IREN: 85% of a $4B ARR target is contracted, but a 63.2x P/E and three quarters of falling Adj. EBITDA leave no room for slip

IREN Ltd (IREN $40.58 -1.70%) is the company the market has started pricing as an AI infrastructure platform rather than a Bitcoin miner, and the most recent developments explain why the conversation has changed. On the Q3 FY26 earnings call, co-CEO Daniel Roberts framed the supply picture bluntly: "We are not chasing demand. We are racing to build supply fast enough to meet it." Two months later, the 8-K filed 2026-07-20 raised the year-end AI Cloud annualized run-rate revenue target from $3.7 billion to more than $4 billion, with approximately 85% already under contract after $2.8 billion in new multi-year cloud deals.

The thesis is that IREN's marginal investor is no longer underwriting a Bitcoin mining hashprice story: they are underwriting a contracted AI Cloud revenue ramp whose economics, duration and counterparty quality are now firm-specific, not coin-price-driven. The sharpest evidence sits in the segment mix: Bitcoin Mining Revenue still accounted for 97% of FY2025 revenue on the company's own tagged fiscal-year figures, and AI Cloud Services contributed 3%, so the question is not whether AI Cloud is big today but whether the contracted trajectory, customer prepayments funding roughly 45% of GPU capital expenditure and customer prepayments plus GPU financing covering 95% of Microsoft-related GPU spend, lets the platform re-rate ahead of realized revenue. The bear case is that 63.2× trailing earnings against a TTM revenue base of $757.1M and a $14.7B market cap leaves no room for the 5 gigawatt power portfolio, the $9.7 billion Microsoft contract or the 1.2 gigawatt 2027 capacity target to slip.

Key takeaways

• IREN's Q3 FY26 earnings call reports a 63.2× trailing P/E against a $757.1M TTM revenue base, a -$408.4M TTM EBIT line, and a $14.7B market cap, with year-end 2026 AI Cloud annualized run-rate revenue guided to $4.0B per the 8-K filed 2026-07-20.
• The company's own non-GAAP Adj. EBITDA has fallen for three consecutive quarters from $91.7M in Q1 2026 to $75.3M in Q2 2026 to $59.5M in Q3 2026, while the EBITDA non-GAAP series sits at -$139.8M in Q3 2026.
• Trailing revenue mix on the company's own fiscal-year figures still places Bitcoin Mining Revenue at 97% of FY2025 revenue against AI Cloud Services at 3%, with $484.6M of FY2025 Bitcoin Mining Revenue dwarfing the $16.4M of AI Cloud Services.
• The 10-Q balance sheet as of March 31, 2026 shows $2.2B in cash and cash equivalents against $4.6B of total liabilities and $2.7B of total stockholders' equity, with -$614.8M of retained earnings and -$1.7B of additional paid-in capital absorbing the gap.
• IREN closed at $41.28, off the $76.41 52-week high by 46.0%, a drawdown that sits alongside the miner cohort (HIVE -54.2%, Bitdeer -52.7%) rather than the Capital Markets peers, while the $14.7B market cap and 21.3% short interest both place the stock in the miner bracket on positioning.

The market is pricing a miner; the contracts say a utility

Price performance vs peers (Capital Markets and cross-industry comps)

The earlier section set the thesis. The setup is the gap between the picture the chart implies and the picture the deals imply.

On price, IREN Ltd closed at $41.28, up 9.3% year-to-date against a 52-week change of +122.1% and a 52-week high of $76.41, a drawdown of 46.0% from that high. The cross-industry Bitcoin-mining comps HIVE Digital Technologies Ltd. (HIVE $3.18 -0.31%) and Bitdeer Technologies Group (BTDR $12.17 -0.73%) carry similar scars: HIVE sits 54.2% off its 52-week high, Bitdeer 52.7% off. The Capital Markets peers did not draw down at all. Morgan Stanley -4.4%, Goldman Sachs -4.7%, Schwab -6.0%. If the marginal holder is bracketing IREN with the miners, the price action is the consistent, expected result.

The 52-week move also tells a second story. IREN's +122.1% ranks above every Capital Markets peer in the table (MS +56.0%, Goldman Sachs Group Inc (GS $1,074.72 -2.14%) +56.8%, Schwab Charles Corp (SCHW $100.80 +0.84%) +5.7%, Robinhood Markets, Inc. (HOOD $104.48 -1.77%) +2.6%) and above the miner cohort (HIVE +27.6%, BTDR -16.3%, Bit Digital, Inc (BTBT $1.56 -1.27%) -52.3%). The +122.1% landed in a year when trailing revenue was $757.1M and EBIT was -$408.4M, on a market cap of $14.7B, a 63.2× P/E versus Goldman Sachs at 20.1×, Morgan Stanley at 19.8×, Schwab at 20.0×, Robinhood at 50.7×. The re-rating has already happened; the question is whether the deal book that drove it is being priced as a one-off or as a base.

