IREN: Five-Year DCF and Buildout Valuation

Thesis

A current preliminary cohort-cash DCF mapped to the live Google Drive workbook. The analysis distinguishes IREN's official capacity stages, cohort financing, five-year primary GPU life, non-core post-contract tail economics, residual support, revenue sharing, and public community-model reference outputs.

Key chart

The argument

The correct valuation question is not “How many megawatts does IREN control?” It is:

How much capacity is operating, contracted, financed, under construction, or merely optional—and what cash flow remains for common shareholders after growth capital, GPU replacement, debt, and dilution?

The model assigns the highest evidentiary weight to operating campuses, executed customer commitments, filed financing, and official aggregate buildout milestones. It assigns lower weight to later community-configured sites whose exact timing, GPU mix, contracts, and funding path remain uncertain.

InferenceA later site can gain value before revenue begins when a binding customer agreement, committed financing, equipment procurement, or construction milestone raises its completion probability. The reverse is equally important: a large power pipeline should not automatically receive the same value per MW as an operating contracted cohort.

Key findings

• IREN's 480 MW and 1.21 GW milestones have stronger official support than the later community-configured site schedule.
• The up-to-5 GW strategic framework is not treated as a fully contracted five-gigawatt backlog.
• The Microsoft contract supplies a strong contracted-revenue, GPU-capex, prepayment, and project-margin anchor for Horizon 1-4.
• The $3.65 billion GPU financing materially reduces immediate equity funding for the Microsoft cohort but creates senior claims and repayment obligations.
• Community-model outputs are reproduced for transparency, but they are treated as stabilized earnings-power references rather than present-value conclusions.
• The terminal calculation deducts a normalized five-year GPU replacement reserve net of residual value before applying the terminal multiple.
• A new GPU Capital Duration layer tests five-, six- and seven-year financing, non-core Years 6–8 tail revenue, NVIDIA support, revenue-share offsets and contribution per critical kW.
• Thirty-one core workbook checks and nine GPU-duration controls pass, confirming formula integrity and valuation invariance rather than forecast accuracy.

Counterarguments

A more optimistic interpretation is that IREN's power and land portfolio has strategic scarcity value that a conventional probability-weighted DCF understates. Large customers may reserve future capacity before detailed site economics are disclosed, and repeat financing could lower the incremental equity burden over time.

A more cautious interpretation is that the model still relies on aggressive GPU revenue density, high project margins, large capital programs, and terminal economics that are difficult to observe. Even when a customer contract is strong, construction delays, customer acceptance, debt amortization, technology turnover, and dilution can reduce common-equity value.

The workbook preserves both arguments by showing full-build values, probability-weighted values, a normalized replacement-reserve floor, and the separate community earnings-power reference.

Risks

Construction and capacity delivery

Company GuidanceIREN has disclosed official aggregate AI Cloud milestones of 480 MW for year-end 2026 and 1.21 GW during 2027. Those milestones do not prove that every modeled site cohort will be completed on the exact community schedule or reach full utilization immediately.

Customer concentration

The Microsoft contract provides a high-quality contracted anchor, but it also creates material customer concentration. Acceptance, service performance, remedies, contract renewal, and the treatment of equipment at the end of the term remain relevant.

Growth capital and leverage

The model deducts cohort growth capital when spent. Customer prepayments and GPU financing reduce corporate equity requirements, but the financing creates senior claims, interest expense, amortization, covenants, and refinancing risk.

GPU replacement, primary life and residual value

The Base case retains a five-year primary accounting and economic normalization with a 12.5% unsupported Year-5 residual. That is not a hard physical shutdown date. An older GPU may continue producing cash flow after Year 5, but it may command a lower rental rate, use power less efficiently than a new system, or be displaced because scarce critical power earns more with newer hardware.

The model therefore keeps explicit Years 6–8 tail cash flows outside the core DCF. A tail can replace part of Year-5 residual value; both cannot be counted in full.

Merchant GPU economics

Merchant utilization, hourly pricing, discounts, and margins are less directly observable than contracted economics. Public community inputs improve reproducibility but do not convert user-generated assumptions into company guidance.

Dilution and capital-structure reconciliation

The base model uses 385 million diluted shares versus an official starting share count of approximately 357.4 million. The denominator must continue to incorporate convertibles, equity awards, financing issuance, NVIDIA-related rights, and subsequent corporate actions.

Mining runoff

The base model reduces mining and legacy EBITDA from $250 million in 2026E to zero by 2030E. Bitcoin price, network difficulty, hardware efficiency, power prices, and the pace of AI conversion could produce materially different results.

