IREN’s (NASDAQ: IREN) latest results marked a turning point in its transformation from bitcoin miner to AI cloud operator. They also left some shareholders uneasy. After the company outlined fiscal 2027 capital spending of $25 billion to $30 billion, Bernstein argued that the size of that program, rather than the underlying economics, is what unsettled the market.
The company itself supplied the figure. On its fiscal 2026 results call, management said the coming year’s investment would cover remaining 2026 deployments tied to already contracted capacity, air-cooled GPU rollouts scheduled through calendar 2027, and early work on liquid-cooled facilities at Childress and Sweetwater slated for the second half of that year.
Actual outlays, executives cautioned, will still depend on construction timing, GPU delivery, contract wins and financing conditions.
Bernstein’s Gautam Chhugani and colleagues told clients that the headline number may have “spooked the market.”
Their counterpoint is that payback on incremental GPU spending has improved.
The firm now estimates roughly two years to recoup that capital, versus about three years under IREN’s 2025 Microsoft arrangement.
In their view, investors are focusing on the cash going out and underweighting the faster return on each new dollar of compute.
The operating picture behind that argument has shifted quickly.
IREN reported $4 billion of contracted annualized run-rate revenue against 2026 capacity, with $1 billion already live after Microsoft accepted the first Horizon hall.
Management described 2026 capacity as largely sold out and said late-stage talks cover a substantial slice of 2027 megawatts, with 2028 conversations underway.
Recent three-year deals have cleared $20 million of revenue per IT megawatt; active discussions are nearer $25 million.Funding is the other half of the debate.
Over the prior twelve months IREN assembled about $19 billion through customer prepayments, GPU-backed loans, convertibles and equity.
Roughly $14 billion of that sits in cash or undrawn commitments.
Management is aiming for another $8 billion of GPU financing and prepayments, with data-center debt, operating cash flow and other corporate sources meant to close the rest of the $25–30 billion range.
Recent prepayments have covered 45–55 percent of associated GPU cost; some non-investment-grade facilities fund about 90 percent of the chips.
That structure is designed to limit, but not eliminate, equity risk.
The share count has already risen, and any shortfall in prepayments or credit markets would reopen the dilution question.
Execution risk remains as well: Horizons 2–4 are targeted for the December quarter, mining hardware is being written off as sites convert, and the company posted a large fiscal 2026 net loss driven by those impairments even as AI cloud revenue multiplied.
Bernstein’s claim is not that the spending is small. It is that unit economics are improving fast enough that the market’s first reaction overstates the danger. Whether that proves right will turn on signed 2027 contracts, on-time hall deliveries and whether IREN can keep financing GPUs largely off the common-equity line.