The market's framing, to the extent one can infer it from the price table, treats IREN as a Bitcoin miner with optionality. The comp set the platform sits inside, HIVE, Bitdeer, Bit Digital, is a miner comp set. Short interest is the miner-level read: 21.3% of IREN's float, alongside HIVE at 21.0%, Bitdeer at 20.4%, Bit Digital at 15.1%. The Capital Markets names carry 1.1% to 4.2%. If the marginal book is sorting IREN by hashrate, the bear case is the standard miner case: BTC price, network difficulty, power cost, rig refresh. None of the four legs of that bull case appear in the most recent deal disclosures.

The contracting evidence cuts the other way. The 8-K filed 2026-07-20 describes a portfolio where approximately 85% of the raised $4.0B AI Cloud annualized run-rate revenue target is already under contract, weighted-average customer contract tenor approximately 4 years, and recent contracts carry customer prepayments representing approximately 45% of the associated GPU capital expenditure. That is the language of a contracted-revenue platform, not a hashprice-and-see-it business. The 2026-07-20 release adds that contracted "pricing continues to strengthen", a phrase that does not map onto a Bitcoin miner, where the issuer is paid in BTC and pricing is the spot market.

The call side makes the same point with sharper edges. On the Q3 FY26 earnings call, co-CEO Daniel Roberts said We are not chasing demand. We are racing to build supply fast enough to meet it, and added that "all of our operational capacity is fully contracted… there are no idle GPUs." Chief Commercial Officer Kent Draper, on the same call, extended that to pricing: "older generations, A100s, H100s, H200s, all fully, effectively, fully utilized across the industry… you're actually seeing pricing for older generation units climbing significantly." Forward visibility reached into 2028: "'27 people were thinking that there was a relatively decent amount of capacity available. We're already seeing that capacity available in '27 is extremely scarce and that is continuing to push into '28 now as well." A miner setup does not produce language of that shape: multi-year capacity scarcity, contracted pricing strengthening, and idle-GPU rejection as a planning case.

The stock-versus-deal gap is the setup. The 46.0% drawdown from the 52-week high occurred in the same two-month window in which the ARR target was lifted from $3.7B to more than $4B and the share of ARR under contract moved to approximately 85%. The miner bracket says the drawdown is a beta move; the contract book says the drawdown is a window into a setup the market has not yet fully re-read.

What management teams are saying

Across both the capital-markets peers and the cross-industry crypto-treasury names, management is telling the same story: AI demand is binding, supply is the bottleneck, and pricing power has shifted to the operator. The divergence shows up in what each side is doing with that setup. IREN is converting it into contracted ARR, the diversified banks are monetizing it through capital markets revenue, and the Bitcoin-pure names are absorbing the price without a comparable contract cushion.

Morgan Stanley frames the demand side as durable rather than peaky. Co-CEO Ted Pick described the current period as a backdrop where there is some pricing leverage, depending on the type of product, while CFO Sharon Yeshaya added that fixed-income macro results were roughly flat versus the prior year because FX volatility traded near historic lows. Pick's framing of cycle risk, that the firm has to demonstrate higher highs and higher lows to deserve a healthy P/E multiple, anchors a multi-quarter setup, not a one-quarter print. Net new assets stepped from $118B in Q1 to $148B in Q2, and Pick said just over half of the Q2 NNA came from IPOs, with the firm holding 70% of the top 100 unicorns by market cap in its workplace pipeline. That is the language of a flow engine with backlog, not a market-taker.

Goldman Sachs is reporting the same acceleration through the capital markets window. CEO David Solomon told the Q2 2026 call that momentum across the franchise has accelerated as clients pursue greater scale, with large-cap corporate M&A volumes up 90% in the first half of 2026, and that the firm became the first bank to cross $1 trillion in announced M&A volumes over a six-month period. CFO Denis Coleman added that advisory 1H26 volumes of $1.2 trillion left Goldman with roughly a $425 billion lead over the nearest peer. On the equities side, the firm delivered record revenue as single-stock volatility and dispersion remained high, and Asia was particularly strong in part from AI capital formation. Solomon also pointed to wallet-share gains, noting the firm is top three with 127 of the top 150 names, against roughly 44 of the top 100 at the January 2020 investor day. The framing, multi-year capacity buildout in client financing, with capital ready to deploy, is the same shape IREN's leadership is using to justify the contracted ramp.

