IREN Drops Over 20% in a Single Week; CEO Reassures: 96% of GPU Costs Covered by Prepayments and Loans
nashnova research
AI cloud-compute firm IREN fell more than 20% in a single week, prompting CEO Daniel Roberts to publicly address financing concerns — the headline number is 96% of GPU costs locked in via prepayments and loans, but the market's real worry is whether those contracts can deliver on schedule.
Down 20% in a week — what is the market afraid of?
IREN's stock dropped over 20% in one week. The immediate trigger: rising debt pressure at peer CoreWeave, plus a research firm warning that by 2030, more than 80% of new cloud providers may fail to operate independently.
This means → the sell-off is not just about IREN — the market is re-examining whether the entire "new cloud compute" business model can work.
CEO Daniel Roberts posted a point-by-point rebuttal on X, attempting to rebuild confidence with contract-level data.
"Zero debt on data centers" — where did the money come from?
Roberts stressed that IREN's data centers carry no debt today; construction costs have been paid entirely in cash.
In plain terms = build the house with your own money first, then once a paying tenant moves in, let the house borrow on its own to expand.
This reflects a "build first, finance later" strategy — the upside is no early interest burden; the risk is rapid cash burn if clients are late to come online.
Where does the 96% figure come from?
The contract mechanism centers on "take-or-pay" — clients sign multi-year deals and must pay whether or not they actually use the compute.
In recent three-year contracts, prepayments cover 45%–55% of GPU costs; the remainder is funded by lenders, with the GPUs themselves as collateral.
Example: IREN's Microsoft contract is backed by an A-rated loan at roughly 6% interest. Loan plus prepayment together cover 96% of GPU spending.
This means → IREN's own out-of-pocket share is roughly 4% of GPU costs — provided both client and lender perform as agreed.
Could rising rates kill the economics?
At the McKenzie site in British Columbia, a non-investment-grade client borrowed at a fixed 9% rate; loan plus prepayment exceeded total GPU cost.
Roberts acknowledged that future debt will be priced at market rates, but offered a key number: recent contracts pay back in roughly two years.
In plain terms = a business that recoups costs in two years can absorb a percentage point or two of higher interest without breaking — that is the CEO's core argument.
Is demand really outstripping supply?
2026 capacity is essentially sold out. Recent three-year contracts are priced at more than double last year's levels.
2027 capacity is oversubscribed by several times; 2028 negotiations have already begun.
This means → from a bookings standpoint, demand is not the problem — the real bottleneck is in the next section.
"Dormant capacity" — what is the biggest risk behind those contract numbers?
Roberts himself pointed out that a large share of the industry's announced capacity sits in a "dormant" state — contracts signed, but GPUs not yet online.
This reflects a core contradiction: the 96% coverage ratio depends on contracts being fulfilled on time. If "dormant capacity" widens, prepayment receipts and loan disbursements could both be delayed.
Put simply = the CEO's math checks out, but the premise is "hardware delivered on time, cash received on time." The market is selling precisely because it doubts that premise.
市场有风险，内容仅供研究参考，不构成投资建议。