IREN's Debt Reassurance Leaves the GPU Financing Question Open
IREN's co-CEO said its data centers carry no construction debt and customer prepayments cover much of recent GPU spending. The remaining question is whether lender-funded equipment and contracted demand can support the next wave of expansion.
IREN shares briefly fell 6% before recovering most of the decline after co-CEO Daniel Roberts addressed concerns about how the company funds its AI cloud expansion. His response clarified an important distinction: IREN says it built its data center sites with cash and carries no debt against those facilities, while customer prepayments and lender financing support much of the GPU fleet.
That structure is different from saying the business has no financing risk. Data centers, power infrastructure and graphics processors are separate layers of the operation. A site built without property-level debt can still depend on outside capital for expensive computing equipment, and those obligations must be serviced from customer payments and utilization.
Roberts said recent multiyear customer contracts include prepayments covering a large share of GPU costs, with lenders providing most of the balance. He also said IREN was largely sold out for 2026 and that demand for 2027 capacity was several times oversubscribed. Those statements support the demand case, but they do not disclose the full economics of the contracts, including financing costs, customer concentration, collateral terms or the returns IREN expects after operating expenses.
The debate extends beyond one company. IREN, TeraWulf and Applied Digital began as cryptocurrency miners before expanding into AI and high-performance computing infrastructure. IREN owns its sites and rents GPU capacity directly, while Applied Digital and TeraWulf rely more heavily on leasing data center capacity to tenants. Direct ownership offers more upside if utilization and pricing remain strong, but it also leaves IREN responsible for procuring, financing and refreshing the hardware.
Reuters reported that higher interest rates and a market selloff were testing the broader AI data center financing boom. The risk is not simply that AI demand disappears. Large buildouts can struggle if the cost of capital rises, customers delay deployments or newer chips make existing fleets less competitive before lenders and investors earn an adequate return.
The next evidence should come from execution rather than social-media reassurance. Investors need to see scheduled capacity delivered to contracted customers, prepayments converted into productive equipment, financing terms that preserve project returns and cash generation after interest. A sold-out year is encouraging only if the contracts are profitable and the capacity arrives on time.
IREN's explanation narrows the concern by separating cash-funded sites from financed GPUs. It does not eliminate the underlying question: whether the combined contract and financing structure can support rapid growth through a full technology and credit cycle.