Stock Analysis · Bulls vs Bears

IREN Ltd IREN

🐂 Bulls vs Bears 🐻Report #19 Oct 2026
★ Willem owns this stock · 1.8% of the portfolioOlder version — see the latest →
67
high risk
🐂 7vs9 🐻
Bears lead
45%
▲ Chance of the biggest tailwind the forecasts miss
60%
▼ Chance of the biggest headwind the forecasts miss
The verdict

IREN is a founder-run bet that grid power bought cheaply for bitcoin can be turned into a top-tier AI cloud faster than anyone else. The customers and prices are real, but the next 12 months demand a step-up from ~40 MW to ~300 MW of working AI capacity and $25–30B of spending funded largely by other people's money. That puts dilution, delivery and GPU price risk ahead of the long-term power advantage.

The Bulls vs Bears podcast is being made
Two AI hosts debate this report, NotebookLM-style. It appears here automatically within a few hours.

Written by Claude (Anthropic’s AI) from public sources · Willem owns IREN · AI can make mistakes — numbers in brackets point to the sources · Not financial advice

🐂 Bull arguments 7

  1. 1Signed, top-tier demand that covers 2026 capacityStrong
  2. 2Energised power in the places that are closing their doorsStrong
  3. 3Unusually strong unit economics on new contractsMedium
  4. 4Proven access to very large amounts of capitalMedium
  5. 5Nvidia is a customer, supplier and would-be shareholderMedium
  6. 6Builder discipline: big milestones mostly metMedium
  7. 7Optional upside in the land bank beyond 2027Weak

🐻 Bear arguments 9

  1. 1The funding gap and continuing dilutionHigh
  2. 2The December quarter is a huge execution step, with termination rights attachedMedium
  3. 3A short-lived, fast-depreciating assetMedium
  4. 4Heavy reliance on two customers who are also IREN's biggest competitors and suppliersMedium
  5. 5Operating quality at the older sites is under public attackMedium
  6. 6Rising leverage and a widening cost of moneyMedium
  7. 7GAAP numbers lag the story, and cash flow is flatteredMedium
  8. 8Founder control and big equity grants without insider buyingLow
  9. 9Regulatory and trade frictions in every geographyLow

Bull arguments show how well they survive the devil’s advocate (strong / medium / weak); bear arguments show the damage they could do after the counter-argument (high / medium / low).

Bull case — all positive points (most important first)

🐂 Bull #1Signed, top-tier demand that covers 2026 capacity

  • The point: IREN is not building on spec for 2026. Its capacity is pre-sold to the most creditworthy buyers in AI, with large upfront payments.
  • Evidence:
    • Microsoft ~$9.67B, with ~$1.93B prepaid [14].
    • Nvidia ~$3.4B for 5 years [8].
    • ~$2.8B of further contracts in July [9], plus a frontier-lab contract in August [2].
    • Capacity for 2026 is described as fully contracted at >$4B ARR [3].
    • Operating ARR went from ~$0.5B (June) to ~$1B (late August) once Microsoft accepted Horizon 1 [3,18].
  • Why it matters: contracted revenue lets IREN borrow against GPUs at investment-grade rates (~6%) instead of selling shares. The 20% Microsoft prepayment plus the $3.6B loan fund ~95% of the Microsoft GPU cost [4,16].
  • 😈 Devil's advocate:
    • Revenue is earned only after acceptance. The Microsoft contract allows immediate termination of a tranche if delivery misses the (redacted) "Delivery Delay Window", with the prepayment refunded [14].
    • Microsoft and Nvidia are also the two parties with the most bargaining power over IREN.
    • A $4B ARR figure on ~300 MW still has to show up as GAAP revenue. Q4 FY26 AI revenue was $70.5M [2].
  • Verdict: Strong the contracts are real and filed with the SEC. The risk is delivery, not demand.

🐂 Bull #2Energised power in the places that are closing their doors

  • The point: in AI, getting electricity is now slower than getting chips. IREN already has power where newcomers face queues and audits.
  • Evidence:
    • Childress (750 MW) is energised, and the Sweetwater 1 substation (1.4 GW) was energised in the March 2026 quarter [1,8].
    • The 10-K lists ~5 GW of executed grid agreements or equivalents [1].
    • Texas rules have tightened for newcomers [44,45]:
      • Abbott's 2026-08-03 audit of all ERCOT queue projects.
      • Delayed "Batch Zero" grouping of large-load studies (results due 2027-04-09).
    • Northland reports that ERCOT provisionally classified Sweetwater as baseload in Batch Zero [24] (secondary source; not verified in an ERCOT document).
  • Why it matters: the 2027–2028 build (liquid-cooled Sweetwater 1 from H2 2027, Sweetwater 2 and Kiowa from 2028) can sell into a market where new power is scarce. That supports pricing [3].
  • 😈 Devil's advocate:
    • "Grid agreements or equivalents" includes letters of agreement, not only firm connections.
    • Sweetwater 2 and Kiowa are 2028 projects, and the 10-K warns the Texas audit could delay energisation even at current projects [1].
    • British Columbia now caps new data-centre power through a BC Hydro competition: 300 MW of AI capacity province-wide over two years, max 145 MW per project. That limits growth at the Canadian sites [46].
  • Verdict: Strong the Childress and Sweetwater 1 power is real. The further GW are options, not certainties.

🐂 Bull #3Unusually strong unit economics on new contracts

  • The point: IREN says new contracts pay back the GPU in about two years, and customers fund about half the GPU upfront.
  • Evidence:
    • More than $20M revenue per MW of IT per year, ~2-year payback, talks near $25M/MW [2,3].
    • Prepayments of 45–55% of GPU capex on recent deals [3,9].
    • 3-year pricing up ~125% since November 2025 [3].
    • Outside data agrees that pricing firmed in H1 2026. SemiAnalysis's 1-year H100 rental index rose from $1.70 (Oct 2025) to $2.35/hour (Mar 2026) [37].
    • Peers raised prices too: CoreWeave raised list prices ~25% in July 2026 (secondary) [39].
  • Why it matters: if a GPU pays for itself within a 3-year contract, IREN carries no residual-value risk. Anything earned after the contract is extra.
  • 😈 Devil's advocate:
    • Pricing is management's own, unaudited, and quoted per MW rather than per GPU-hour, so it can't be checked.
    • Redburn initiated (2026-09-21) by cutting CoreWeave and Nebius to Sell because of falling GPU rental prices [25].
    • Management itself says GPU and data-centre cost per MW is up 15–20%, so returns on the next tranche are lower than on deals priced off older GPU purchases [3].
  • Verdict: Medium the direction is supported by outside data, but the precise payback claims are unverifiable and peak-cycle.

