Iren CEO, NVIDIA Partner: AI Computing Supply Struggles to Keep Up with Demand
nashnova research
Iren co-CEO Daniel Roberts says global compute supply "can barely keep pace with demand." The company raised roughly $19 billion in the past year to build data centers — a heavy-asset bet whose payoff hinges on whether the compute shortage persists.
Why does he say compute will "never be enough"?
Roberts's core argument: every new unit of AI compute tends to generate several times more additional demand, rather than absorbing the existing shortfall.
This means → the compute market doesn't follow traditional infrastructure logic of "build it and it saturates." It runs on a supply-creates-demand feedback loop.
Two forces keep pushing consumption higher: the proliferation of AI agents — autonomous programs that execute tasks — and rising inference speeds. Both compound demand continuously.
What's stopping data centers from expanding without limit?
Three hard-world constraints — power, land, and political resistance — are capping how fast data centers can be built.
Goldman Sachs projects U.S. data-center capacity will double by end-2027 and reach roughly 125 gigawatts by 2030 — more electricity than 100 cities the size of San Francisco.
In plain terms = it's not that nobody wants to build. It's that there isn't enough power, enough land, or enough local consent — and that bottleneck is precisely what prevents an oversupply glut.
Who is Iren? How did it go from near-bankruptcy to a $17 billion valuation?
Iren, formerly Iris Energy, was founded in Sydney in 2018 as a Bitcoin miner. It once faced loan defaults on mining rigs and a going-concern warning from its auditor.
Over the past three years it pivoted into a core Nvidia "neocloud" partner — a new breed of cloud provider purpose-built for AI workloads. Its clients now include Microsoft, Perplexity, and Nvidia itself.
Market cap rose from roughly $60 million at the crypto low in late 2022 to about $17 billion; the stock climbed from around $1 to a high of $76. This means → a nearly 280× increase in just over two years, driven almost entirely by one strategic pivot: mining to compute.
Where did the $19 billion in funding go?
Over the past 12 months, Iren raised approximately $19 billion through convertible notes — debt that can convert into equity — chip-collateralized borrowing, customer prepayments, and share issuance.
Most of the capital went to purchasing Nvidia processors and building data centers. This summer it closed a $3.65 billion chip-backed facility at 6%, tied to a Microsoft procurement agreement.
Last week it secured $2.4 billion from Blue Owl and Pimco-affiliated funds at 9%. This means → the rate jumped from 6% to 9% because this tranche's end-customers (Cohere, Perplexity) carry sub-investment-grade credit compared to Microsoft — higher risk, higher cost of capital.
Why is Nvidia locking itself into Iren?
In May, Nvidia signed a five-year, $3.4 billion compute-lease deal with Iren, plus the right to buy up to $2.1 billion in Iren stock at $70 per share.
That equity option unlocks progressively as Iren takes delivery of up to 600,000 Nvidia chips. In plain terms = the more chips Iren buys, the more Iren stock Nvidia can acquire — their interests are tied to the same rope.
Roberts calls the structure "extremely powerful": the core supplier is no longer just a vendor but a shareholder, incentivized to help Iren scale its GPU fleet as large as possible.
What is the central risk of this bet?
Iren plans to deploy up to $30 billion in AI infrastructure over the next year — a textbook heavy-asset wager.
Roberts himself acknowledges: whether the compute supply-demand gap persists long-term is the key variable that determines if this spending pays off.
This reflects a fundamental tension in the AI infrastructure sector: everyone is betting demand will keep exploding, but if supply catches up — or demand growth slows — heavy-asset models will be the first to feel the pressure.
市场有风险，内容仅供研究参考，不构成投资建议。