DCF Drop: IREN Limited. The compute infrastructure repricing nobody's underwriting correctly

A full 14-tab DCF on the stock that's up 866% in twelve months. Base case $55.61, bull $116.74, stress $39.05, blended $66.01.

There’s a particular kind of stock that punishes both bulls and bears at the same time. The one where the headline thesis is right, but the path to get there runs through a balance sheet that hasn’t been built yet.

IREN is that stock.

The pivot from Bitcoin miner to vertically integrated AI Cloud provider has been the most successful repricing in the sector by some distance. Twelve months ago, the stock was $7. Today it is $56. Market cap $20 billion. Two signed hyperscaler contracts worth over $13 billion in contracted revenue. An NVIDIA strategic partnership that grants the chipmaker the right to invest $2.1 billion in the equity at a $70 strike. And as of yesterday, an upsized $2.6 billion convertible priced at 1.00% coupon and a $73.07 conversion price.

The question for anyone looking at IREN today is no longer whether the pivot worked. It worked. The question is whether $56 reflects the structural reality of the business or the narrative momentum that delivered the rerating. And, more importantly, where the asymmetry sits if the answer is “approximately right at the central estimate.”

I built a 14-tab DCF and sum-of-the-parts model to answer that. This post walks through what the model says, where the assumptions are most defensible, and where they could be wrong. Paid subscribers get the full Excel workbook attached to this post. The scenario switch on the assumptions tab toggles bear, base, and bull. Every output cell is live. Every input is editable. Flex the assumptions you disagree with and watch the SOTP recompute.

That is what Mindful Returns exists to do. Not opinions about what a stock should be worth. Models you can argue with.

The business, briefly.

IREN was founded in 2018 in Sydney by Dan and Will Roberts. They tried to list on the ASX in 2021 and were turned away because Australian regulators did not know how to think about a Bitcoin miner. They listed on NASDAQ instead at the peak of the pandemic crypto bubble. The stock priced at $28 on November 17, 2021. It fell to $1.02 by December 28, 2022. It is now $56. The volatility is a feature of what the business is, not a bug.

The AI Cloud pivot landed in two announcements. Microsoft signed a $9.7 billion five-year contract on November 3, 2025 for 200 megawatts of IT load at the Childress, Texas campus using NVIDIA GB300 GPUs. Then six days ago, NVIDIA followed with a $3.4 billion managed cloud contract plus a 5-gigawatt strategic partnership and warrants for 30 million IREN shares at $70.

The Roberts brothers built renewable-power infrastructure first, then mined Bitcoin to monetise it, then positioned the same infrastructure as AI-ready data center capacity when the AI capex cycle arrived. That sequence is the entire story. The company that exists today is what happens when you build power infrastructure carefully for a decade and the market for that infrastructure arrives unexpectedly.

Why IREN is different from CleanSpark.

I cover both names. They look like siblings and they are not. The distinction matters more than most write-ups acknowledge.

IREN runs a neocloud business model. It buys NVIDIA GPUs through Dell, deploys them in its own data centers, and rents managed cloud capacity to customers. Revenue is GPU-hours rented. The competitive set is CoreWeave, Nebius, Lambda. Revenue per megawatt is higher than a landlord business. So is capex per megawatt, GPU depreciation risk, and operational complexity.

CleanSpark runs a triple-net landlord model. It leases power and shell building to a hyperscaler tenant who owns the GPUs and operates the workload. Revenue is lease payments. The tenant funds the GPUs. Lower revenue per megawatt, lower capex intensity, lower execution risk on any single contract.

Same starting point, two different bets on where in the AI value chain to position. IREN went up the stack. CleanSpark stayed at the property level. The market has paid IREN for the executed strategy and is making CleanSpark earn it on an unsigned deal. That asymmetry in how the market handicaps signed versus unsigned contracts is the behavioural finance frame at the centre of this piece.

The model and what it produces.

The workbook is fourteen tabs. AI Cloud revenue build, Bitcoin mining wind-down, cost structure, capex schedule, financing and capital structure, two DCFs (AI Cloud and Bitcoin mining), Bitcoin treasury, pipeline optionality, sum-of-the-parts consolidation, comparables, sensitivity, plus a master assumptions tab that drives everything through a scenario switch and a publication-grade summary tab.