🐂 Bull #4Proven access to very large amounts of capital

  • The point: few companies of this size can raise this much money this quickly, and IREN has shown it can.
  • Evidence:
    • The CFO says ~$19B was raised in 12 months: ~$16B from prepayments, GPU financing and converts, ~$3B from equity [3].
    • The Microsoft GPU financing is investment-grade rated at ~6% [16].
    • The Mackenzie GPU loan ($2.4B from Blue Owl and PIMCO) shows financing for non-investment-grade customers [2].
    • The data centres themselves are "100% unencumbered", so they can still be borrowed against [3].
  • Why it matters: the business is capital in, contracted cash out. Cheaper and more plentiful capital means less dilution per MW.
  • 😈 Devil's advocate:
    • Access depends on open credit markets and a high share price. Rising Treasury yields hit the stock on 2026-10-07/08 [28].
    • Non-investment-grade money already costs 9% against 6%.
    • The FY27 gap is still large (Bear #1).
  • Verdict: Medium impressive so far, but it is a skill that the market can take away.

🐂 Bull #5Nvidia is a customer, supplier and would-be shareholder

  • The point: IREN is tied unusually closely to the most important company in AI hardware.
  • Evidence:
    • Nvidia is a $3.4B customer [8].
    • It holds a 5-year right to buy up to 30M IREN shares at $70 (up to $2.1B), vesting as Nvidia GPU deployments reach up to 600,000 GPUs [1,8].
    • It signed a non-binding "5 GW partnership", with Sweetwater as the main site [1].
    • IREN got Nvidia "Exemplar Cloud" status in August 2026 [18].
    • Roberts says Nvidia refers customers [4].
  • Why it matters: Nvidia decides who gets scarce GPUs and steers demand. Its $70 strike, about twice today's ~$36 share price [28], signals what it would pay if deployments grow.
  • 😈 Devil's advocate:
    • The 5 GW partnership is explicitly conditional, "with no assurance" [1].
    • The $70 right is an option for Nvidia, not a commitment.
    • Nvidia supports many neoclouds (CoreWeave, Nebius, Lambda), so it is not exclusive to IREN.
  • Verdict: Medium a real relationship with real optionality, but none of it is binding beyond the $3.4B contract.

🐂 Bull #6Builder discipline: big milestones mostly met

  • The point: IREN has delivered large physical projects close to schedule, which is rare in this industry.
  • Evidence:
    • Sweetwater 1 substation on schedule [8].
    • Horizon 1 delivered and accepted in August 2026, within the September-quarter promise despite Nvidia component shortages [1,18].
    • ARR targets raised three times [6,8,9].
  • Why it matters: in a market where "bringing GPUs online is the bottleneck" [3], speed earns the contracts and the prices.
  • 😈 Devil's advocate:
    • The step-up ahead is far bigger than anything done so far: 150 MW of liquid-cooled Microsoft capacity plus BC and Childress air-cooled halls in one quarter, against ~40 MW running in June [1].
    • Interim targets have slipped by about a quarter (see section 1).
    • SemiAnalysis's reports on the BC sites question run-quality, not build speed [27].
  • Verdict: Medium a good record on fewer, smaller projects. The December quarter is the real test.

🐂 Bull #7Optional upside in the land bank beyond 2027

  • The point: the planned 2027 capacity (~0.8 GW IT cumulative) is only a small part of the ~5 GW portfolio [2].
  • Evidence:
    • Sweetwater 2 (600 MW), Kiowa, Oklahoma (1.6 GW), Bundey, South Australia (800 MW) and Badajoz, Spain (300 MW) are all in development [1].
  • Why it matters: each additional GW, if contracted at anything like current prices, is worth several billion of annual revenue.
  • 😈 Devil's advocate:
    • These are 2028+ projects in new jurisdictions (Oklahoma ratepayer rules, Spanish sustainability rules), each needing new financing [1].
    • By 2028 power scarcity may have eased.
  • Verdict: Weak to Medium — real option value, but too far out and too conditional to rely on.

Bear case — all negative points (most important first)

🐻 Bear #1The funding gap and continuing dilution

  • The point: the build plan costs far more than IREN has, so existing shareholders keep getting diluted.
  • Evidence:
    • FY27 capex is guided at $25–30B. It excludes GPUs for the liquid-cooled halls planned for H2 2027 [3].
    • In hand: ~$14B of cash, committed GPU financing and prepayments, with ~$8B more "targeted" [3].
    • Consensus FY27 free cash flow is about −$21B (TIKR, third-party) [22].
    • Ordinary shares went from 257.2M (2025-06-30) to 394.1M (2026-08-14): +53% in 14 months [1].
      • FY26 included 70.2M ATM shares and 39.7M shares in a December 2025 registered direct offering at $41.12 [1].
    • ~$3.5B of the $6B ATM remains (my arithmetic: $6B minus ~$2.49B sold) [1,17].
    • Convertibles add 127M potential shares (32%) if all convert. Only the small 2029/2030 notes ($13.64 and $16.81 conversion prices) are in the money today; the big 2031–2033 notes convert at $51–86 [1].
    • Jim Chanos (2026-03-05) read the $6B ATM filing as a sign that material good news was not imminent; the stock fell 8.5% the next day [31,32].
  • Why it matters: if new equity is sold at ~$36, the value of each existing share in future MW and ARR shrinks. The business could succeed and the shares still underperform.
  • 🛡️ Counter-argument:
    • Most of the gap is meant to be filled by non-dilutive money: prepayments, GPU loans and loans against the unencumbered data centres [3].
    • The 2033 convert ($3B) converts only at $73.07, about double today's price [15].
    • No ATM shares were sold between 2026-06-30 and 2026-08-14 [1].
  • Severity: High even with the counter-argument, roughly $3–8B of the plan has no named non-equity source.

