DBP

Vertically integrated clouds compound ownership. Asset-lite neo-clouds compound lease exposure. Every day, the gap becomes more real.

$IREN Highlights Prince George construction is for LIQUID-COOLED data center. The new building we saw earlier, was for $NVDA GB300 NVL72 systems, which will host more than 4,500 GB300 GPUs. Back-up generators and UPS systems to be installed for all GPUs. (10.9K GPU's by December 2025). The number 10.9K is 2500 more GPUs than they currently have on order. Can anyone tell me where the additional 2500 GPUs are going to? Will this be GB300s for in Prince George? Correct me if I'm wrong, but I don't see any mention of an additional 2500 GPU purchase. Childress: Horizon 1 (50MW IT Load) liquid-cooled AI data center on track for Q4 2025. This building is mentioned to be able to host ">19,000 GB300s", does it mean it will be IRENcloud and not colocation? Would they mention the quantity of GPUs if it was for colocation? Horizon 2 (50MW IT Load) liquid-cooled AI data center site works and procurement underway. Great to hear them confirming Horizon 2. I have a very good idea about this project, and I think this will be completed in Q1 2026. Bookmark this if you think I'm making it up. $200-250m annualized revenue from AI Cloud at 10.9k NVIDIA GPUs by Dec 2025 The most important topic that we were really looking for. ARR guidance by end of year 2025. $200-250m is a very decent number. It's almost up 10x from the last $28m and as soon as the GPU financing flywheel starts to turn, I'm confident they will increase the cluster at Prince George. The new liquid cooled data center will probably not be part of the ARR guidance for December, so I think we will see a strong uptick in H1 2026, if this building will indeed be used for AI Cloud Services. The final highlight for me: Secured a second round of GPU financing for 100% of hardware purchase price, with further financing workstreams underway I am not sure if this is for the additional 2500 GPUs or for the recent purchase of 4200 B200s. Either way, it's good that this non-dilutive financing is becoming more and more used by $IREN. By the way, there is something about the NYDIG settlement in the filing: IREN agreed to pay a settlement amount to NYDIG in connection with the August 12, 2025 Settlement Agreement. The filing states that $18.2 million exceeds amounts previously accrued by the Company for these matters. That phrasing means the incremental impact on IREN’s financials is $18.2 million above what was already provisioned. The total cash settlement is therefore: - Previously accrued liabilities (undisclosed, but already on the books before June 30, 2025) - $18.2 million incremental (new expense recognized as part of the settlement) So the settlement cost to IREN is at least $18.2 million, but the all-in payment obligation could be higher depending on the size of the earlier accrual. Very bullish report. Will write my thoughts after the call and an actual night of sleep.

I wonder if $IREN is trying to leverage partial prepayments and GPU contributions from Microsoft to finance large-scale, bare-metal CSP buildouts using project-level debt. This would create a self-funding cycle where each contracted phase pays for the next, enabling rapid, non-dilutive expansion of data-center capacity. In this structure, IREN enters into a long-term, bare-metal CSP contract with $MSFT modeled on the $17 billion, five-year Nebius precedent. MSFT commits to roughly 200 MW (e.g., Horizons 1-4, with Horizon 1 already fully funded and nearing completion) of AI-optimized compute capacity at about $11.3 million per MW annually, representing a total contract value near $11.3 billion. Unlike a full prepayment, Microsoft would advance only about 25–35% upfront, with the remainder paid over five years as recurring service fees. The prepayment serves as quasi-equity, giving IREN liquidity to launch construction without issuing new shares. Project-level, non-recourse debt would provide funding for greater scaling (i.e., beyond the contracted 200 MW), secured by the Microsoft take-or-pay contract and the underlying physical assets. Under this arrangement, MSFT ships its GPUs to IREN’s facilities under a barter-style “compute-credit” structure (somewhat similar to the arrangement between $CRWV and $CORZ). IREN operates the GPUs within its own vertically integrated data centers, providing power, cooling, and operational management. In return, Microsoft receives discounted GPU-hour credits equal in value to its hardware contribution. Because IREN owns the infrastructure, it captures recurring, high-margin service revenue. With build costs around $8 million per MW, the initial 200 MW phase requires roughly $1.6 billion of total CapEx, fully covered by the prepayment and project debt. At $11.3 million of annual revenue per MW, approximately $4.3 million per MW flows to IREN as cash infrastructure fees, while the remainder represents the in-kind GPU offset. The unlevered payback period is about two years, and with project-level leverage the equity payback could compress to less than one year, yielding an internal rate of return well above 80 percent over the five-year term. Crucially, this model transforms prepayment and contracted cash flow into a repeatable financing flywheel: the first 200 MW phase self-funds its buildout and generates sufficient residual liquidity to greatly expand data center capacity without additional equity. Over time, each tranche can be refinanced or monetized, allowing IREN to scale several gigawatts of capacity non-dilutively. Strategically, the model positions IREN as a power-integrated, neutral AI-CSP partner to hyperscalers, rather than a mere colocation landlord. MSFT gains productive use of idle GPUs and secure low-cost compute capacity without new data-center capex. IREN gains recurring revenue, credit-enhanced project financing, and control of high-value infrastructure assets. The result is a self-reinforcing capital loop in which each contracted project finances the next, creating a non-dilutive, self-funding growth engine capable of expanding hundreds of megawatts per tranche while preserving shareholder value. PS: This theory is pretty close to my earlier discussions about an optimized colocation arrangement (e.g., colo 2o and 3o). Also, I'm just brainstorming, so the above could be totally off.