IREN: almost all of the value sits in 2027 capacity that is not contracted yet, and the verdict is hold at $24.46

IREN fell about 11% on November 4 because Texas elected a governor who campaigned for an emergency freeze on new data centers. The more useful number is in the model below. Of my Base value of $23.14 a share, $21.74 comes from capacity IREN plans to switch on in 2027 and has not sold yet. The contracted fleet plus net cash is worth about $10 a share, which does not cover the cost of running the company.

At $24.46 on the November 3, 2026 close, IREN trades 5.7% above my Base value of $23.14. The Bear value is $3.20, 86.9% below the price, and the Bull value is $59.03, 141.3% above it. The range is that wide because the company plans to spend two and a half to three times its market value in one year. I assign no probabilities to the cases. The verdict is Hold, with new money waiting for $10.41 or less.

Valuation route: a former bitcoin miner now renting out GPU computing in its own data centers, with no stable free cash flow yet. I value each wave of capacity from its expected contract cash flows, building costs, customer prepayments, the second life of the GPUs and the remaining value of the buildings, then deduct head office costs and debt and add the new shares the plan needs. Values are present fair-value estimates, not 12-month price targets.

What IREN is

IREN, formerly Iris Energy, built data centers on cheap grid power to mine bitcoin and is now turning all of them into AI cloud capacity. It owns the land, the grid connections, the buildings and the GPUs, and rents the computing to customers on contracts of three to five years. Mining ends by the end of December 2026. The founders, Daniel and William Roberts, run the company as co-chief executives.

The anchor customer is Microsoft, with a $9.7 billion contract for four 50 megawatt liquid-cooled halls, called Horizons, at Childress, Texas. Horizon 1 was delivered in August and the other three are due in the December quarter. Other customers named on the August call include Together AI, Fireworks AI, Cohere, Perplexity, Figure AI and Prometheus, plus a frontier AI lab whose name is not disclosed yet. NVIDIA has a separate cloud contract worth about $700 million a year that ramps up in 2027.

IREN says it has more than 5 gigawatts of grid connections announced. About 2.75 gigawatts of that is in Texas, at Childress, which already runs, and Sweetwater, which still needs permission to switch on. The plan is roughly 300 megawatts of IT load delivered in 2026 and another half a gigawatt in 2027.

The latest results

Management said contracted yearly revenue, which it calls ARR, was about half a billion dollars at the end of June, $1 billion after Microsoft accepted Horizon 1, and should pass $4 billion by the end of December, all of it already under contract. Much of the December capacity comes on late in the quarter, so reported revenue catches up mainly in the March quarter. Cash overhead in the September quarter was guided up $40 to $50 million from June, and headcount, which nearly tripled in fiscal 2026, is expected to grow at a similar pace again.

Which numbers to trust

• Reported revenue describes the old business. Of $707 million in fiscal 2026, $578.2 million was bitcoin mining, which is being switched off. The figure that matters is contracted yearly revenue and when each hall starts earning it.
• Operating cash flow is flattered by prepayments. Customers paid $1,722 million in advance in the June quarter. That cash is real, but it is revenue IREN will later recognize without receiving cash, so I treat it as an advance, not as earnings.
• The losses are mostly accounting for now. The June quarter loss includes a $450.4 million impairment, and earlier quarters swung with the value of convertible notes. None of it says much about what a rented GPU earns.
• Share-based pay is a cost. It was $42.9 million in the June quarter. I count it with cash overhead at about $700 million a year in total.

Debt, cash and funding

On paper IREN has slightly more cash than debt: about $8.26 billion including restricted cash and my estimate of $0.45 billion raised by selling shares after August 14, against $7.93 billion of debt and leases at face value. That is a snapshot before the spending. For the year to June 2027 management guides $25 to $30 billion of capital spending, says about $14 billion is secured in cash, committed GPU financing and prepayments, expects about $8 billion of further GPU spending to be covered by GPU financing and prepayments, and expects the rest from data center financing, operating cash and corporate sources.