🐻 Bear #2The December quarter is a huge execution step, with termination rights attached

  • The point: IREN must move from ~40 MW of operating AI capacity to ~300 MW IT within months. That includes 150 MW of liquid-cooled GB300 capacity for Microsoft, whose contract carries real penalties.
  • Evidence:
    • ~40 MW was operating at 2026-06-30 [1].
    • Horizons 2–4 are targeted for the December quarter, mostly "late in the quarter", with contractual grace periods to the start of Q2 2027 [1,3].
    • The filed contract sets out what happens on a miss [14]:
      • Daily delay credits accrue (rate redacted).
      • Missing the "Delivery Delay Window" lets Microsoft terminate a tranche "with immediate effect", with the prepayment refunded.
      • Missing an agreed updated date means the tranche is "deemed terminated".
    • The IREN parent guarantees any debt shortfall if a tranche is terminated and the GPUs can't be re-rented [16].
    • Headcount must keep tripling [3].
  • Why it matters: a terminated Microsoft tranche (~$2.4B of contract value each, my arithmetic: $9.66B ÷ 4) would hurt ARR, force a refund, and test the parent guarantee.
  • 🛡️ Counter-argument:
    • Excluded delays (force majeure, Nvidia supply failure) don't count against IREN [14].
    • Horizon 1 shows the template works [18].
    • Microsoft has every reason to accept: it wants the capacity.
    • Even a terminated tranche leaves IREN with GB300 GPUs in a market where management says pricing is up.
  • Severity: Medium High — termination is unlikely, but a slip of one or two quarters is plausible, and the stock is priced off the December ARR exit.

🐻 Bear #3A short-lived, fast-depreciating asset

  • The point: IREN owns GPUs that lose value quickly, rented on contracts shorter than the GPUs' useful life.
  • Evidence:
    • HPC hardware is depreciated over 5 years [1].
    • Contracts average ~4 years (weighted by ARR) [9].
    • Nvidia's Vera Rubin NVL72, the next generation, was running at CoreWeave, Google Cloud, Azure and Oracle by July 2026 [41].
    • The FT reported (2026-09-29) that Nvidia shared chip-depreciation data with an insurer, in talks over cover for neocloud loans should GPUs not resell. Talks may come to nothing [38] (secondary write-up of the FT story).
    • Redburn's Sell calls on CoreWeave and Nebius rest on falling GPU rental prices [25].
    • The 10-K now adds risk factors on "GPU technology obsolescence and shorter useful lives" [1].
  • Why it matters: if older GPUs re-rent for much less, or cannot be re-rented, IREN faces impairments. Its 5-year depreciation would then overstate earnings, and the real return per MW would be lower than management's payback math.
  • 🛡️ Counter-argument:
    • With a ~2-year payback and 45–55% prepayment, the GPU is paid back within the contract [3].
    • The SemiAnalysis index showed older H100s being re-rented at higher prices in H1 2026 [37].
    • Draper says older H100/A100 GPUs are fully used, with pricing "climbing significantly" in places [4].
  • Severity: Medium the risk is real but shows up in 2028–2030, and early payback limits it, if the payback claims are true.

🐻 Bear #4Heavy reliance on two customers who are also IREN's biggest competitors and suppliers

  • The point: Microsoft and Nvidia are a "substantial majority" of contracted revenue [1]. Both have options other than IREN.
  • Evidence:
    • Concentration as stated in the 10-K [1].
    • Microsoft is shifting capacity to operating leases and building its own data centres ($175B 2026 capex) [36].
    • The Microsoft contract includes a change-of-control termination right if Amazon, Oracle or Alphabet gains control of IREN [14].
    • Nvidia supplies IREN's GPUs and also supports rivals [1].
    • The smaller customers are often private AI start-ups with credit risk [1].
  • Why it matters: if either big customer doesn't renew in 2031, or pushes hard on price, IREN has few comparable buyers. The change-of-control clause also limits who could buy IREN without losing Microsoft.
  • 🛡️ Counter-argument:
    • The contracts run to 2031 with prepayments.
    • Customer count is growing: 10+ named.
    • Management says it avoids exclusivity [3].
  • Severity: Medium locked in for now, but it caps negotiating leverage and narrows the list of possible acquirers.

🐻 Bear #5Operating quality at the older sites is under public attack

  • The point: the air-cooled British Columbia halls, converted from mining, may not meet the standard customers expect from a cloud.
  • Evidence:
    • On 2026-10-07 SemiAnalysis, an influential chip-industry research firm, posted customer accounts of multi-day outages, storage failures and air-quality problems at Prince George and Mackenzie, calling one the "#1 worst site in the industry" [27,28].
    • It reportedly placed IREN in its lowest ClusterMAX tier and argued that the 10-K overstates IREN's managed-cloud capability [28].
    • It said Childress and Sweetwater look "much better" [29].
    • The stock fell ~6% on Oct 7 and ~8% on Oct 8 to ~$35.71 [27,28].
    • IREN had not responded publicly as of the latest reports I found (2026-10-08) [29].
    • The BC capacity matters near term: it was the ~$1.5B-ARR British Columbia piece of the original $3.4B plan [6], it is funded by the 9% Mackenzie loan [2], and it is due in the December quarter [3].
  • Why it matters: cloud customers sign with providers they trust, and they talk to each other. Outages mean service credits now and lost renewals later. Smaller customers (the ones using BC) are exactly the market IREN bought Mirantis to win.
  • 🛡️ Counter-argument:
    • The claims are anecdotal user reports, not audited uptime data [27].
    • The Microsoft and Nvidia capacity is at Childress, not BC.
    • IREN has already decided to convert Canal Flats to liquid cooling [3].
  • Severity: Medium not thesis-breaking, but it hits the part of the business (smaller customers, managed cloud) where IREN needs to show it is more than a landlord with GPUs.