Schwab is reading the demand side through engagement rather than transactions. CEO Rick Wurster, on the Spring 2026 Business Update, said investors opened 1.3 million brokerage accounts, brought $140B of core net new assets in 1Q26 and a further $120B in 2Q26, with daily average trades reaching a record 11.9 million in Q2. CFO Mike Verdeschi said bank deposit account fees grew 20% year-over-year as lower-yielding fixed-rate obligations matured into higher yields. The pricing language here is more cautious. Wurster and Verdeschi both described smaller trade sizes generating less revenue per trade, even as trade counts hit records, but the underlying engagement is up, and Verdeschi said the firm was tracking higher than the $5.70 to $5.80 EPS scenario laid out in January.

Robinhood's management is reading the same retail engagement and re-routing it. CEO Vlad Tenev, on the Q1 2026 call, pointed to Robinhood Cortex reaching nearly a million customers and to the rollout of the AI assistant to all Gold subscribers. The firm funded the quarter with a $2.2B zero-coupon convertible due 2029, used roughly $290M to repurchase 2.743 million shares concurrently, and layered $123.2M of capped calls to push the no-net-dilution threshold to roughly $237.85 per share, and factoring in the buyback, to roughly $303.95. CFO Shiv Verma said Q2 was off to a good start, with equities and options tracking as the highest-volume month of the year and net deposits already approximately $5 billion month-to-date. The mix shift away from crypto dependence, cryptocurrencies revenue was down 47% year-over-year at $134M even as total revenue grew 15%, shows the demand engine broadening without losing torque.

The cross-industry crypto-treasury comps split into two readings. HIVE's executive chairman Frank Holmes and CEO Aydin Kilic, in the FY2026 release, described a year of dual-engine scaling: installed hashrate up roughly four-fold to 25.1 EH/s while HPC contracted ARR reached $35M against a $660M year-end 2028 target. Kilic framed execution as the bottleneck, not customer demand. The pricing read on the legacy side was harsher: average Bitcoin price fell roughly 27% sequentially in Q4 FY26 and network difficulty rose roughly 27% year-over-year, and Q4 adjusted EBITDA flipped negative even as full-year adjusted EBITDA reached $72.9M. Bitdeer's management delivered a similar message through CFO Pretesh Dahya, who described Q1 with $188.9M of revenue against a $39.0M gross loss driven by Bitcoin price pressure, accelerated rig depreciation, and seasonal power costs at Norway and Bhutan. CSO Harris Bassett, on the same call, said the supply and demand imbalance for AI compute has widened and is expected to persist well into 2027 and beyond, and noted that H100 hourly rates were raised 40% on renewals. CEO Jihan Wu's framing was that most AI Cloud contracts are now three- to five-year commitments and that customers are quite competitive on the demand side. Bit Digital is the clearest contrast: CEO Sam Tabar described compute itself as becoming a new asset class, but the Q1 FY2026 transcript shows revenue mix continuing to shift away from mining and toward ETH staking, cloud services and colocation, with cloud services contributing roughly 60% of FY2025 segment revenue against a TTM total of $116.4M and an EBIT loss of $182.4M. The capital side is leveraged, CFO Erke Huang used a 20% leverage-to-ETH-bals metric as the funding decision threshold, and the ETH holdings gap between call and filing (a spoken figure near 155,444 against the 10-Q's 140,195.9 as of March 31, 2026) sits as a footnote on the management narrative.

The synthesis is sharp. The capital-markets desks, Morgan Stanley, Goldman, Schwab, Robinhood, are all reporting durable demand, with Solomon's $1 trillion M&A milestone and Pick's 70%-of-top-100-unicorns pipeline as the most concrete corroboration of the contracted-compute scarcity IREN's leadership is citing. Within the crypto-treasury comp set, HIVE and Bitdeer both confirm multi-year AI compute scarcity and reported pricing power on H100 renewals and on Tier-III colocation rates, HIVE's Bell B200 cluster was signed at roughly $2.90 per GPU-hour, above the $2.20 planning rate, while their Bitcoin-mining economics took a sequential hit on price and difficulty. The divergence is structural: IREN is the only name in the table pairing a contracted AI Cloud ARR ramp with a 5 gigawatt secured power position, $9.7 billion in contracted Microsoft revenue, and a customer-prepayment-funded GPU stack, while peers either lack the AI book (Bit Digital), lack the contracted multi-year tenor (Bitdeer's H100 renewals), or are operating from a much smaller HPC base (HIVE's $35M ARR against a year-end 2028 target of $660M). The peer set is unanimous on direction; IREN's edge is the contract structure sitting on top of that direction.