Catalysts

Evidence that could justify higher completion probabilities or lower required returns includes:

• commissioning and customer acceptance for the 480 MW stage;
• contracted deployment and financing evidence for the full 1.21 GW stage;
• signed Sweetwater customer commitments;
• site-level capex and financing packages;
• GPU delivery and installation milestones;
• permits, interconnection, and construction notices for later projects;
• realized AI Cloud utilization, pricing, margin, and cash conversion;
• observable GPU residual transactions; and
• updated capital-structure disclosures.

Scenario analysis

The website model uses three published stage anchors rather than pretending that every project belongs in one current value.

The scenario weights are analytical inputs, not statistical guarantees. Readers should inspect the stage conditions rather than treating the weighted summary as a forecast of market price.

Buildout-stage valuation

The relatively small difference between the probability-weighted 5.0 GW and raw 5.6 GW estimates illustrates an important point: low-probability distant capacity contributes far less present value than its headline MW would suggest.

Valuation analysis

The primary method is a five-year cohort-cash DCF. It deducts modeled site and GPU growth capital in the years when it is spent. The terminal year then deducts a normalized five-year GPU replacement reserve net of the selected residual value.

The bridge is:

Present value of explicit cohort cash flow + present value of replacement-adjusted terminal value − adjusted net debt = equity value.

Equity value is divided by a fully diluted share count that includes modeled financing and other potential claims.

The model also shows a normalized-reserve floor that deducts annual GPU capital recovery throughout the explicit forecast. This is intentionally conservative and should not be confused with the central cohort-timing DCF.

Contract and financing anchors

Microsoft

The workbook uses this disclosure to reconcile approximately $1.94 billion of annual contract revenue and the contracted project-margin framework for Horizon 1-4. It does not assume that all later merchant or hyperscaler cohorts receive identical terms.

GPU financing

FactIREN announced a $3.65 billion investment-grade GPU financing facility tied to the Microsoft deployment at a disclosed 6.00% blended debt cost. Together with the approximately $1.94 billion customer prepayment, the funding bridge covers approximately 96% of the $5.81 billion Microsoft GPU capex at a disclosed 3.31% average financing cost.

The model keeps the zero-cost customer prepayment and 6.00% senior debt separate. The 3.31% blended figure is not used as the debt coupon, and later cohorts are not assumed to obtain identical terms.

NVIDIA managed-services cohort

No public disclosure reviewed for this update establishes a separate NVIDIA revenue share, residual-value guarantee, final GPU ownership allocation or end-of-term hardware right for this signed cohort. The public model therefore applies no additional NVIDIA revenue share or residual support to it.

GPU capital duration and post-contract tail

Longer financing can reduce annual capital service and lower the revenue required for a GPU deployment to remain viable. It does not improve physical performance, preserve the original GPU-hour price or automatically extend IREN's accounting depreciation life.

The lower annual burden can support a longer profitable tail, but longer amortization can increase total interest and leave debt outstanding after the initial contract expires.

The six- and seven-year rows hold project EBITDA constant and are intentionally aggressive sensitivities, not forecasts. The five-year / 12.5% row remains the core Base treatment.

NVIDIA residual support and revenue-share offset

NVIDIA's announced financing platforms are intended to mobilize more than $500 billion over time, with possible case-by-case support of up to 25%. Model AssumptionThe IREN model applies 0% IREN-specific support in the core, tests a 25% ceiling, and shows an 18.75% expected residual only in a 50%-eligibility sensitivity.

At 50% eligibility, the workbook calculates approximately $215 million of present-value uplift. Full eligibility produces approximately $430 million. Both are excluded from the published stage values.

A broad 10% revenue-share stress on the full $1.94 billion annual Horizon revenue would surrender approximately $194 million per year. That exceeds the $145 million annual capital-recovery benefit of raising residual value from 12.5% to 25% before utilization and financing benefits. Actual economics could be better if sharing applies only to the capacity NVIDIA backstops.

Power opportunity cost

An older GPU can remain cash-flow positive while still being economically obsolete. The relevant decision is whether continued use produces more risk-adjusted contribution per critical kW than a new system after upgrade capex and downtime. NVIDIA's Rubin performance claims are used only as a technology-cycle stress, not as realized IREN economics.

Official pipeline versus community configuration

The live workbook uses the public IREN Community model as a site-level operating engine and reconciliation tool. It does not treat the community configuration as official company guidance.

The official near-term milestones anchor the 480 MW and 1.21 GW stages. Later sites receive lower completion probabilities based on their evidence state:

• operating and contracted cohorts: 100%;
• Horizon 5-6: 85%;
• Childress Block 7-10: 80%;
• Sweetwater 1: 55%;
• Sweetwater 2 and Oklahoma: 35%;
• Nostrum: 25%; and
• Kiowa: 20%.

Bundey is tracked as an official development option but is excluded from the additive DCF to avoid double counting the up-to-5 GW framework.