🐻 Bear #6Rising leverage and a widening cost of money

  • The point: IREN is turning from a debt-light miner into a highly leveraged financier of GPUs.
  • Evidence:
    • Debt principal was $7.71B at 2026-06-30, of which $6.77B is recourse to the parent [1].
    • The Microsoft GPU financing has a 1.05x debt-service-coverage covenant [16]: cash flow must cover debt payments 1.05 times.
    • Non-investment-grade GPU money costs 9% against ~6% for investment grade [2,16].
    • Q4 FY26 adjusted EBITDA was only $19.2M [2].
  • Why it matters: a thin covenant cushion and a two-tier funding cost mean a few points of utilisation or price can decide whether a facility performs.
  • 🛡️ Counter-argument:
    • The heavy GPU debt is non-recourse or limited-recourse and matched to contracts.
    • The convertible coupons are low (0–1% on most).
    • Cash was $5.9B at 2026-06-30 [1].
  • Severity: Medium

🐻 Bear #7GAAP numbers lag the story, and cash flow is flattered

  • The point: the reported figures show far less than the ARR headlines.
  • Evidence:
    • FY26 AI cloud revenue was $128.8M, against a >$4B ARR target [2].
    • FY26 net loss was $702.6M, including $638.8M of impairments on mining hardware [2].
    • Operating cash flow of +$2.1B includes $1.84B of customer prepayments [1].
    • Q4 adjusted EBITDA of $19.2M missed BTIG's $35M estimate [33].
    • The shares fell ~6–12% after the 2026-08-27 results [33].
  • Why it matters: investors can't yet check margins on AI cloud from audited numbers. Prepayment cash will reverse as lower invoices from month 25 [14].
  • 🛡️ Counter-argument: this is a timing effect of a fast transition. The first full quarter of Microsoft revenue (the March 2027 quarter) will show the real margin.
  • Severity: Medium Low — mostly a transparency issue until early 2027.

🐻 Bear #8Founder control and big equity grants without insider buying

  • The point: the Roberts brothers control the vote and just received very large grants. No insider has bought shares in 2026.
  • Evidence:
    • B-class shares carry 15 votes per ordinary share held [1].
    • Each co-CEO received 9,099,328 RSUs on 2026-07-01 [12,20]:
      • They vest over 4 years, with a 2-year holding period after vesting.
      • At ~$36 that is roughly $330M each (my arithmetic).
    • There were no open-market insider purchases in 2026. The last sales were 1M shares each by the co-CEOs at $33.13 on 2025-09-11 [20].
  • Why it matters: outside shareholders have little say over capital allocation, and capital allocation is the main decision at this company.
  • 🛡️ Counter-argument:
    • The founders each hold ~23.6M shares including RSUs [20].
    • The holding periods run to ~2032, aligning them with the long term.
    • No new grant is due until FY2031 [12].
  • Severity: Low Medium.

🐻 Bear #9Regulatory and trade frictions in every geography

  • The point: each of IREN's regions is adding rules for big power users.
  • Evidence:
    • Texas:
      • SB6 requires remote-disconnect equipment and backup or curtailment for new loads of 75 MW or more connected after 2025 [1].
      • Abbott's audit is in force; ERCOT says verification takes "several months" [44,45].
      • PUCT adopted interconnection standards on 2026-09-21 [42].
    • British Columbia: caps on new data-centre power (Bill 31) [46].
    • Oklahoma: ratepayer-protection and groundwater rules [1].
    • Trade:
      • US Customs challenged the origin of imported mining machines, with exposure of up to ~$100M under protest. IREN warns GPUs could face similar notices [1].
      • A Section 232 chip tariff exists, with data-centre exemptions (secondary) [47].
  • Why it matters: these rules add delay, cost or curtailment risk to the 2027–2028 growth sites.
  • 🛡️ Counter-argument:
    • Childress and Sweetwater 1 are already energised, and management says tighter rules favour incumbents [3].
  • Severity: Low Medium for the 2026–27 plan; higher for 2028+.

Tailwinds NOT in the forecasts

What is already priced in: consensus FY27 (to June 2027) revenue is $2.8B and FY28 ~$7.3B (TIKR, 2026-09-28) [22]. Interpretation: that already assumes the >$4B December ARR exit converts to revenue, and that a large part of the 2027 capacity (0.5 GW more IT) is contracted and delivered.

  • A binding hyperscale-size contract at Sweetwater (≥300 MW IT) — 45% chance within 36 months · Impact if it happens: High

    • What it is: a multi-year contract with a hyperscaler, Nvidia or a frontier lab covering a large block of Sweetwater 1 (1.4 GW gross), at or near current per-MW pricing.
    • Why it's not priced in: FY28 consensus covers 2027 capacity. A GW-scale Sweetwater anchor would mostly add revenue from 2028+, beyond most models. The Nvidia 5 GW partnership is non-binding [1].
    • What would trigger it / the sign to watch:
      • An 8-K naming a Sweetwater customer.
      • Nvidia's $70 purchase right starting to vest.
      • ERCOT Batch Zero results (2027-04-09) [45].
    • Why this probability: demand and power are there, and IREN has signed five-plus large deals in 12 months. But a 2028 start needs liquid-cooled GPUs ordered by 2027 and financing at a scale IREN hasn't yet done, so it is a bit worse than a coin flip.

  • Large non-dilutive data-centre financing — 60% chance within 36 months · Impact if it happens: Medium-High

    • What it is: a loan secured on the unencumbered Childress and Sweetwater buildings and land (the CFO's "data centre financing"), or a sale-leaseback of them, raising several billion without new shares.
    • Why it's not priced in: no deal has been announced. The bear case and the stock's fall since May assume equity fills the gap [3].
    • What would trigger it / the sign to watch: an 8-K with a data-centre facility, or a statement on the Q1 FY27 call (~early November).
    • Why this probability: peers (Applied Digital, Hut 8, TeraWulf) have raised this kind of debt against contracted halls, and IREN says it has "lots of preliminary conversations". But lenders prefer long leases to 3–5-year GPU contracts.