The trailing print is a hashprice print, not a platform print

Valuation vs peers (Capital Markets and cross-industry comps)

Performance vs peers (Capital Markets and cross-industry comps)

IREN Ltd management guidance — from the 8-K filed 2026-07-20

Metric: AI Cloud annualized run-rate revenue (ARR); Period: year-end 2026; Guidance: $4.0B.

IREN's own stated operating metrics are the scoreboard this case is judged against, and the scoreboard so far contradicts the contracted-revenue narrative the deals imply. On the company's own tagged fiscal-year figures, Bitcoin Mining Revenue carried 97% of FY2025 revenue and AI Cloud Services carried 3%, a mix so concentrated that the trailing twelve months on the comparative table, $757.1M of revenue against -$408.4M of EBIT, on a $14.7B market cap and a 63.2× P/E, still reads as a miner print. The forward case lives in the contracted ARR pipeline; the trailing print is what's in the income statement.

The Adj. EBITDA series, the metric IREN itself reports as its non-GAAP operating scoreboard, has been falling for three straight quarters. Q1 2026 came in at $91.7M, Q2 2026 at $75.3M, and Q3 2026 at $59.5M, a sequential decline of roughly $16M each period with no inflection. The series is reading peak-then-fade, the opposite of the ARR ramp the contracts imply. The number carries the direction: management is guiding ARR up while the metric IREN itself highlights is heading the other way.

The gap between the headline non-GAAP metric and the underlying GAAP-style result is the second tell. EBITDA (non-GAAP) sat at -$243.9M in Q2 2026 and -$139.8M in Q3 2026, with the negative narrowing by roughly $104M quarter-over-quarter. The improvement is the better-Adj.-EBITDA number combined with whatever the non-GAAP reconciliation adds back: the more relevant read is that Adj. EBITDA's $59.5M is the metric IREN asks the market to score it on, and that metric is in decline.

The segment mix makes the contradiction concrete. Bitcoin Mining Revenue grew from $75.5M in FY2023 to $184.1M in FY2024 to $484.6M in FY2025, a roughly 6.4× increase in two years, while AI Cloud Services went from $0 in FY2023 to $3.1M in FY2024 to $16.4M in FY2025. Even at the FY2025 mix, AI Cloud is 3% of revenue, and the dollar increment from FY2024 to FY2025 in AI Cloud is a fraction of the Bitcoin Mining increment. The trajectory of the company-stated segment lines is a mining company growing into AI optionality, not an AI platform with mining as a residual.

The peer segment mixes shown alongside are not directly comparable, since each company names its own segments, but the contrast reads. On its own reported lines, Morgan Stanley (MS $218.50 +0.97%) put 46% of FY2025 revenue in Institutional Securities and 44% in Wealth Management; Goldman Sachs Group Inc put 71% in Global Banking and Markets and 29% in Asset and Wealth Management; Schwab Charles Corp put 79% in Investor Services and 21% in Advisor Services; Robinhood Markets, Inc. had four transaction-based lines ranging from 6% to 44% of revenue; HIVE Digital Technologies Ltd. put 93% in Digital Currency Mining and 7% in HPC Hosting; Bit Digital, Inc put 60% in Cloud Services, 24% in Digital Asset Mining, 8% in Colocation and 6% in ETH Staking. Of the seven peers, only Bit Digital has a cloud-services majority, and its cloud share is being built on a total revenue base of $116.4M, a fraction of IREN's $757.1M. The bracketing on segment mix places IREN closer to HIVE (97% mining) than to any of the diversified banks. The market's bracket is the right one on the trailing print.

The valuation and performance tables then layer the question of whether the bracket is being repriced. IREN's market cap of $14.7B sits in the same range as the miner cohort (HIVE Digital Technologies Ltd. at $863.8M, Bit Digital, Inc at $544.7M, Bitdeer Technologies Group at $3.0B) and a fraction of the diversified Capital Markets peers (Morgan Stanley at $344.6B, Goldman Sachs Group Inc at $324.0B, Schwab Charles Corp at $175.3B, Robinhood Markets, Inc. at $94.1B). The 63.2× P/E is the second-highest in the table, ahead of Robinhood Markets, Inc. at 50.7× and several multiples above the diversified banks (Morgan Stanley 19.8×, Goldman Sachs Group Inc 20.1×, Schwab Charles Corp 20.0×). The multiple is paying for the ARR ramp, not the trailing print.

