Bernstein Cuts IREN Target to $100 as Miner Pivots to AI Cloud, Assigns Bitcoin Unit Zero Value

Bernstein dialed back its price target on IREN to $100 from $125 while maintaining an Outperform rating, framing the company less as a crypto miner and more as an emerging hyperscale AI cloud operator. The adjustment stems from a reduced Bitcoin mining footprint and higher share count from recent issuances—not a pullback in AI conviction. At $43.78, IREN fell over 9% amid a broader AI selloff tied to a report of OpenAI underperformance, though the stock remains up nearly 25% over the past month. The new target implies roughly 128% upside from current levels.

The fulcrum of the thesis is IREN’s rapid redeployment of capital and infrastructure into GPU cloud, anchored by a five-year Microsoft contract that locks in an estimated $1.94 billion in annualized revenue across 77,000 of IREN’s 150,000 GPUs. The remaining capacity is being sold on-demand, with about $400 million of customer contracts signed as of February. To scale supply, IREN entered a $5.8 billion purchase agreement with Dell for Nvidia GB300 processors and lined up $3.6 billion of GPU-backed financing at a sub-6% rate. Combined with Microsoft prepayments, Bernstein estimates roughly 95% of the capital for that anchor contract is already covered.

By 2030, Bernstein expects IREN to operate around 275,000 GPUs, up from 150,000 today, driving AI cloud revenue from $2.6 billion in 2027 to about $6 billion by decade-end. At scale, adjusted EBITDA margins are projected to hover near 82%, implying close to $5 billion of EBITDA. That level of profitability often requires not just hardware, but advantaged power and real estate—areas where IREN appears to be leaning in.

IREN controls roughly 4.5 gigawatts of power capacity across Texas, British Columbia, and Oklahoma. Bernstein values 3.6 gigawatts of undeveloped capacity in Sweetwater and Oklahoma at $3 million per megawatt, contributing approximately $10.8 billion to its sum-of-the-parts view. This power optionality is the quiet driver of the model: it lowers unit costs, accelerates deployment timelines, and creates a structural moat for high-density compute.

The controversial call is on Bitcoin mining: Bernstein assigns it no value and models mining revenue tapering to zero by fiscal 2030 as IREN swaps ASICs for GPU racks and repurposes facilities for AI workloads. Many miners have chased similar opportunities recently, with some already exiting crypto mining as AI economics look more compelling than post-halving proof-of-work returns.

A few numbers to anchor the setup: - Price target: cut to $100 from $125; rating: Outperform - Microsoft anchor: 77,000 GPUs; five-year term; ~$1.94B annualized revenue - On-demand cloud: $400M in signed contracts as of February - Hardware/financing: $5.8B Dell agreement for Nvidia GB300s; $3.6B GPU-backed financing at <6% - Scale: 275,000 GPUs by 2030; $2.6B revenue in 2027; $6B in 2030 - Margins: ~82% adjusted EBITDA at scale; ~\$5B EBITDA by 2030 - Power assets: 4.5 GW total; 3.6 GW undeveloped valued at ~$10.8B - Bitcoin mining: modeled to zero contribution by FY2030

What matters here is not simply “AI good, mining bad,” but the discipline of turning energy plus land into consistently utilized, premium-priced compute. The Microsoft deal solves utilization early, but it also concentrates counterparty risk and bakes in pricing that may or may not keep up with future GPU performance curves and cost of capital. The 82% margin assumption looks achievable at high utilization with long-dated, low-cost power; it becomes fragile if demand cycles, GPU lifespans shorten, or on-demand pricing compresses as more supply floods the market.

Valuing mining at zero is a statement of focus as much as a model choice. It strips out optionality that often tempts operators when BTC hash economics briefly improve. If IREN truly exits mining, execution risk narrows to building on time, filling capacity, and defending price. If AI unit economics tighten, some investors will wonder whether a reversible pathway back to hashrate would have been worth keeping. For now, the power platform plus contracted demand argues that the pivot can work—provided the company keeps utilization high, manages dilution, and avoids overreliance on any single buyer.