  • GPU rental pricing stays high into the first re-contracting wave (2028–2029) — 40% chance within 36 months · Impact if it happens: Medium

    • What it is: H100/B200/GB300 capacity re-rents at or above today's prices, so older GPUs earn well after their first contract.
    • Why it's not priced in: sell-side bears (Redburn) and most models assume GPU price deflation [25].
    • What would trigger it / the sign to watch: SemiAnalysis's GPU rental index, CoreWeave and Nebius price moves, IREN's renewal pricing disclosures.
    • Why this probability: hyperscaler capex is heading toward ~$1T in 2027 [51], which supports demand. But each new Nvidia generation historically cuts the price of the last.

  • Nvidia exercises a meaningful part of its $70 share-purchase right — 15% chance within 36 months · Impact if it happens: Medium

    • What it is: Nvidia invests up to $2.1B at $70 a share [1,8].
    • Why it's not priced in: the strike is about twice the current price, and vesting requires large GPU deployments.
    • What would trigger it / the sign to watch: a vesting disclosure in a 10-Q, or new Nvidia deployments.
    • Why this probability: it needs both the share price above $70 and very large deployments within 5 years.

  • Mirantis lifts IREN from GPU landlord to software-rich cloud — 25% chance within 36 months · Impact if it happens: Low-Medium

    • What it is: managed and on-demand services sold on Mirantis software earn a premium and win stickier customers.
    • Why it's not priced in: consensus models IREN on per-MW economics only, and the SemiAnalysis critique questions its managed capability [28].
    • What would trigger it / the sign to watch: disclosure of managed-service revenue or on-demand launches; SemiAnalysis's ClusterMAX rating improving.
    • Why this probability: software is a new skill for IREN, and the October 2026 reliability critique shows how far it has to go.

Headwinds NOT in the forecasts

  • Heavy further dilution (>60M new shares, ≥15%, within 12 months) — 60% chance · Impact if it happens: High

    • What it is: ATM sales, a registered direct offering, or new deals that come with shares.
    • Why it's not priced in: consensus targets (average PT ~$81 [30]) use the current share count, while a $3–8B unfunded slice at ~$36 would need ~85–220M shares (my arithmetic).
    • What would trigger it / the sign to watch: share count in the Q1 FY27 10-Q (~November), new 424B prospectus supplements, a new ATM.
    • Why this probability: IREN has issued ~137M shares in 14 months, and $3.5B of ATM capacity is already registered. Only a large data-centre loan avoids it.

  • Microsoft Horizons 2–4 slip beyond March 2027 — 25% chance · Impact if it happens: Medium-High

    • What it is: one or more tranches miss the December quarter and land at the end of the grace window or later, with delay credits. In a worse case a tranche is terminated (I put termination of at least one tranche at ~7%).
    • Why it's not priced in: consensus March-quarter revenue of ~$983M assumes the December ARR exit [22].
    • What would trigger it / the sign to watch:
      • An 8-K announcing each acceptance (as for Horizon 1 on 2026-08-13).
      • Q1 FY27 call wording on "late in quarter".
      • Nvidia GB300 component supply.
    • Why this probability: Horizon 1 slipped on component shortages, and three buildings at once is far harder. The grace period cushions this but doesn't remove it.

  • GPU rental prices fall >30% by 2028 on new capacity and Rubin — 35% chance · Impact if it happens: Medium

    • What it is: a supply wave (hyperscalers, CoreWeave's $35–39B capex, new neoclouds) plus Rubin makes Hopper and Blackwell capacity much cheaper to rent [39,41].
    • Why it's not priced in: management's "~$25M/MW" talk and FY28 consensus assume firm pricing.
    • What would trigger it / the sign to watch: the SemiAnalysis index falling back below ~$2/hour for 1-year H100 contracts; peers cutting prices; IREN prepayments shrinking below 45%.
    • Why this probability: every GPU generation has deflated the last, but contracted demand and power scarcity can delay it. Somewhat less likely than not within 36 months.

  • Customer losses or service credits from BC reliability problems — 30% chance · Impact if it happens: Low-Medium

    • What it is: customers at Prince George and Mackenzie claim credits, don't renew, or delay acceptance of the December-quarter BC capacity.
    • Why it's not priced in: guidance includes BC in the >$4B ARR, and IREN has not answered the claims [3,29].
    • What would trigger it / the sign to watch: an IREN response, ClusterMAX changes, BC ARR disclosures on the Q1 call.
    • Why this probability: user reports point to real problems, but the affected capacity is small next to Childress.

  • Texas rules delay Sweetwater or impose curtailment costs — 20% chance · Impact if it happens: Medium

    • What it is: the Abbott audit, SB6 backup or curtailment rules, or the PUCT large-load transmission-cost review (due by 2026-12-31) slows Sweetwater energisation or makes it costlier.
    • Why it's not priced in: management treats energised sites as safe [3]. The 10-K says delays are possible [1].
    • What would trigger it / the sign to watch: the PUCT transmission-charge decision, ERCOT audit findings, Batch Zero results (2027-04-09).
    • Why this probability: Sweetwater 1's substation is already energised, which lowers the risk. But the follow-on build and transmission costs aren't locked.

  • Credit markets close for neocloud GPU debt — 20% chance · Impact if it happens: High

    • What it is: an AI sentiment shock or a peer default pushes non-investment-grade GPU loan spreads up sharply or shuts the market, stalling the ~$8B "targeted" GPU financing.
    • Why it's not priced in: the funding plan assumes continued access [3].
    • What would trigger it / the sign to watch: CoreWeave or Nebius bond spreads; the FT-reported Nvidia insurance talks [38]; a failed IREN or peer financing.
    • Why this probability: credit has been plentiful, but the sector's ~$35B+ debt stacks (CoreWeave alone [39]) make a sentiment-driven pause plausible.