The operating-performance table shows IREN's revenue growing 167.7% year-over-year against the peers' mid-single-digit to high-single-digit prints (Morgan Stanley 5.5%, Goldman Sachs Group Inc 8.9%, Schwab Charles Corp -9.3%, Robinhood Markets, Inc. 51.6%), and short interest at 21.3% of shares against the Capital Markets names at 1.1% to 4.2%. IREN's 21.3% short sits alongside the miner cohort (HIVE Digital Technologies Ltd. 21.0%, Bitdeer Technologies Group 20.4%, Bit Digital, Inc 15.1%): the same bracket again. The year-to-date return of +9.3% against the 52-week change of +122.1% and a -46.0% drawdown from $76.41 captures the same gap: the trailing print miner, the forward print platform.

Management's own guidance on the 8-K filed 2026-07-20 names AI Cloud annualized run-rate revenue at year-end 2026 as $4.0B, anchored by approximately 85% under contract after $2.8B in new multi-year deals. The guidance is the binding item. The 8-K does not provide a quarterly Adj. EBITDA trajectory or a revenue-mix forecast, so the bridge from the contracted ramp to the scoreboard is management's argument, not the table's. Until the AI Cloud segment line in the company's own tagged fiscal-year figures moves off 3%, the table is what's being scored, and the table is a miner.

Liquidity is the cushion; the GPU stack is the claim

IREN Ltd balance sheet — as of Mar 31, 2026 (10-Q filed May 8, 2026)

The Q3 FY26 earnings call ended April 2026 with $2.6B of cash and cash equivalents, and the 8-K filed 2026-07-20 reports approximately $7.6B in cash and cash equivalents as of June 30, 2026, inclusive of $1.7B of restricted cash tied to the GPU financing for the Microsoft Horizon 1-4 contract. The 10-Q table that sits at the end of March 2026 is the snapshot before the second-quarter financing wave: $2.2B in cash and cash equivalents, $2.4B of total current assets, $651.4M of total current liabilities, $4.6B of total liabilities and $2.7B of total stockholders' equity against $7.3B of total assets. The earlier prompt framed it as a miner print; the balance sheet is the source of the contradiction, since it is being assembled to fund a contracted platform, not a hashprice bet.

The customers are paying ahead of the revenue line. Accounts receivable of $69.1M sit against deferred revenue (current) of $21.8M and deferred revenue (non-current) of $98.6M, and the 2026-07-20 release states that recent contracts include customer prepayments representing approximately 45% of the associated GPU capital expenditure. Co-CEO Daniel Roberts, on the Q3 FY26 call, described the funding dynamic as financing 95% of that CapEx at an average interest rate of about 3% through prepayments and GPU financing, and the Q2 FY26 call had previously disclosed the structure: a $3.6B delayed-draw term loan from Goldman Sachs and JPMorgan, at an interest rate expected to be less than 6%, combined with $1.9B of Microsoft prepayments, covering 95% of GPU-related CapEx for Horizons 1 through 4. The customer-funded structure is the cushion underneath the $4.0B ARR target.

The composition of the asset base is the second tell. Goodwill is $1.2M on the table, long-term investments are $0, and operating lease assets are $2.9M, so the $7.3B of total assets is overwhelmingly physical and contracted, not goodwill- or investment-portfolio-driven. The acquisitions referenced in the Q3 FY26 earnings release (Nostrum's 490MW in Spain and Mirantis) are additions to the power and operating stack, not balance-sheet paper. Roberts's framing on the call: that the capital is "out there as long as you sign good contracts and you show that you can execute and operate this capacity": describes a balance sheet built to be drawn against contracts, not a balance sheet built to be marked-to-market.

The NVIDIA equity instrument is the third piece. The Q3 FY26 release states IREN issued NVIDIA a 5-year right to purchase up to 30 million shares of ordinary stock at an exercise price of $70 per share, a right to invest up to $2.1 billion, subject to certain conditions including regulatory. The structure vests against deployments rather than running as a dilutive raise on day one, and it ties the customer to the GPU supply line. The 8-K of 2026-07-20 adds that customer prepayments plus GPU financing, the NVIDIA right, and the contracted Microsoft and NVIDIA pipelines stack together to back the $4.0B ARR target.

The other side is the equity book. Additional paid-in capital is -$1.7B, retained earnings are -$614.8M and accumulated other comprehensive income is -$39.4M, summing to a $2.7B total stockholders' equity that is materially below the $7.3B of total assets. The negative APIC is the Aurora legacy; the negative retained earnings are the cumulative effect of investing ahead of revenue. Liabilities of $4.6B against $2.7B of equity is a leverage ratio that a regulated bank would not run, but IREN is a contracted-asset builder, not a regulated bank, and the borrowing is tied to specific contracted GPU deployments at sub-6% rates, per the Q2 FY26 disclosure. The risk is execution: if the contracted ramp slips, the carrying cost of the GPU stack is on the equity holders, not the customers.

