Thursday, August 27, 2026

IREN Thesis

SharkWater Trading  •  Equities Desk • IREN • AI Infrastructure

IREN's $4 Billion ARR Is Real. So Is the $27 Billion Bill Sitting Behind It.

August 27, 2026

Bottom Line Up Front

IREN Limited (NASDAQ: IREN) closed FY26 with $707.0m in revenue, a $702.6m net loss, and $638.8m of non-cash impairments as it scrapped its bitcoin fleet. It also closed the year with $4bn of contracted ARR against 2026 capacity and $1bn already operating as of August 26, 2026. The market cap into the print was roughly $14.5bn at $40.53.

The risk is not demand and it is not price per megawatt. The risk is that management guided FY27 capex of $25bn to $30bn against roughly $22bn of identified funding, only $14bn of which is actually committed. This is a financing trade wearing an AI costume.

What Actually Happened in FY26

IREN stopped being a bitcoin miner this year. Not gradually. Mining revenue fell from $111.2m in Q3 to $66.7m in Q4 as the company physically pulled hardware to clear floor space for GPUs. Management said on the call that mining will be effectively decommissioned by the end of December 2026. The $450.4m Q4 impairment is what that decision looks like on a balance sheet.

What replaced it is growing faster than the mining book shrank. AI Cloud Services revenue went from $16.4m in FY25 to $128.8m in FY26. In Q4 alone it hit $70.5m, or 51.4 percent of total revenue. That is the first quarter in IREN's history where the AI business was larger than the mining business.

The headline number missed. Q4 revenue of $137.2m came in under a roughly $157m consensus, and adjusted EBITDA collapsed to $19.2m from $59.5m the prior quarter. Shares closed up 2.4 percent at $40.53 and then dropped 7 to 8 percent after hours.

The FY26 Scoreboard

Metric FY26 FY25
Total revenue $707.0m $501.0m
AI Cloud Services revenue $128.8m $16.4m
Bitcoin mining revenue $578.2m $484.6m
Adjusted EBITDA $245.7m $269.7m
Net income / (loss) $(702.6)m $86.9m
Non-cash impairments $(638.8)m n/a
Cash and equivalents (incl. restricted) $7,619.5m n/a
Total debt $7,593.0m n/a

Source: IREN FY26 results release, August 27, 2026. Balance sheet figures as of June 30, 2026, and include $1,670.3m of restricted cash tied to Microsoft GPU financing. Consensus revenue estimate is a third-party compilation, not a company figure.

The Number That Runs the Thesis

Forget the loss. The impairment is the cost of a decision already made, and it does not consume cash. The number that decides this stock is FY27 capex.

Management guided $25bn to $30bn of capital spending for FY27. Against that, CFO Anthony Lewis laid out roughly $14bn of existing cash plus committed GPU financing and customer prepayments, a further $8bn targeted in additional GPU financing or prepayments, and then, in his words, the balance coming from data center financing, operating cash flows, and corporate sources.

Do the arithmetic the company did not put on a slide. Committed funding is $14bn. Midpoint capex is $27.5bn. The unfunded portion is $13.5bn, of which $8bn is a target rather than a signature and $5.5bn is a sentence. The entire enterprise value of the company is about $14.5bn.

IREN is a boat that has pre-sold four years of catch and financed the nets against the buyers' signatures. If every buyer shows up at the pier, the math is beautiful. If one does not, the nets still have to be paid for.

Where the Funding Stands

Source Amount Status
Cash on hand (June 30, 2026) $7.6bn Held, $1.7bn restricted
GPU financing, Microsoft deployments $3.6bn Committed, ~6.0% rate
GPU financing, other customers $2.8bn Committed, ~9.0% rate
Additional GPU financing / prepayments $8.0bn Targeted, not committed
Data center debt, op cash flow, corporate $3bn to $8bn Unspecified
FY27 capex plan $25bn to $30bn Guided

Source: IREN FY26 results release and Q4 FY26 earnings call, August 27, 2026. The first three rows sum to the ~$14bn management describes as committed. The $2.8bn tranche includes $2.4bn led by Blue Owl and PIMCO. The fourth and fifth rows are management's stated intent, not executed facilities, and the fifth is my arithmetic on the guided range rather than a company figure. Rate figures are as disclosed by the company and are approximate.

The Part the Bears Keep Underweighting

Pricing is going the right way, hard. Three to five year contracts are now signing in excess of $20m per megawatt of IT load, with active discussions around $25m per megawatt. That is roughly a 125 percent increase since November 2025. Customer prepayments are now covering 45 to 55 percent of GPU capital expenditure, which means the customer is funding the asset before IREN turns it on.