Investment Risk Score: 67/100

Risk area Weight Score (0–100) Why
Business & competition 25% 60 Contracted demand and energised power are real, but the product is a fast-depreciating, increasingly commoditised GPU rental. Big, well-funded rivals and the hyperscalers' own builds compete for the same customers.
Financial health & dilution 20% 80 $25–30B of FY27 capex against ~$14B in hand; ~$7.7B of debt; +53% share count in 14 months; ~$3.5B ATM left; ~127M potential convert shares.
Execution (delivering the plan) 20% 70 From ~40 MW to ~300 MW IT in about six months, with termination rights on the Microsoft tranches. Interim targets have slipped ~1 quarter, the BC sites face reliability criticism, and headcount is tripling.
Dependency & concentration 15% 70 Microsoft plus Nvidia are a "substantial majority" of contracted revenue. Nvidia is supplier, customer and would-be investor, Dell assembles the big orders, and the founders control the vote.
Regulation, policy & geopolitics 10% 55 Texas SB6, the Abbott audit and the PUCT cost review; BC Bill 31 caps; a US Customs origin dispute and chip tariffs. Mostly a 2028+ issue, as the 2026–27 sites are energised.
Sentiment & expectations 10% 60 Consensus PT ~$81 against ~$36 and FY28 revenue of $7.3B already assume a lot. The stock is ~53% below its 52-week high of $76.87 with ~22% short interest (aggregator data) [30], so some fear is priced.

Overall = 0.25×60 + 0.20×80 + 0.20×70 + 0.15×70 + 0.10×55 + 0.10×60 = 67 (high). The score is driven by money and execution, not demand: IREN has the customers, but it must fund and build an order of magnitude more than it runs today, inside 6–12 months, with assets that lose value quickly. First deep dive, so there is no change to report.

Conclusion

  • What kind of company this is: a capital-hungry, founder-controlled infrastructure company in the middle of a total business-model switch. It is turning cheap grid power, bought for bitcoin mining, into GPU rental to Microsoft, Nvidia and AI labs. Owning it is a bet on three things: power stays scarce long enough; IREN can build faster than rivals; and it can fund the build with mostly borrowed and prepaid money rather than new shares.

  • The 2–3 things that will decide the outcome:

    1. Whether Horizons 2–4 and the BC/Childress air-cooled capacity go live and are accepted by ~March 2027, turning ~$1B of ARR into >$4B.
    2. How the FY27 funding gap is filled: data-centre debt and prepayments, or shares at depressed prices.
    3. Whether GPU rental pricing holds through 2027–2028, so new contracts (2027 capacity, Sweetwater) are signed at $20M+/MW with big prepayments.

  • Signposts:

    • Confirming the bull case:
      • 8-Ks announcing acceptance of Horizons 2, 3 and 4 by 2027-03-31.
      • A multi-billion data-centre financing with no share issuance.
      • A named Sweetwater anchor contract.
      • A flat share count (~394M plus Mirantis and RSUs) in the next two 10-Qs.
    • Breaking the bull case:
      • Any Microsoft tranche terminated or pushed past the grace period.
      • Share count above ~455M by mid-2027 without matching new contracts.
      • New contracts signed with prepayments below ~30% or prices below ~$15M/MW.
      • Customer departures or credits at the BC sites.

  • Balanced verdict: The strongest bull version is that IREN is one of very few companies with energised, multi-GW power and a contract book led by Microsoft and Nvidia. It is signing new capacity at about twice the per-MW price of its first big deal, with customers funding half the GPUs, and Texas is shutting the door behind it. The strongest bear version is that a ~$14B company (market cap at ~$36) is trying to spend $25–30B in one year on assets that depreciate in five, with ~40 MW of working AI capacity today. It has already diluted shareholders by half, and the first public reliability complaints came this week. Today the evidence leans toward real demand but unresolved funding and execution risk. The next two quarters (December acceptances, the Q1 and Q2 FY27 10-Qs) will settle much of the debate either way.