The bear case on the balance sheet is the speed of the spend. The 2026-07-20 release places the contracted capex obligation across a 5 gigawatt secured power portfolio, a 1.2 gigawatt 2027 capacity target, and the GPU financing for Microsoft Horizons 1-4, with the 1.7B of restricted cash tied specifically to that GPU financing. The gap between $7.6B of cash and the multi-year capex pipeline is the working capital of an AI infrastructure build-out, and the bull case is that customer prepayments and contracted GPU financing fund the build without diluting equity below the $2.7B book. The structure is sound on the figures; the question is whether the contracted counterparties deliver on the timeline.

What needs to go right

For the contracted-revenue thesis to deliver, four things must go right in lockstep over the next twelve months. First, the $4.0 billion year-end 2026 AI Cloud annualized run-rate revenue target, of which approximately 85% is already under contract after the $2.8 billion in new multi-year deals announced on July 20, 2026, must convert from contract paper into recognized revenue inside the company's own tagged fiscal-year segment line. The figure management has anchored to is the contract value, not the realized revenue, and the segment mix on the company's own fiscal-year figures still shows AI Cloud Services at 3% of FY2025 revenue against Bitcoin Mining Revenue at 97%, so the bridge from approximately $16.4 million of FY2025 AI Cloud to a $4.0 billion ARR target requires the segment to grow roughly two orders of magnitude in eighteen months while Bitcoin Mining stays a residual. Second, the GPU commissioning for Microsoft Horizon 1 must ramp on the timeline management laid out, with CFO Anthony Lewis saying on the Q3 FY26 call that Microsoft revenue and revenue from the additional 50,000 GPUs secured during the quarter were expected to begin ramping in Q3 2026. Third, the 1.2 gigawatt 2027 capacity target must be energised on the schedule the Q3 FY26 release set out, with 480 megawatts being delivered in calendar 2026 and 1.2 gigawatts targeted for 2027, and the Sweetwater One and Kiowa sites plus the Nostrum 490-megawatt Spain footprint plus the Mirantis operational layer all have to come online without the multi-month slip typical of hyperscale builds. Fourth, the financing structure has to hold: the $3.6 billion delayed-draw term loan from Goldman Sachs and JPMorgan at an interest rate expected to be less than 6%, combined with $1.9 billion of Microsoft prepayments, covers 95% of GPU-related CapEx for Horizons 1 through 4, with the NVIDIA $2.1 billion equity right vesting against deployments and customer prepayments representing approximately 45% of the associated GPU capital expenditure on recent contracts, per the July 20, 2026 release. The 5 gigawatt secured power portfolio must convert into contracted megawatt-hours at the pricing implied by management's commentary that contracted pricing continues to strengthen, with no idle GPUs and pricing for older-generation A100s, H100s and H200s climbing significantly per Chief Commercial Officer Kent Draper's remarks on the Q3 FY26 call.

The bear case has equal weight and lives in the same evidence. The trailing print is a miner print, and Adj. EBITDA, the non-GAAP metric IREN itself highlights as its operating scoreboard, has fallen for three consecutive quarters from $91.7 million in Q1 2026 to $75.3 million in Q2 2026 to $59.5 million in Q3 2026, with no inflection in the series and the EBITDA non-GAAP series at -$243.9 million in Q2 2026 and -$139.8 million in Q3 2026, indicating the Adj. EBITDA line is masking a much larger GAAP-style gap. The 63.2× trailing P/E against a $757.1 million trailing-revenue base and a -$408.4 million EBIT line is paying for the contracted ramp, not the trailing income statement, so any slip in the ARR conversion widens the air pocket. Co-CEO Daniel Roberts's framing on the Q3 FY26 call that we are racing to build supply fast enough to meet it is the bull case in one sentence; the bear case is that the same sentence describes a capacity-execution company whose trailing three quarters have produced declining Adj. EBITDA while raising the contracted target twice. If commissioning slips on Horizon 1 or the 1.2 gigawatt 2027 build, the deferred revenue line of $21.8 million current and $98.6 million non-current sits against $651.4 million of total current liabilities and $4.6 billion of total liabilities, and the negative APIC of $1.7 billion plus the -$614.8 million retained earnings on the March 31, 2026 balance sheet leave the equity cushion of $2.7 billion as the absorber. The cross-industry cohort confirms the execution risk is real: HIVE's executive chairman and CEO in the FY2026 release described a year where Q4 Bitcoin-mining revenue fell roughly a quarter sequentially on a roughly 27% decline in average Bitcoin price and a roughly 27% year-over-year increase in average network difficulty, with Q4 adjusted EBITDA flipping negative even as full-year adjusted EBITDA reached $72.9 million, and Bitdeer's CFO on the Q1 2026 call attributed a $39.0 million gross loss against $188.9 million of revenue to Bitcoin price pressure, accelerated rig depreciation and seasonal power costs at Norway and Bhutan, with the Chief Strategy Officer noting that the supply and demand imbalance for AI compute is expected to persist well into 2027 and beyond but also that the firm's BTC treasury shrank from 1,502 coins a year earlier to roughly 150 by mid-2026 as mined bitcoin was monetised.

