mapped in the AI capital stack and in Nscale’s credit-line-not-a-raise: the buildout is moving from equity risk to contracted cash flows. Prepaid contracts are what that transition looks like at the neocloud layer.
AI and business tech news, verified by a physician who reads the filings. One email a week, no noise.
