$IREN - The Bottle Neck-Line break (?)IREN LimitedBATS:IRENDillyDallyGallyneed some decisive volume for a breakout. The thesis is a contracted backlog colliding with a very full valuation. IREN has pivoted from Bitcoin mining to AI cloud with real hyperscale contracts — Microsoft $9.7b (with a ~$1.9b prepayment covering ~95% of GPU capex), NVIDIA $3.4b plus a 5GW strategic partnership — and is targeting $3.4b AI Cloud ARR by end-CY26 off 140k GPUs, scaling a 480MW→1,210MW→5GW power pipeline. That backlog is what makes it less speculative than a typical miner. But note the transition valley: total revenue is falling right now ($240m→$185m→$145m/qtr) because Bitcoin rigs are being decommissioned faster than GPUs bill — the AI ramp (early 2027) is ahead, not here. Valuation is the crux. At $66.6 the EV (~$28.8b incl. ~$5b net debt) is ~22x CY26 sales; even the base case (revenue 14×-ing to ~$17.8b by 2035) returns only ~7%/yr to ~$127, and bull (~$237, 14%/yr) still sits below the 15% cost-of-equity hurdle line ($269) — only the blue-sky case (~$504, 22%/yr) clears it. In other words, you are largely paying today for the buildout you hope happens. The reverse-DCF cuts the other way, though: at a flat EV the price implies only ~$7.6b of 2035 revenue — less than half the base path — so if you do believe the contracted ramp, the market isn’t fully crediting it. The tension between “22x near-term sales” and “under-crediting the out-year backlog” is the whole debate. Devil’s advocate: (1) the expensive entry multiple eats most of the base/bull return — multiple compression from 22x→3.8x offsets a 14× revenue ramp; (2) neocloud oversupply is the central risk — GPU rental rates could fall hard as CoreWeave/Nebius/hyperscalers add capacity, and the bear rail ($12, -16%/yr) prices exactly that; (3) massive capex + dilution — >$9b spend, $3.69b converts, $3.65b GPU facility, ATM up to $6b, NVIDIA’s 30m-share right at $70; base assumes ~36% more shares by 2035, bear ~65%; (4) Bitcoin price still swings near-term revenue and is being wound down at a loss (impairments); (5) execution — delivering 5GW of powered, GPU-filled data centers on schedule is the entire equity. Blue-sky “what has to be true”: the world stays structurally short compute, IREN fills its full 5GW+ at durable rates, GPU economics hold, and it re-rates as a vertically-integrated neocloud peer — then ~$504. A higher conviction in AI-infra demand (lower cost-of-equity, e.g. 12%) would pull the hurdle down and make bull clear it. Engine: Fair value = forward total revenue × EV/Sales − net debt, ÷ diluted shares, for four scenarios; log axis (speculative pre-cash-flow asset). Bitcoin mining revenue is being decommissioned/converted to GPU capacity, so TOTAL revenue dipped through the transition (Q1 FY26 $240.3m → Q2 $184.7m → Q3 $144.8m) even as AI Cloud revenue surged ($7.3 → $17.3 → $33.6m). 9mo FY26 revenue $569.8m. AI Cloud ARR >$500m end-Q1'26, targeting $3.4b by end-CY26 (140k GPUs); $3.1b ARR under contract. Contracts: Microsoft $9.7b (5-yr, ~$1.9b prepayment covering ~95% of GPU capex), NVIDIA $3.4b (5-yr air-cooled Blackwell, 60MW Childress, ramp early 2027) + 5GW NVIDIA strategic partnership (NVIDIA holds a 5-yr right to buy 30M shares at $70 ≈ $2.1b), plus Together AI/Fluidstack/Fireworks. Capacity: 480MW (2026, fully contracted) → 1,210MW (2027, Childress H5-6 + Sweetwater) → 5GW+ secured power (>4.5GW; +1.6GW Oklahoma, 800MW South Australia, 490MW Spain via Nostrum). Balance sheet (Mar 31 2026): cash $2.21b, convertible notes $3.69b, PP&E $4.37b, CIP $2.09b, GPU finance leases $274m; ATM up to $6b; just closed $3.65b IG GPU financing facility (June 1). Adj EBITDA ~$60–92m/qtr (~41% margin); GAAP losses driven by non-cash convert/capped-call swings + ASIC-to-GPU mining impairments. Spot June 3 2026 ~$66.60, ~357m shares, ~$23.8b mkt cap (+92% 12mo). Street PTs: Canaccord $79, B.Riley $96. Cost of equity 15% (speculative); IRR rows price-only. Bought Mirantis ($625m, 650 engineers) for AI-cloud orchestration. Probabilities are judgment. NOT investment advice.