Hashprice is the second leg of the bear case. Bitcoin Mining Revenue on IREN's own fiscal-year lines grew from $75.5 million in FY2023 to $184.1 million in FY2024 to $484.6 million in FY2025, but that trajectory was on the way up in the BTC cycle; a sustained move lower in BTC, paired with network difficulty continuing to rise, would compress the mining line that is still funding the operating base while AI Cloud ramps, and IREN's existing mining fleet has no idle-GPU cushion because management has explicitly steered capacity into AI. The Bitdeer management commentary on the Q1 2026 call was that the rig market is so weak that selling mining rigs would happen at a very bad price, so the firm is steering output into self-mining and co-mining rather than rig sales, a setup that raises hashrate exposure rather than lowering it; IREN's equivalent decision shows up in the same language of all of our operational capacity is fully contracted on the Q3 FY26 call, applied to GPUs rather than ASICs, but the mining-rig book still rides BTC. The short interest at 21.3% of shares, against 1.1% at Morgan Stanley, 2.2% at Goldman Sachs and 1.4% at Schwab, places IREN in the miner bracket on positioning even though the deal book sits in the capital-markets bracket. The capital-markets peers' management commentary points the other way for risk: Ted Pick at Morgan Stanley, asked on the Q2 2026 call whether equity-and-FICC pricing power would persist, said there is some pricing leverage though depending on the type of product, while Sharon Yeshaya noted that FX volatility traded near historic lows, the kind of cycle-sensitivity language that describes a franchise whose revenue tracks volatility rather than volume. If that volatility regime fades, the bid for IREN's GPU scarcity narrative fades with it, because the hyperscaler M&A and equity-capital-raising pipeline Goldman Sachs CEO David Solomon tied to AI capital formation on the Q2 2026 call is the upstream driver of demand for the air- and liquid-cooled capacity IREN is contracting into.

Power and regulatory execution form the third leg. The 5 gigawatt secured power portfolio referenced on the Q3 FY26 call, the Nostrum 490-megawatt Spain acquisition and the Sweetwater and Kiowa sites plus the European and APAC additions, all sit on interconnection, permitting and regulatory timelines outside IREN's control. The NVIDIA $2.1 billion equity right at a $70-per-share exercise price, disclosed in the Q3 FY26 release as a 5-year right to purchase up to 30 million shares subject to certain conditions including regulatory, vests against deployments, so any regulatory delay on NVIDIA's side delays the equity-funded portion of the GPU stack. Bitdeer's commentary on the Q1 2026 call about the 570-megawatt Clarington, Ohio project captures the same risk: the project is under contract with AEP, but litigation by American Heavy Plate Solutions is described by management as under extensive influence from MHR, a New York-based PE firm, and the case has not closed, leaving the site in a legal overhang while the rest of the pipeline moves forward. IREN's project list sits in the same kind of permitting matrix; the July 20, 2026 release announces $2.8 billion in new contracts but does not disclose commissioning dates for the gigawatts beyond the 1.2 gigawatt 2027 figure.

The two sides share one anchor. The contracted-revenue thesis is supported by approximately 85% of the $4.0 billion year-end 2026 ARR target already under contract, weighted-average customer contract tenor of approximately 4 years, and customer prepayments plus GPU financing covering roughly 95% of Microsoft GPU-related CapEx, with contracted pricing described by management as continuing to strengthen. The miner-thesis bear case is supported by three sequential quarters of declining Adj. EBITDA from $91.7 million to $75.3 million to $59.5 million, a trailing-revenue base of $757.1 million against a $14.7 billion market cap and a 63.2× P/E, and a segment mix on the company's own fiscal-year figures that places AI Cloud at 3% of FY2025 revenue. The thesis works if the contract-to-revenue conversion runs ahead of the hashprice drag; it breaks if a single quarter of commissioning slip or a sustained BTC move lower pushes the Adj. EBITDA series down another leg while the contracted target stays static, leaving the equity holders absorbing the gap.