The funding model is clever. Customers prepay 45% to 55% of the GPU cost on recent deals and lenders finance about 90% of it against the contract, so the GPUs can be more than fully funded. The buildings are not, and the buildings are where the shareholders' money goes. The share count rose 49% from the September 2025 quarter's average, and I assume another $3 billion of new shares in the Base.

The Texas question

On August 3, 2026 the governor of Texas ordered the state utility commission and the grid operator, ERCOT, to audit every data center going through grid approval before more could move forward. ERCOT has paused approvals to switch on large new data center loads until that check is complete and expects to report by December 10, 2026. IREN's Sweetwater campus is in the first group under review, known as Batch Zero.

On November 3 Gina Hinojosa defeated Governor Greg Abbott. During the campaign she called for an emergency moratorium on new data centers until the legislature reconvenes. Childress already runs on its existing grid connection, so the Microsoft halls should not depend on the review. Sweetwater, where the first building is going up for 2027, does. I do not know how much of the 2027 half gigawatt sits there, because IREN does not split it by site, so the Base moves a fifth of it one year later and the Bear moves a third of it two years later.

Primary valuation: each wave of capacity on its own

I value three waves of capacity separately, because they are at very different stages. The 2026 fleet is contracted and mostly built. The 2027 build is funded in part and not contracted. The 2028 sites are land, power and plans. From the total I deduct head office costs at 7 times a year's cost, add net cash and add the new shares needed to fund the plan. The Base value is $23.14.

• Economics of a 2027 megawatt: $45 million to build, 73% of it GPUs. It rents for $20 million a year at an 85% site margin for four years, with half the GPU cost prepaid. Then the GPUs earn half the price for two more years, and the building keeps 76% of its cost for the next generation. At an 11% discount rate and 15% cash tax, that is worth about $18.9 million more than it costs. The whole valuation rests on this.
• Why the 2026 fleet is worth less per megawatt: the Microsoft contract pays about $9.7 million per megawatt a year over five years, half the rate of recent deals, and by my estimate about $9 billion of its cost, including the $1.83 billion of unpaid bills, is still to be spent. Contracts are with stronger customers, so I use 9.5%.
• Why 7 times head office: the cost supports the next waves too, so I do not deduct it forever at full weight, but it is not optional. Cash overhead was about $85 million in the June quarter by my arithmetic, and was guided $40 to $50 million higher for September.
• No terminal value: each wave ends with its GPUs retired and its buildings at a fraction of cost. Any value from waves after 2028 is outside the Base, and is part of the Bull.

The price a megawatt rents for and how much of the 2027 build switches on in time drive almost all of the answer:

The rest of the 500 megawatts arrives a year late, at $2 million less. Each $2 million of rent per megawatt moves the value by about $4.70 a share, while a year's delay on 100 megawatts costs about $1.30. Price matters more than timing.

Cross-check: what is under contract

The second view counts only what is signed. The contracted 2026 fleet is worth $9.22 a share and net cash $0.81, together $10.03. Head office costs, capitalized the same way, are $12.11. Without new capacity that earns more than its cost, the signed business does not carry the company, which is why I do not give a separate positive cross-check value.

A market multiple tells the same story from the other side. At $24.46 the company is valued at about $9.6 billion including net cash, or 2.4 times the more than $4 billion of yearly revenue contracted for December. That looks cheap for revenue with an 85% site margin, until you add the $25 to $30 billion IREN has to spend to grow it.

Valuation triangulation

• Primary (each wave valued on its own, Base): $23.14. This is the central value.
• Contracted business only: $10.03 a share before head office costs of $12.11, so no positive value on its own.
• Nearby assumptions (rent $2 million higher or lower, discount rates one point higher or lower, overhead $100 million higher or lower): $13.68 to $33.10.
• Price per $1 of contracted yearly revenue for December: about 2.4 times.

I keep the wave-by-wave model as the central value because it is the only one that counts both the spending and the contracts. The cross-check is the reason the margin of safety is the widest I use.