Open questions & data gaps

  • Per-customer revenue and per-GPU-hour pricing: not disclosed. This matters because concentration and unit economics can't be tested independently. The $20M/MW and ~2-year payback claims are management's alone.
  • Microsoft contract key numbers are redacted: the delay-credit rate, the Delivery Delay Window and the GPU counts [14]. This matters for sizing the worst case in Bear #2.
  • Funding gap size: the split of the $25–30B capex between data centres and GPUs, and which part is uncovered, was not found. High importance.
  • IREN's response to SemiAnalysis: none found as of 2026-10-08. Also not verified: SemiAnalysis's own ClusterMAX publication, since the report was read through secondary coverage.
  • Consensus EBITDA and EPS for FY27–28: not found from a reliable source. Only revenue and FCF (TIKR) are available.
  • Texas: whether the Abbott audit touches Sweetwater's further build, and the scope of a reported 2026-09-22 permit halt (Bloomberg headline only, UNCONFIRMED).
  • British Columbia: whether Bill 31 restricts the conversion of IREN's existing BC capacity (160 MW) from mining to AI. Not found, though the Mackenzie GPU financing suggests it is going ahead.
  • Q1 and Q2 FY26 call transcripts (Nov 2025, Feb 2026) were not read. The Q4 FY26 transcript was read from a third-party copy (Motley Fool).
  • Analyst rationales (JPMorgan, Northland, Redburn, Needham) come from secondary summaries. The full reports are paywalled.
  • Today's share price: not confirmed. The last reported close was ~$35.71 on 2026-10-08 [28].
Sources
  1. IREN Ltd Form 10-K, fiscal year ended 30 June 2026 — SEC EDGAR — 2026-08-27 — https://www.sec.gov/Archives/edgar/data/1878848/000187884826000052/iren-20260630.htm
  2. IREN Reports FY26 Results (8-K Ex. 99.1) — SEC EDGAR / GlobeNewswire — 2026-08-27 — https://www.sec.gov/Archives/edgar/data/1878848/000187884826000051/irenreportsfy26results.htm
  3. IREN Q4 FY26 Earnings Call Transcript — Motley Fool (third-party copy) — call 2026-08-27, published 2026-08-31 — https://www.fool.com/earnings/call-transcripts/2026/08/31/iren-iren-q4-2026-earnings-call-transcript/
  4. IREN Q3 FY26 Results Call Transcript (8-K Ex. 99.1) — SEC EDGAR — call 2026-05-07, filed 2026-05-11 — https://www.sec.gov/Archives/edgar/data/0001878848/000187884826000028/irentranscript.htm
  5. IREN Form 8-K (Microsoft Partner SOW and Dell purchase agreement) — SEC EDGAR — 2025-11-03 — https://www.sec.gov/Archives/edgar/data/1878848/000114036125040072/ef20058139_8k.htm
  6. IREN Reports Q1 FY26 Results — SEC EDGAR / GlobeNewswire — 2025-11-06 — https://www.sec.gov/Archives/edgar/data/1878848/000187884825000079/q1fy26resultspressreleas.htm
  7. IREN Reports Q2 FY26 Results — SEC EDGAR / GlobeNewswire — 2026-02-05 — https://www.sec.gov/Archives/edgar/data/1878848/000187884826000014/a991irenreportsq2fy26res.htm
  8. IREN Business Update and Q3 FY26 Results (incl. $3.4B Nvidia contract) — SEC EDGAR — 2026-05-07 — https://www.sec.gov/Archives/edgar/data/0001878848/000187884826000025/irenreportsq3fy26results.htm
  9. IREN Signs $2.8bn in New Customer Contracts, Raises 2026 ARR Target to over $4bn (8-K Ex. 99.1) — SEC EDGAR — 2026-07-20 — https://www.sec.gov/Archives/edgar/data/1878848/000114036126028871/ef20078253_ex99-1.htm
  10. IREN Form 8-K, Dell purchase agreement (~$1.6B) — SEC EDGAR — 2026-05-26 — https://www.sec.gov/Archives/edgar/data/1878848/000187884826000030/iren-20260519.htm
  11. IREN 2025 Proxy Statement (DEF 14A) — SEC EDGAR — 2025 — https://www.sec.gov/Archives/edgar/data/1878848/000114036125037383/ny20053824x2_def14a.htm
  12. IREN Form 8-K, Item 5.02 (co-CEO RSU grants) — SEC EDGAR — 2026-07-01 — https://www.sec.gov/Archives/edgar/data/1878848/000114036126027202/ef20077190_8k.htm
  13. IREN Form 10-K, fiscal year ended 30 June 2025 — SEC EDGAR — 2025-08-28 — https://www.sec.gov/Archives/edgar/data/1878848/000187884825000063/iren-20250630.htm
  14. Microsoft–IREN GPU Partner Statement of Work (10-Q Ex. 10.3, partly redacted) — SEC EDGAR — 2026-02-05 — https://www.sec.gov/Archives/edgar/data/1878848/000187884826000015/exhibit103-partnerstatem.htm
  15. IREN Form 8-K, closing of $3.0B 1.00% convertible notes due 2033 — SEC EDGAR — 2026-05-14 — https://www.sec.gov/Archives/edgar/data/1878848/000114036126021285/ef20073507_8k.htm
  16. IREN Form 8-K, $3.6B Microsoft GPU financing — SEC EDGAR — 2026-06-01 — https://www.sec.gov/Archives/edgar/data/1878848/000114036126023427/ef20075181_8k.htm
  17. IREN Form 8-K, $6.0B ATM prospectus supplement — SEC EDGAR — 2026-03-04 — https://www.sec.gov/Archives/edgar/data/1878848/000114036126007919/ny20064909x2_8k.htm
  18. IREN press release, Horizon 1 delivered to Microsoft (8-K Ex. 99.1) — SEC EDGAR — 2026-08-13 — https://www.sec.gov/Archives/edgar/data/1878848/000114036126032638/ef20080141_ex99-1.htm
  19. IREN Form 8-K, Mirantis acquisition closing — SEC EDGAR — 2026-08-04 — https://www.sec.gov/Archives/edgar/data/1878848/000114036126031053/ny20079234x2_8k.htm
  20. Form 4 filings, Daniel and William Roberts, CFO and directors — SEC EDGAR — 2026-07-01 and 2025-09-11 — https://www.sec.gov/Archives/edgar/data/1878848/000095010326009958/ownership.xml
  21. IREN Prices $2B Convertible Notes Offering (Ex. 99.1) — SEC EDGAR — 2025-12-03 — https://www.sec.gov/Archives/edgar/data/1878848/000114036125044095/ny20060080x4_ex99-1.htm
  22. IREN Stock Is Up 17% in 2026. Here's Where the Stock Could Go (consensus figures) — TIKR — 2026-09-28 — https://www.tikr.com/blog/iren-stock-is-up-17-in-2026-heres-where-the-stock-could-go
  23. JPMorgan upgrades IREN, $65 target — Blockspace (secondary summary) — 2026-09-14 — https://blockspace.media/insight/jpmorgan-upgrades-iren-65-target/
  24. Northland initiates IREN at Outperform, $99 target — goldesel.de (secondary) — 2026-09-18 — https://goldesel.de/aktien/news/northland-stuft-iren-mit-outperform-ein-und-setzt-kursziel-auf-99-dollar