Apply the pricing to the pipeline and you see why the equity trades where it does. The 2027 target is 0.8GW of IT load. At $25m per IT megawatt, 800 megawatts is $20bn of annualized revenue. That is not a forecast and I am not presenting it as one. It is the arithmetic of the company's own two disclosed inputs, and it explains why a $14.5bn market cap does not look absurd to the people buying it.

The Capital Structure Is Already Crowded

IREN raised roughly $19bn across the last twelve months. Inside FY26 that included $2.1bn of share issuance, $3.0bn of convertible notes, and $938.0m of financing facilities. The May 2026 convertible carries a 1.00 percent coupon, matures in 2033, and was struck at a 32.5 percent conversion premium with a capped call at $110.30 against a $55.15 reference price.

Two things follow. First, at $40.53 the stock sits 27 percent below the reference price of its own convertible, which means the capped call the company paid $201.3m for is a long way out of the money and the dilution protection is not doing much work today. Second, gross debt of $7,593.0m against FY26 adjusted EBITDA of $245.7m is 31 times. That ratio only makes sense if you believe the ARR ramp, and the market is being asked to underwrite the ramp with the debt already on the books.

There is a smaller overhang on top of it. The August 4, 2026 filing tied to the Mirantis acquisition registered up to 11,981,668 ordinary shares for potential resale, roughly 3.4 percent of the 357.4m shares outstanding.

The Bull Case

  • The revenue is contracted, not projected. $4bn of ARR against 2026 capacity is signed paper with a weighted average term of about four years, spread across Microsoft, NVIDIA, Perplexity, Figure AI, Together AI, Fluidstack, and others. $1bn of it is already operating as of August 26, 2026.
  • Customers are pre-funding the assets. Prepayments now cover 45 to 55 percent of GPU capex. That is the single most important line in the release. It converts a balance sheet problem into a working capital problem.
  • Delivery is proven, not theoretical. Microsoft accepted Horizon 1, a 50MW liquid-cooled unit at Childress, in August 2026, and IREN achieved NVIDIA Exemplar Cloud status on GB300 NVL72. Three more Horizon phases are targeted for Q4 to reach 200MW.

The Bear Case

  • $13.5bn of FY27 capex has no signature on it. $8bn is targeted GPU financing and the remainder is described as data center financing, operating cash flow, and corporate sources. Corporate sources at a $14.5bn market cap means equity, and equity at these levels is expensive.
  • The operating business does not currently fund anything. Q4 adjusted EBITDA of $19.2m is down 68 percent sequentially. Mining cash flow is being switched off by December 2026 and the AI Cloud gross margin has not been disclosed. IREN is asking to be financed through the gap.
  • The positioning is a loaded spring in both directions. Short interest reached 26 percent of float as of August 16, 2026, with implied volatility near 109 percent on the earnings-week expiration. A stock this crowded does not reprice gently when the financing narrative changes.

The SharkWater Take

I am long this story and short this entry. The demand side is settled. Microsoft signed for $9.7bn, the pricing per megawatt has more than doubled in nine months, and customers are wiring money before the racks are energized. You do not get that combination in a business the market has correctly valued.

But I will not own the common outright into a fiscal year where planned capex is nearly twice the entire enterprise value and a third of it is unfunded. The bear case here is not that AI demand cracks. It is that credit markets have one bad month in the middle of a $27bn build and IREN has to sell equity at whatever the tape gives it. That is a gap risk, not a drift risk, and it does not respect your stop.

The tempting trade is selling puts into 100-plus implied volatility with 26 percent of the float short. I am not taking it. When a premium is that fat on a name that has already traded from $76.87 to $22.04 inside a year, the market is not mispricing the risk. It is describing it. Defined risk or no risk on this one.

Execution Notes

Premium basis: NO LIVE CHAIN. I could not pull an executable IREN options chain at time of writing. Every options reference below is structural. Do not treat any of it as a quote, and price your own fills.

Structure: If you want exposure, a defined-risk long call spread dated past the December 2026 mining decommissioning and the $4bn ARR commissioning deadline. That window is where the thesis either proves out or does not. Short-dated anything is a coin flip on a 109 IV name.

What I am not doing: Naked short puts. Cash-secured or otherwise. The tail here is a financing event, and financing events gap through strikes.

Confirmation to wait for: Executed facilities against the $8bn targeted tranche, and Horizons 2 through 4 accepted by Microsoft to complete the 200MW. Both are checkable. Neither requires you to guess.

Invalidation: A large equity raise priced at a discount, a slip in the December 31, 2026 date for $4bn of operational ARR, or any customer prepayment being renegotiated. Any one of those turns this from a build story into a funding story, and funding stories trade at a different multiple.

Tight lines. — SharkWater

Educational and informational purposes only. Not personalized investment advice. All figures sourced as noted and accurate as of publication. Options involve substantial risk of loss. The author may hold positions in securities discussed. Do your own work.

No comments:

Post a Comment