What is the market already pricing in?

The market pays for the 2027 build being rented at about the rate IREN says it signs today, with a fifth of it late. After the November 4 fall it allows for slightly less. That is not a stretched assumption in a market where IREN reports three-year pricing up about 125% since November 2025, but it leaves nothing for the risk that the halls are late, cheaper or never switched on.

Bear, base and bull scenarios

Each case also includes net cash of $0.33 billion, and the value per share counts the new shares.

• Bear: the contracted fleet earns $3.8 billion a year at an 82% margin after delays and customer losses. Rent falls to $16 million per megawatt for 300 megawatts in 2027 and $15 million for 150 megawatts that the Texas review holds back two years. Prepayments shrink to 35% to 40% of GPU cost, building costs rise to $47 million per megawatt, GPUs earn only 35% of the price in their second life and the buildings keep 60% of cost. Nothing comes from the 2028 sites. Head office is cut to $550 million, and $4 billion of new shares are sold at $15. Before those shares, debt and costs exceed the value of the fleet, so the new money takes most of what is left.
• Base: as described above.
• Bull: the contracted fleet reaches $4.2 billion a year, all 500 megawatts switch on in 2027 at $22 million, 500 megawatts from the 2028 sites rent at $20 million, prepayments reach 50% to 55%, and only $1.5 billion of new shares are needed, sold at $40.

Discount rates run from 9.5% for the contracted fleet to 13% for the 2028 sites in the Base, half a point to a point lower in the Bull and higher in the Bear. The stock's beta of about 4.3 would give an absurd cost of equity, so the risk sits in the cash flows and the funding instead.

How the buy price is set

Confidence is D, which carries a default margin of safety of 55%. I add nothing to it, because the Texas, pricing and funding risks already sit in the scenarios. That gives a maximum buy price of $10.41.

With rent $2 million higher or lower, discount rates one point higher or lower and overhead $100 million higher or lower, the same margin gives buy prices from $6.16 to $14.89. The price of $24.46, and the November 4 level near $21.70, are above all of them, so the decision does not depend on which is right. The 52-week low is $21.05 and the high $76.87.

IREN pays no dividend. Any return comes from the 2027 build being contracted and switched on, and from funding it without selling many shares at low prices.

Reasons to own IREN

• It owns the scarce part of the chain: grid connections, land and buildings, with more than 5 gigawatts announced against about 0.8 gigawatt of IT load planned by the end of 2027.
• Recent deals are very profitable on paper: above $20 million per megawatt a year on three-year terms, paying back the GPUs in about two years, with half of their cost prepaid.
• Horizon 1 was delivered to Microsoft and rated an NVIDIA Exemplar Cloud, which shows IREN can run the whole stack, not just the building.
• Funding has worked across credit quality: $3.6 billion at about 6% behind Microsoft and $2.4 billion at 9% fixed without an investment-grade customer.
• The November 3 close was 68% below the 52-week high of $76.87, and the price now asks only for the Base, not the Bull.

What could go wrong?

• Texas. ERCOT is not approving new large data center loads until its audit is done, and the incoming governor campaigned for a moratorium. Sweetwater, a big part of the 2027 and 2028 plan, needs that approval.
• Funding. The year to June 2027 needs $25 to $30 billion, and about $14 billion is secured. Selling shares at today's price to fill the gap costs existing owners more than it did at $70.
• Pricing. Rents for AI computing rose fast in 2026. If they fall, the GPUs earn less in their second life and new contracts sign at the low end of my grid, where $16 million gives a value near $14.
• Customers. Besides Microsoft and NVIDIA, many customers are young AI companies without investment-grade credit. A default leaves GPUs and loans behind.
• Dilution and claims. Shares are up 49% in a year, $6.75 billion of convertible notes sit ahead of shareholders, and the two oldest series met their conversion price trigger in June.