The model produces four scenarios. Two downsides on purpose, because they answer different questions.

Stress case at $39.05. A 31 percent drawdown. Hyperscaler digestion through 2027, multiple compression toward 18x, contract execution slippage of 8 percent, and the funding gap requirement creating equity dilution. This is a capital-markets event scenario. The base assumption that breaks here is access to funding at reasonable terms. I weight this at 15 percent.

Consensus bear at $46. This is JPMorgan’s Underweight target. Goldman Neutral is $44. Both reflect concerns about the circular economics of the NVIDIA deal, the convertible debt stack, and the undefined GPU allocation in the partnership. This is a fundamentals slip scenario, not a financing scenario. The base assumption that breaks here is GPU pricing and utilisation. I weight this at 25 percent.

Separating these matters. My stress case is a capital markets event. The consensus bear is a fundamentals miss. They have different catalysts. They require different hedges. Anyone publishing a single bear case is collapsing two different downside paths into one number.

Base case at $55.61. Essentially current price. Implied return is minus 1.7 percent. The base case assumes 89 percent utilisation, a 7 percent execution risk haircut on contracted ARR, 10.25 percent WACC reflecting hyperscaler counterparty credit quality, 24x exit EV/EBITDA, and 25 percent probability on the 5GW NVIDIA partnership converting to firm contracts. By FY30 the model has IREN generating $3.2 billion of AI Cloud revenue at a 76 percent EBITDA margin across 560 megawatts of deployed capacity, on cumulative capex of $20.3 billion from FY26 through FY30. I weight base at 35 percent.

Bull case at $116.74. A 107 percent gain from current. The 5GW NVIDIA pipeline converts at 35 percent probability rather than 25. Utilisation runs at 92. Exit multiple expands to 25x, in line with where CoreWeave actually trades. NVIDIA exercises the $70 warrants, which adds $2.1 billion to the balance sheet and 30 million shares of dilution. The cash add more than offsets the dilution drag. I weight bull at 25 percent.

Probability-weighted at 15/25/35/25 across stress / consensus bear / base / bull, the blended target is $66.01. Plus 17 percent. Mildly bullish.

The single most important variable in the entire model.

The funding gap. IREN’s capex schedule through FY31 totals approximately $23 billion. Microsoft’s $1.94 billion prepayment, existing convertibles, the new $2.6 billion convertible priced yesterday, and operating cash flow cover most of it. The residual gap is roughly $6.9 billion. That has to come from future debt or equity issuance over the next five years.

The model applies a 40 percent NPV haircut to the residual gap, scaled by the scenario-active capex multiplier. The mechanism is simple. If management scales capex back aggressively in a stress scenario, the gap shrinks. If management presses ahead in a bull scenario, the gap stays large. In every scenario, the haircut on the remaining gap reflects expected dilution from future raises.

This single line item is worth roughly $6 per share against the base case. It is the largest negative line in the SOTP bridge after net debt.

If you think IREN can fund the gap as efficiently as yesterday’s 1.00 percent coupon convertible at a $73.07 conversion price suggests, the base case lifts toward $63. If you think capital markets close, the stress case widens.

If the model is wrong somewhere, this is the most likely place. Not on the AI demand curve, not on the NVIDIA partnership, not on margins. On the cost of capital required to finish the build.

Dilution has been the silent compounder.

IREN’s basic share count has grown 72 percent in the last twelve months. Adding the new $2.6 billion convertible at a $73.07 conversion price, plus the existing converts, plus the NVIDIA warrants in the bull case, fully diluted shares now exceed 397 million in the base and bear scenarios, and 427 million in the bull.

The previous version of the model used 290 million. Correcting that share count dragged every scenario price target lower by roughly 27 percent. The equity value held up. The per-share figure didn’t.

This is the part of the IREN story that has not been priced cleanly. The headline contracts produced the rerating from $7 to $76. The dilution required to fund the contracts has been steady throughout. The market has been so focused on the numerator that it has under-priced the denominator.

The May 12 capital raise added another 35.58 million shares of potential dilution at $73.07. Useful for clearing a 90-day overhang. Not sufficient for the FY27 through FY29 build.

Comparison to CoreWeave.