  25. Rothschild Redburn downgrades Nebius, CoreWeave to Sell on lower GPU prices — 24/7 Wall St. — 2026-09-21 — https://247wallst.com/investing/2026/09/21/rothschild-redburn-downgrades-nebius-coreweave-to-sell-on-lower-gpu-prices/
  26. Neocloud stocks fall as selling extends after a week of research notes — 24/7 Wall St. — 2026-09-25 — https://247wallst.com/investing/2026/09/25/neocloud-stocks-fall-as-selling-extends-after-a-week-of-research-notes-iren-drops-4-coreweave-slides-3-nebius-eases/
  27. IREN Stock Plunges as SemiAnalysis Flags Data Center Problems — TipRanks — 2026-10-07 — https://www.tipranks.com/news/iren-stock-plunges-as-semianalysis-flags-data-center-problems
  28. IREN stock falls after report flags infrastructure and disclosure concerns — Investing.com — 2026-10-08 — https://ca.investing.com/news/stock-market-news/iren-stock-falls-after-report-flags-infrastructure-and-disclosure-concerns-93CH-4872305
  29. IREN Drops 4% as SemiAnalysis Flags Its Canadian Data Centers — 24/7 Wall St. — 2026-10-08 — https://247wallst.com/investing/2026/10/08/iren-drops-4-as-semianalysis-flags-its-canadian-data-centers-core-scientific-pulls-back-2/
  30. MarketBeat IREN page (short interest, consensus PT, insider activity; aggregator) — MarketBeat — ~2026-10-08 — https://www.marketbeat.com/stocks/NASDAQ/IREN/news
  31. Short-seller Jim Chanos sounds alarm on IREN's $6 billion equity filing — Finviz/Benzinga — 2026-03-05 — https://finviz.com/news/330383/short-seller-jim-chanos-sounds-alarm-on-irens-6-billion-equity-filing-questions-silence-on-500-million-revenue-guidance
  32. IREN stock price down 8.5% — MarketBeat — 2026-03-06 — https://www.marketbeat.com/instant-alerts/iren-nasdaqiren-stock-price-down-85-heres-what-happened-2026-03-06/
  33. IREN stock falls 6% but analysts say buy the dip (BTIG, H.C. Wainwright) — CoinCentral — 2026-08-28 — https://coincentral.com/iren-stock-falls-6-but-analysts-say-buy-the-dip/
  34. IREN: Needham holds rating, raises FY27 revenue — Blockspace — 2026-08/09 — https://blockspace.media/insight/iren-needham-holds-rating-raises-fy27-revenue/
  35. Neocloud market forecast to approach $400B by 2031 — Synergy Research Group — 2026 — https://srgresearch.com/articles/neocloud-market-forecast-to-approach-400b-by-2031-driven-by-surging-ai-infrastructure-demand
  36. Follow the CAPEX: Q2 2026 Scoreboard — Platformonomics — 2026-07 — https://platformonomics.com/2026/07/follow-the-capex-q2-2026-scoreboard/
  37. The Great GPU Shortage (H100 rental index) — SemiAnalysis — 2026 — https://newsletter.semianalysis.com/p/the-great-gpu-shortage-rental-capacity
  38. Nvidia asking insurers to cover what happens if GPUs don't resell (citing FT, 2026-09-29) — Let's Data Science — 2026-09/10 — https://letsdatascience.com/blog/nvidia-asking-insurers-cover-what-happens-if-gpus-dont-resell
  39. CoreWeave bulls just got what they wanted, but its massive debt problem isn't going away — 24/7 Wall St. — 2026-08-12 — https://247wallst.com/investing/2026/08/12/coreweave-bulls-just-got-what-they-wanted-but-its-massive-debt-problem-isnt-going-away/
  40. AI Data Centers (Powered Shell Providers) Q2 2026 Earnings Recap — Rittenhouse Research — 2026-09-02 — https://rittenhouseresearch.substack.com/p/ai-data-centers-powered-shell-providers
  41. Nvidia Vera Rubin ramp — eeNews Europe — 2026 — https://www.eenewseurope.com/en/nvidia-rubin-gpu-samples-vera-rubin-ramp/
  42. Texas PUC adopts softened rules on data center interconnection — Utility Dive — 2026-09-21 — https://utilitydive.com/news/texas-puc-adopts-data-center-interconnection-rules/830899
  43. PUCT approves Batch Zero framework — Sheppard Mullin — 2026-06 — https://www.sheppard.com/insights/blogs/puct-approves-batch-zero-framework-that-will-provide-a-pathway-for-large-load-to-connect-to-the-grid-and-receive-electricity
  44. Governor Abbott Directs Comprehensive Data Center Audit — Office of the Texas Governor — 2026-08-03 — https://gov.texas.gov/news/post/governor-abbott-directs-comprehensive-data-center-audit
  45. What Governor Abbott's data center audit directive means for ERCOT and the Batch Zero study process — Gibson Dunn — 2026-08 — https://www.gibsondunn.com/what-governor-abbotts-data-center-audit-directive-means-for-ercot-and-the-batch-zero-study-process/
  46. Data centre regulation in British Columbia: competing for a limited supply — BLG — 2026-07 — https://www.blg.com/en/insights/2026/07/data-centre-regulation-in-british-columbia-competing-for-a-limited-supply
  47. US Section 232 proclamation imposes 25% tariff on certain semiconductors — EY Tax News — 2026-01-15 — https://taxnews.ey.com/news/2026-0209-us-section-232-proclamation-imposes-25-percent-tariff-on-certain-semiconductors
  48. Hashrate Index Roundup — Hashrate Index — 2026-09-28 — https://hashrateindex.com/blog/hashrate-index-roundup-september-28-2026/
  49. IREN jumps on $3.4B Nvidia deal — Yahoo Finance — 2026-05-07 — https://finance.yahoo.com/markets/stocks/articles/iren-jumps-7-3-4-144258886.html
  50. IREN Stock Fell 6% in a Day After SemiAnalysis Flagged Its Canadian Sites — TIKR — 2026-10 — https://www.tikr.com/blog/iren-stock-fell-6-in-a-day-after-semianalysis-flagged-its-canadian-sites-heres-where-the-stock-could-go
  51. Moody's hyperscaler capex forecasts marked up by $85bn to close in on $1trn by 2027 — DatacenterDynamics — 2026 — https://datacenterdynamics.com/en/news/moodys-hyperscaler-capex-forecasts-marked-up-by-85bn-to-close-in-on-1trn-by-2027/
All IREN Bulls vs Bears reports
Disclaimer. AI-generated research and podcast (Claude by Anthropic, voices by Google Gemini) for educational purposes — not financial advice and not a recommendation to buy or sell. Willem (MR.INVEST) owns this stock. Check the company’s own filings before making any decision.
← All Bulls vs Bears reports