Community-model reconciliation

The workbook reproduces the public community code's stage outputs exactly. Those outputs are materially higher than the cohort-cash DCF because they represent stabilized forward earnings power rather than discounted present value after timing, build capital, financing, dilution, and replacement capital.

The community model is valuable for transparency, site definitions, and sensitivity work. It is not substituted for the present-value model.

Assumption evidence map

The IREN workbook now includes a dedicated “16 Assumption Evidence” tab. It maps material inputs to their model locations, primary or community sources, evidence classifications, confidence levels, unresolved questions, and website source IDs.

The strongest evidence supports the official near-term capacity milestones, Microsoft contract economics, GPU financing, and filed capital-structure starting point. The most uncertain assumptions include later-site timing, merchant GPU rates, utilization, replacement economics, unit capex, execution probabilities, terminal multiple, and pro forma dilution.

Model checks

The workbook currently passes 31 of 31 core checks plus 9 of 9 GPU-duration controls, including:

• community and DCF capacity reconciliation;
• exact reproduction of the public community calculation engine;
• the Microsoft annual-revenue and GPU-capex anchors;
• the disclosed financing-coverage bridge;
• execution probabilities not exceeding 100%;
• no negative site revenue or modeled growth capex;
• full-build value not falling below its probability-weighted counterpart; and
• a positive diluted-share denominator;
• the five-year Base primary life and 12.5% Base residual remaining unchanged;
• no IREN-specific NVIDIA support entering the core DCF;
• all five published probability-weighted Base stage values remaining unchanged; and
• no full double counting of Year-5 residual value and explicit Years 6–8 tail cash flows.

Passing checks validates the workbook's mechanical integrity. It does not prove that the assumptions or future outcomes are correct.

Disconfirming evidence

The current model would weaken if:

• official capacity delivery slips materially;
• customer acceptance or utilization is below the modeled ramp;
• realized GPU pricing or margins fall below assumptions;
• later projects require materially higher capex or equity funding;
• GPU financing is unavailable on comparable terms for new cohorts;
• residual values or post-contract monetization are lower than modeled;
• easier industry financing expands GPU supply and compresses merchant rental rates;
• NVIDIA support or revenue sharing transfers more economics away from IREN than modeled;
• operating expenses or corporate overhead are higher;
• the diluted share count materially exceeds 385 million;
• adjusted net debt is understated; or
• the terminal value becomes an increasing share of the conclusion while near-term cash conversion deteriorates.

What would change the conclusion

The conclusion would become more constructive if IREN demonstrates repeatable conversion of contracted revenue into after-capex cash flow, secures later customer contracts with strong credit and prepayments, funds expansion without severe dilution, and reports residual or redeployment economics that support the replacement assumptions.

It would become more cautious if the 1.21 GW delivery path slips, merchant economics underperform, later site schedules are deferred, financing creates heavier common-equity claims, or GPU replacement consumes more cash than modeled.

Primary and supporting sources

The current IREN source archive and model-specific source map include:

• the March 31, 2026 Form 10-Q for financial statements, commitments, share count, and capital-structure context;
• the Q3 FY26 business update for the 480 MW and 1.21 GW buildout anchors;
• the Microsoft contract materials for contracted revenue, prepayment, GPU capex, Childress deployment, and project economics;
• the NVIDIA strategic-partnership disclosure for the official-capped strategic framework;
• the $3.65 billion GPU-financing announcement and filing;
• the five-year NVIDIA managed-services contract and Blackwell systems purchase;
• NVIDIA's institutional AI-factory financing-platform announcement;
• NVIDIA's Rubin platform announcement as a technology-cycle stress;
• the public IREN Community repository;
• the community site-definition file;
• the community GPU-default file; and
• The Golden Age of Neoclouds as secondary strategic interpretation.

Primary filings and executed obligations control over community and secondary research. Public code is used for reproducibility and cross-checking, not to upgrade unverified assumptions into facts.

Related spreadsheet model

Open the IREN Five-Year DCF and Buildout Model to inspect the current scenario estimates, buildout stages, assumptions, source links, interactive controls, sensitivities, limitations, model checks, and revision history.

Article revision summary

Version 1.2.0-preliminary adds the GPU Capital Duration framework from the revised live workbook. It preserves the five-year primary life, 12.5% Base residual and all published stage values while exposing cohort financing, Years 6–8 tail sensitivities, NVIDIA residual-support and revenue-share offsets, GPU ownership questions, contribution-per-kW replacement logic, 31 core checks and 9 GPU-duration controls.

Version 1.0.0-preliminary moved the public IREN model from the original demonstration scaffold to the live site-level Google Drive workbook basis. The revision added a dated archive copy, an in-workbook assumption-evidence tab, official-versus-community capacity classification, current stage estimates, source-backed contract and financing anchors, replacement-adjusted terminal methodology, and updated public limitations.