IREN Limited

In four months IREN went from a $3.7bn ARR target that was partly aspirational to a >$4bn target with ~85% contracted, a customer list running from Microsoft and NVIDIA to Perplexity and Figure AI, ~$7.6bn of cash, and - the most underrated item - customers now prepaying ~45% of the GPU capex. Demand is explicitly described as exceeding available and planned capacity.

• The ARR target went UP and got MORE contracted at the same time: raised from $3.7bn to >$4bn with ~85% under contract as of Jul 20, 2026, on $2.8bn of new total contract value. Rising target plus rising contracted percentage is the opposite of the usual neocloud pattern of a flat target quietly losing coverage.
• Customer prepayments of ~45% of associated GPU capex on contracts signed since Jun 1, 2026 is the single most important economic change: it converts the biggest bear point (owned-GPU capital intensity funded with debt) into partly customer-funded capex and materially reduces net funding need. Management also states contracted pricing 'continues to strengthen.'
• Counterparty quality plus diversification, which is rare in this cohort: Microsoft (~$9.7bn contract) and NVIDIA (~$3.4bn) anchor the book, but the customer list now also includes Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI and Hume AI - so the thesis no longer rests on two names, and the weighted average contract term is ~4 years.
• NVIDIA is a partner, not just a supplier: a 5GW strategic partnership to deploy NVIDIA-aligned infrastructure across the pipeline, and NVIDIA took a 5-year right to purchase 30M IREN shares at $70 (~$2.1bn). A chip vendor taking equity exposure at that strike is a strong third-party validation of the platform, whatever it says about the current price.
• Funding is committed rather than hoped for: ~$7.6bn of cash at Jun 30, 2026 plus ~$3.6bn of GPU-level financing arranged May 29, 2026 (a $1.5bn Goldman/JPMorgan delayed-draw term loan at SOFR+2.25% and $2.1bn of 5.96% notes due Dec 2031, drawable in tranches to May 2027) at rates that imply lenders underwrote the Microsoft contract cash flows, not the equity story.
• Operating cash flow is already positive: +$289.326M for 9M FY26. Unlike most of the AI-buildout cohort the operating business funds itself; the cash burn is entirely discretionary capex.
• Power is the real scarce asset and IREN owns a lot of it: 5GW of secured power across North America, Europe and APAC, expanded by the Nostrum acquisition (490MW in Spain plus a GW+ development pipeline) and advancing Australian projects. The Mirantis acquisition adds the software layer for deploying and managing that compute.
• The owned-GPU model captures the full services margin rather than a landlord's rent: Q3 AI Cloud gross margin ex-D&A was 86.3%, versus 68.2% for mining - so as the mix inverts, unit economics improve structurally, not just in dollars.
• Execution cadence has been fast and specific, not vague: 3MW to 480MW of self-built AI Cloud capacity inside 12 months, with Horizon 1-4 on track for year-end delivery and the operational capacity fully contracted.

The last actual reported quarter missed revenue by 34% and produced a -$247.8M GAAP loss, and the entire thesis now rests on an ARR metric the company itself disclaims as not GAAP revenue. Getting from ~$580M of annualized GAAP revenue to a >$4bn December ARR is roughly a 7x step-up in six months, funded by a debt stack that has gone from ~$1bn to ~$7bn+ of commitments in twelve months - and the market has voted, taking the stock ~48% below its 52-week high.

• The gap between the metric and the money is enormous and explicitly disclaimed: ARR is defined as GPU/hour pricing for GPUs commissioned as of Dec 31, 2026 multiplied by 8,760 hours plus storage/ancillaries, is 'an operating metric, not a GAAP measure... not a substitute for revenue,' and depends on 'commissioning, testing and customer acceptance.' Q3 FY26 GAAP AI Cloud revenue was $33.6M - about 3% of a $4bn annualized rate.
• The last reported quarter was a bad miss, not a rounding error: revenue $144.8M vs ~$219.9M consensus (-34%) and EPS -$0.30 vs ~-$0.22 (-39%). Revenue was flat YoY and down 21.6% QoQ. Whatever the strategic narrative, the delivery-versus-expectation record at this point is poor.
• Owned GPUs mean owned depreciation and owned obsolescence, and it is already visible: $121.245M of D&A and a $140.411M impairment in a single quarter, taking 9M FY26 impairments to $188.423M. Adjusted EBITDA of 41% exists only because D&A, impairment and stock comp are excluded; on a GAAP basis operating margin was -161%.
• Leverage has exploded and is now structurally senior to the equity: $3.688bn of convertible notes at Mar 31, 2026 plus ~$3.6bn of new Hardware-3 secured GPU financing (with limited parent guarantees), where the maturity is tied to full payment of the Microsoft contract service fees. If Microsoft's tranches slip, the financing structure - not just the P&L - is exposed.
• There is a full quarter of information darkness at exactly the wrong moment: FY2026 ended Jun 30, 2026 and results were still unreleased as of Aug 4, 2026. The $7.6bn cash figure is explicitly 'unaudited preliminary,' the drawn portion of the new $3.6bn facility is undisclosed, and Q4 revenue, impairments and share count are all unknown.
• Bitcoin mining - still 77% of Q3 revenue - is being deliberately dismantled, so near-term revenue gets worse before it gets better. Mining revenue fell to $111.2M from $141.2M a year earlier and the decommissioning is what produced the impairments.
• The 9M FY26 headline net loss of only -$18.6M is an artifact, not performance: it was rescued by a $533.942M unrealized gain on financial instruments (capped call and prepaid forward transactions), against which sat $111.799M of debt-conversion inducement expense. None of it is operating cash.
• Dilution has been the primary funding tool and continues: shares went from 258.1M at Jun 30, 2025 to 341.0M at Mar 31, 2026 to 357.4M by Apr 30, 2026, plus $2.63bn of share issuance in 9M FY26, plus NVIDIA's right to 30M more shares, plus 9,099,328 RSUs granted to EACH of the two co-CEOs on Jun 30, 2026 (~18.2M shares, six-year vest).
• NVIDIA's $70 strike is now ~76% above the $39.75 share price, so the ~$2.1bn of potential proceeds that were part of the funding narrative are, at current levels, not coming.
• Concentration remains real despite the new names: the ~$9.7bn Microsoft contract plus the ~$3.4bn NVIDIA contract still dominate ARR under contract, and the $2.8bn of newly signed TCV is spread across AI developers and startups whose own funding durability is unproven over a ~4-year contract term.
• Supplier and single-site concentration: a single ~$1.6bn Dell purchase agreement (parent-guaranteed) supplies the GPUs for the $3.4bn contract, and the 2026 delivery program is concentrated at Childress, Texas - one site, one grid, one interconnection risk.