Key points
- JPMorgan went straight from underweight to overweight
- Price target lifted to $65 from $46
- IREN co-CEO answered the slowdown call at 6:12 a.m.
- IREN fell least of the comparable AI names
IREN Ltd co-chief executive Daniel Roberts posted at 6:12 a.m. New York time Monday: "Even if models never improved from here, just rolling out what they can already do would take more compute than the world can build for years."
He was responding to the weekend calls to slow frontier-AI development. About three hours later, JPMorgan Chase double-upgraded IREN.
The broader AI complex still sold off after three lab chief executives called for a slowdown. Every adjacent compute and data-center name we checked fell. IREN (IREN) was down about 2% by late morning, making it the smallest decline in the group by a wide margin.
The upgrade came from analyst Richard Choe, who moved IREN two notches, to overweight from underweight, and raised his price target to $65 from $46.
That target is about 48% above Friday's $43.83 close. Choe's argument is that IREN is becoming a top-tier neocloud provider through its Nvidia (NVDA) partnership. He also pointed to higher pricing: AI compute that previously sold for $10 to $15 per watt is now fetching $15 to $20 or more, depending on delivery timing, GPU generation and contract length. Fourteen of the 17 analysts covering IREN rate it buy or strong buy.
The Nvidia agreement is the spring deal whose value has been reported in several ways. IREN's own May 7 announcement put it at $3.4 billion over five years for managed GPU cloud services supporting Nvidia's internal AI and research work. The contract covers air-cooled Blackwell systems using about 60MW at IREN's Childress, Texas campus, with cluster-management software from Mirantis, which IREN later acquired.
What Roberts actually argued
His post ran long, and not all of it is settled fact. On demand, he argued the slowdown debate is about future models rather than current usage, and that existing demand is what people misread. He cited Anthropic chief executive Dario Amodei telling a developer conference in May that the company had planned for 10 times growth and saw 80 times in the first quarter on an annualized basis, which Amodei said was the reason it had struggled with compute. He cited OpenAI president Greg Brockman saying in July that the company will be short of compute no matter what and is choosing which products to scale, and Google saying it processes seven times the tokens it did a year ago.
On supply, he put the constraint in memory. Three companies make the high-bandwidth memory that sits inside every major AI accelerator, all three are sold out this year, and a new plant takes years. He cited TrendForce for HBM shipments growing 50% to 60% next year, and Nvidia's own outlook for about 70% revenue growth. That last figure needs pinning down: it is chief financial officer Colette Kress's fiscal 2028 guidance from the August 26 earnings call, not a fresh number, and Nvidia described the 70% outlook as supply-constrained, not as a forecast of unconstrained demand. Roberts closed with the trade in one line. "The risk to demand continues to seem heavily weighted to the upside. The risk to supply continues to seem weighted toward less capacity getting built, not more."
He has an obvious interest in that conclusion. He also runs one of the companies that has to deliver the capacity, which is a different vantage point from a lab executive arguing about model releases, and his own numbers are the check on him: IREN's contracted revenue and delivery record are public, and we went through them when it reported full-year results in August.
What the tape said
Every figure below is against Friday's close, as of 10:51 a.m. ET.
| Company | Monday |
|---|---|
| IREN Ltd (IREN) | -2.3% |
| Nvidia (NVDA) | -3.6% |
| Core Scientific (CORZ) | -4.6% |
| MARA Holdings (MARA) | -4.7% |
| Nebius Group (NBIS) | -4.9% |
| Cipher Mining (CIFR) | -5.9% |
| Applied Digital (APLD) | -6.0% |
| CoreWeave (CRWV) | -6.4% |
| TeraWulf (WULF) | -7.0% |
A single upgrade absorbing a sector drawdown is not a verdict on who is right about the buildout. It says the desk that moved IREN two notches did it on contracted revenue and delivered megawatts, which is the part of the AI trade that a slowdown in model releases does not obviously touch. We looked at the same split across the neocloud names last month.
What is not settled is most of it. Roberts is making a forecast, not reporting a result, and the claims he stacks up are other people's. JPMorgan's $65 assumes IREN keeps signing customers at the higher end of that per-watt range and delivers the capacity on time, which is the thing the whole sector keeps missing. And one upgrade on one red morning is a day of trading, not a trend.