CoreWeave is the cleanest public comparable. Market cap around $62 billion. Revenue backlog of $99.4 billion. Enterprise value to forward revenue around 12 times. Forward EV/EBITDA between 20 and 31 times depending on which adjustment you accept. Microsoft is 67 percent of CoreWeave’s revenue. Customer concentration is structural in this sector, not unique to IREN.

IREN sits at roughly one-third of CoreWeave’s market cap with one-eighth of the contracted backlog. The ratio is approximately proportionate to the execution timeline. The dilution path is steeper for IREN because the company is at an earlier stage of the build. Both stocks earn or lose their multiples on the quality and pacing of the next contract.

The multiple itself is the bear catalyst. IREN currently trades at 22 times forward EBITDA on the base. Compression to 18 times would take the stock toward the stress case. But the CoreWeave floor in soft conditions has been around 18 to 20 times, which argues against the multiple going materially lower from here. So the question becomes: what moves the base up, not the multiple down.

Where the mispricing actually lives.

It is the pipeline conversion math.

The market is implicitly pricing the unsigned pipeline at roughly 5 percent probability. My base case carries it at 25 percent. The NVIDIA 5GW LOI is the centrepiece, and the $2.1 billion warrant structure gives a structural floor for partnership conversion even in a hyperscaler digestion scenario. NVIDIA has financial incentive to drive at least partial deployment regardless of broader AI demand conditions.

Each contracted MW that comes off the LOI and onto the books reduces my execution haircut and lifts the base case. A signed deal for 1GW of additional capacity at neocloud terms would lift the base case by roughly $8 per share. A 2GW signing would lift it by $15 to $18.

That is the catalyst worth watching. Not earnings prints, not Bitcoin price, not Fed rate cuts. Signed MW.

The behavioural read.

The market has rewarded IREN for executing on the announcement. The pre-announcement state of the equity, which had the same power infrastructure and the same operational team as the post-announcement state, was priced as if the contracts would never come.

This is narrative dominance. Once a contract is signed, the market treats it as certain. Before the contract is signed, the market treats it as speculative. The intermediate state, where the structural conditions for a contract exist but the paper has not yet been signed, gets priced closer to speculative than to certain.

That mispricing is the analytical opportunity. The IREN trade today is no longer cheap. The base case at $55.61 implies current price is approximately right. The asymmetric upside is real but lives in the bull case, which requires the 5GW NVIDIA pipeline to convert at higher probability than I am modelling and the warrants to come into the money.

The same mispricing exists today in CleanSpark at $13. The structural conditions are in place. The signed contract is not. The market is pricing CleanSpark as if no contract is coming, which is the same mistake the market made on IREN in mid-2025. I covered that thesis on Monday in the Thesis Update post.

The catalyst path on IREN.

Three things would move me from watchlist to position.

A senior secured raise that closes the funding gap meaningfully at non-dilutive terms. The cleanest version of this would be a $4 to $5 billion investment grade DDTL similar to the $8.5 billion CoreWeave printed last quarter. That would compress the SOTP funding gap haircut from 40 percent toward 25 percent and lift the base case toward $63.

A second hyperscaler contract beyond Microsoft and NVIDIA. The conversation has been about Amazon, Google, and Meta. Any of those landing at scale would justify lifting the 5GW pipeline conversion probability in the base case from 25 percent toward 35 to 40 percent and add another $10 per share to the base.

A customer prepayment that does the same work as a debt raise. IREN received $1.94 billion from Microsoft as a 20 percent prepayment. A similar structure from a second customer would directly reduce the residual funding gap without diluting the equity.

Until at least one of those lands, the base case is the base case. Fair value at $56 is the model’s verdict. The position size question matters more than the directional question.

The trade is binary on a single variable.

How the next $6.9 billion gets funded.

Watch that, and the rest of the model takes care of itself. The AI demand is real. The contracts are signed. The infrastructure is being built. The only open question is the price of capital required to complete the build, and the second-order question of how much of the equity gets diluted to fund it.

I sit on the sidelines until the funding path clears. Either a senior secured raise that closes the gap, or a customer prepayment that does the same. The May 12 convertible resolved the near-term anxiety. It did not solve the FY27 through FY29 capital requirement.