Daave “3 and 30” Tepper
Institutional Endowment Investor
IREN is a stock to look for exit liquidity in: it is a scam that will be sold for parts, most likely to Nebius, and whether or not it ends in Chapter 7, Nebius will own its GPUs. Since that call was first made, IREN has fallen 20% while Nebius has risen 80%, and the price gap keeps widening. x.com
IREN might be exposed to a Microsoft capacity-delivery delay. Microsoft said some third-party GPU-as-a-service contracts have been delayed to the second half of FY27, while RHouse Research asked which neocloud committed to deliver capacity in Q4 2026. The provider is unidentified, but that possibility leaves IREN at risk. x.com
IREN's predatory management and markedly weaker execution than Nebius make it hard to take seriously while Mike Alfred remains on its board. Barring an oversold bounce, IREN looks cooked big time with more time. x.com x.com
IREN's $363 million in GPU-backed financing makes it a bet to survive neocloud consolidation. Elon Musk warned that drying debt markets could kill or consolidate many of the 200-plus neoclouds, leaving only a few survivors; IREN secured funding in that very market. x.com
IREN's lack of transparency on its biggest contract makes it hard to underwrite. It signed a multi-year deal with a leading frontier AI lab without naming the lab or the value, so the credit risk could be OpenAI's or a far weaker lab's. Nebius names and sizes its deals with Microsoft, Meta and Reflection. NDAs are usually the customer's demand, but the thesis rests on contracted revenue converting while IREN lost $684 million last quarter and funds its buildout partly through a $6 billion ATM. … Show more
IREN keeps falling behind Nebius: it misses estimates, its story keeps getting pushed back, and despite its capacity it fails to sign contracts or bring in meaningful revenue. Its management, software and engineering are all weaker, and it has not beaten Nebius on any front in recent years. @SpoonKidnapper counters that IREN is still converting its Canadian sites from bitcoin mining to AI, so ARR doesn't yet reflect where it will be in a quarter or two. x.com
IREN's investment case and its moat in power and land are unchanged by SemiAnalysis's criticism of its cloud offering. Nebius and CoreWeave clearly lead in technology, managed clusters and higher-stack services, but the Mirantis acquisition gives IREN a path to improve, and SemiAnalysis noted progress at its Sweetwater and Childress sites. NVIDIA Exemplar Cloud status and an NVIDIA-certified hypervisor show a more developed service, though adoption and profitability remain unproven. x.com x.com
IREN's risk is who pays for its contracts, not how much compute people want. Its August release shows Microsoft's contract financed at 6.0% investment grade, while $2.4bn for non-investment-grade customers at Mackenzie, led by Blue Owl and PIMCO, costs 9.0%. SoftBank's 10.7% drop, the steepest on the board, fits a funding story. IREN has $4bn of contracted 2026 ARR against $707m of fiscal-year revenue. If the next non-Microsoft financing prices below 9.0%, lenders are looking through the worry. … Show more
IREN is an extreme diluter of shareholders, so peers offer better equity exposure to the data center buildout. Its share count rose from 50M in 2022 to 350M through ATMs, repeated convertibles and 40M shares sold at $42; a new $6B ATM could dilute 35% or more, and it issued 30M warrant shares to help pay for Nvidia GPUs. Its new $3B convertible runs seven years with a $110 cap; if the stock doesn't get there, Cipher, Nebius, Riot or Marathon look better. Bondholders may fare better. x.com x.com
IREN's owned land, power infrastructure and data centers should retain value even if hyperscalers' own chips compress GPU-leasing margins. Enterprise AI-compute demand is enormous, while most enterprises will not build their own data centers, as Palo Alto Networks CEO Nikesh Arora noted. CoreWeave faces a bigger long-term risk because it relies heavily on third-party infrastructure. x.com
IREN and other neoclouds are valued on annualized run rate, a weak metric that wrongly treats infrastructure revenue like SaaS ARR. A two-year take-or-pay deal amortizes a finite GPU cluster rather than renewing like a subscription, and it hides customer concentration and utilization. It also ignores the capital, power, debt service and refresh capex behind the revenue; prepayments and GPU financing do not change that. IREN's stock compensation also rose to $205 million in FY26 from $43 million. … Show more