The bull case rests on contract conversion; the bear case rests on the next quarter

The verdict falls on which side of the line management moves in the next reported quarter. On the bull side, IREN has $4.0 billion of year-end 2026 AI Cloud annualized run-rate revenue guided, approximately 85% already under contract after $2.8 billion in new multi-year deals, with a weighted-average customer contract tenor of approximately 4 years and customer prepayments plus GPU financing covering roughly 95% of Microsoft GPU-related CapEx. On the bear side, the same company's own non-GAAP operating scoreboard has declined for three sequential quarters from $91.7 million in Q1 2026 to $75.3 million in Q2 2026 to $59.5 million in Q3 2026, with the EBITDA non-GAAP series at -$243.9 million in Q2 2026 and -$139.8 million in Q3 2026, and the segment mix on the company's own tagged fiscal-year figures still places AI Cloud Services at 3% of FY2025 revenue against Bitcoin Mining Revenue at 97%. The evidence currently better supports the bull framing on contracted future revenue but the bear framing on what is in the income statement today.

The bull scenario is bounded by the contracted ARR target and the customer-funded structure behind it. If the $4.0 billion year-end 2026 ARR target converts on the timeline management laid out, with Microsoft Horizon 1 revenue and the additional 50,000 GPUs secured during the quarter beginning to ramp in Q3 2026 per CFO Anthony Lewis on the Q3 FY26 earnings call, the AI Cloud segment line moves off 3% of the company's own fiscal-year figures and the contracted-revenue bracket begins to overwrite the miner bracket. The 1.2 gigawatt 2027 capacity target, the 5 gigawatt secured power portfolio and the 480 megawatts being delivered in calendar 2026 provide the supply side. The customer-prepayment-led funding structure, with prepayments representing approximately 45% of GPU capital expenditure on recent contracts per the 8-K filed 2026-07-20 and approximately $7.6 billion in cash and cash equivalents as of June 30, 2026 inclusive of $1.7 billion of restricted cash tied to the GPU financing for the Microsoft contract, removes the dilution risk from the ramp. Co-CEO Daniel Roberts's framing on the Q3 FY26 call that the capital is out there as long as you sign good contracts and you show that you can execute and operate this capacity is the bull case in one sentence.

The bear scenario is bounded by the trailing print and the gap between contract value and realized revenue. The 63.2× trailing P/E on the valuation table sits against a $757.1 million trailing-revenue base and a -$408.4 million EBIT line, the multiple is paying for the contract book, not the income statement, and any slip in ARR conversion widens the air pocket. The Adj. EBITDA series is heading the wrong way for three quarters running, with no inflection in the data, and the EBITDA non-GAAP series at -$139.8 million in Q3 2026 shows the magnitude of the GAAP-style gap the Adj. line masks. The balance sheet, with $4.6 billion of total liabilities against $2.7 billion of total stockholders' equity as of March 31, 2026, has the $651.4 million of total current liabilities and the negative APIC of $1.7 billion plus the -$614.8 million retained earnings as the cushion; a single quarter of commissioning slip on Horizon 1 or a sustained move lower in BTC that compresses the $484.6 million FY2025 Bitcoin Mining revenue line pushes the absorption onto that equity. The 46.0% drawdown from the 52-week high of $76.41 already absorbed much of the contraction scenario, but the year-to-date return of +9.3% against the 122.1% 52-week change shows the bulk of the move happened before the contract book landed.

The next three events settle the contest. The Q4 FY26 earnings call, expected in August 2026, will show whether the Microsoft Horizon 1 revenue ramp that Lewis described as beginning in Q3 2026 has begun to flow through the AI Cloud segment line on the company's own tagged fiscal-year figures. Any movement of that segment off 3% of revenue toward double-digit share is the first concrete refutation of the miner bracket. The next earnings release will also produce a fresh Adj. EBITDA print, with the trailing series at $59.5 million in Q3 2026, and a fourth sequential decline below that level forces a re-read of the customer-funded structure. The next 8-K with a capacity commissioning update, expected as Sweetwater One energisation continues and Horizon 1 GPU commissioning proceeds per the Q3 FY26 release, will test the 1.2 gigawatt 2027 target and the operational conversion of the 5 gigawatt secured power portfolio. A slip in any of these three readouts opens the bear case; execution on all three closes the contest in favor of the bull.

---

Figures are sourced from SEC filings and Equibles' deterministic datasets; management commentary comes from earnings-call transcripts and filings gathered by Equibles research agents. This article is for information only and is not investment advice.