Management execution

• The pivot from mining to AI cloud happened quickly and on a large scale, and Horizon 1 arrived when promised.
• Contracting has broadened from one hyperscaler to a list of AI developers, with renewals from Together AI and Fireworks AI.
• Financing has been creative and cheap for the GPUs, but much of the growth so far was paid with new shares and convertible notes, including a $111.8 million inducement expense in fiscal 2026 when it bought back part of the older notes.
• Disclosure is detailed on contracts and pricing but thin on the split of the 2027 plan by site, which is exactly what the Texas question needs.
• Overhead is rising ahead of revenue by design, which is reasonable if the 2027 build fills and expensive if it does not.

On my ten-point management check (two points each), IREN scores 10 of 20: credit for delivery, contracting and funding access, little for dilution and per-share results.

Stock Analyza scorecard

The valuation score is 22.5 of 100.

The growth runway gets the most credit, because recent contracts are profitable and the power is in hand. Return on capital and the balance sheet get little, because neither is proven yet and the funding for the next twelve months is not complete. Business quality and valuation are shown separately and are not averaged.

Confidence score

The grade measures how well the model is identified, not the chance of a loss. Here the two happen to be related: the inputs I cannot pin down are the ones that decide the value.

What I would watch from here

The September quarter report on November 5 should show four things for the Base to hold: Horizons 2 to 4 still on track for the December quarter, the more than $4 billion of yearly revenue confirmed for December, new 2027 contracts at or above $20 million per megawatt, and financing for the roughly $8 billion of further GPU spending. After that, ERCOT's report, expected by December 10, is the next test for Sweetwater.

The three most important thesis breakers

1. The 2027 build is rented well below $20 million per megawatt, which takes the value toward the low teens.
2. Texas holds Sweetwater back for years, and IREN keeps paying for buildings and overhead that earn nothing.
3. The funding gap is closed with shares at low prices, so the growth arrives but each share owns much less of it.

Stress test and independent check

I recomputed every scenario with a separate script that writes out each year's cash flows and discounts them in a different way, and the results match to the cent. That checks the arithmetic, not the forecasts.

The weakest point is the 2027 rent, because $2 million per megawatt moves the value by about $4.70 and nothing for 2027 is signed. The strongest bull argument is that recent deals pay back the GPUs in two years with half of the cost prepaid, so every megawatt switched on at today's prices adds value. The strongest bear argument is that the signed business does not cover the company's costs, and the rest depends on a Texas approval, a rental market at a peak and capital markets that have taken the stock down 68% from its high.

Final verdict: hold at $24.46

IREN has built a real business: Microsoft, NVIDIA and a growing list of AI developers are paying more than $4 billion a year for capacity that is mostly built. At $24.46 you are not paying for that business. You are paying for the next one, at roughly today's prices, before a single 2027 contract is signed and while Texas decides whether Sweetwater can switch on.

Verdict: HOLD at $24.46, watchlist for new money. Base $23.14, 5.4% below the price, Bear $3.20, Bull $59.03, Confidence D (12.5/25). Required margin of safety 55%, maximum buy price $10.41, range $6.16 to $14.89 under nearby assumptions.

New money buys only at or below $10.41. For an existing holder the original cost is irrelevant: with the price slightly above the Base, the holding is a bet on the 2027 build, and I would size it as one. The next tests are the September quarter report on November 5 and ERCOT's report expected by December 10.

Sources

Primary sources are IREN's annual report on Form 10-K for fiscal 2026 (filed August 27, 2026), the fourth quarter fiscal 2026 earnings call of August 27, 2026, and the proxy statement filed October 13, 2026. The Texas review is described in the quarterly reports of Cipher Digital (filed November 2, 2026) and American Electric Power (filed October 28, 2026). All were accessed through Bigdata.com and are available on SEC EDGAR.

Quarterly financial statements, the share count and prices come from market data via Bigdata.com. The election result and the moratorium pledge come from news reports of November 4, 2026, including Benzinga, and the Batch Zero status of Sweetwater from MT Newswires (September 15, 2026). The reference market price is $24.46, the November 3, 2026 Nasdaq close.