IREN IREN

At a glance

One-line thesis. IREN is a vertically integrated data-center owner-operator (ex-Iris Energy, Sydney-based, sites in Australia/Canada per its filed profile) pivoting from Bitcoin mining to AI cloud — the expert panel calls it "one of the clearest fundamental setups" and consensus pencils revenue rising ~8× from FY26E to FY28E, but the reported numbers today show three straight quarters of falling revenue, negative operating margins, ~$1.8B of trailing FCF burn and heavy dilution, so this is a Watch: the exponential is real on paper, and we want the print — not the promise — before paying beta-4.28 risk.

1. What it is

IREN Limited (Nasdaq: IREN, formerly Iris Energy; renamed November 2024) is a vertically integrated data-center business incorporated 2018 and headquartered at 55 Market Street, Sydney, Australia — it owns its computing hardware, electrical systems and the facilities themselves. Its filed FMP profile describes operations spanning Australia and Canada with Bitcoin mining as a primary undertaking; the forward story (per the expert panel, §2) is the conversion of that secured power into AI cloud / GPU compute. ~257 full-time employees; IPO 2021-11-17. The profile lists William Roberts as CEO, and the July 2026 Form 4s show Daniel John Roberts and William Gregory Roberts as Co-Chief Executive Officers. Fiscal year ends June 30 (FY25 = year ended 2025-06-30; Q4 FY26 reports 2026-08-27).

Revenue mix — what the data file actually shows:

• By geography (FY25, FMP segment data): Australia $484.6M · Canada $16.4M. This looks like a billing/parent-entity split rather than end-customer demand — the file contains no product-level segmentation (seg_prod is empty), so we cannot separate Bitcoin-mining revenue from AI-cloud revenue from this dataset. That is an honest gap.
• Classification caveat: FMP tags IREN as "Financial Services / Financial – Capital Markets," which is simply wrong for a data-center operator — it also pollutes the FMP peer list (§8). Judge it against miners-turned-neoclouds, not banks.

The strategic pivot the panel is underwriting: ~3GW of secured power and NVIDIA-preferred-partner status turning a Bitcoin miner into an AI-cloud landlord (jensen_huang-Tj_LZgN0oWM:4e470dfe86 — panel claim, not in the financial file).

2. The expert thesis — why the panel is bullish (traceable)

The Synthos KB holds 13 claims on IREN across 4 voices — all bullish, none bearish. Eight claims carry traceable claim_ids; five (Anthony Pompliano ×3, Natalie Brunell ×2) are sector-level miner-to-AI claims without claim IDs or dates, so we cite them only as thematic color. All voices carry skill 1.0 — there is no high-skill (2.0) anchor here as there was on MRVL. Three threads:

• "One of the clearest fundamental setups" — vertical integration + secured power. The Jensen-Huang-attributed voices (conviction 85–88) argue IREN is "vertically integrated, NVIDIA preferred partner, ~3GW secured power — positioned to sell all capacity in a compute/power-constrained world" (jensen_huang-Tj_LZgN0oWM:4e470dfe86, 2026-02-04; echoed at jensen_huang_ai-Tj_LZgN0oWM:2342211048, conviction 88). The same voice cites the concrete ramp math: "$3.4B AI cloud ARR by end of 2026 (~$850M/quarter into Q1 2027)… via the $9.7B Microsoft deal" (jensen_huang-Tj_LZgN0oWM:46df35ac1b, conviction 75) and "23,000 GPUs operational by end Q1 2026… +40,000 GPUs in BC by end-2026" (jensen_huang_ai-Tj_LZgN0oWM:245aa7778d). Honest weighting: these are company-projected figures relayed by a bullish commentator, and none of that ARR is visible in reported revenue yet (§5).
• Financing read as strength, not weakness. "Upsized/fully-exercised $2.96B 1%-coupon convert on good terms… signals strong demand" (jensen_huang--dSTakWQaR8:e7de899f59, 2026-05-18, conviction 75; also jensen_huang_ai--dSTakWQaR8:b27c9b2b49). Honest weighting: a 1% coupon is genuinely cheap capital, but it is still ~$3B of future dilution/debt layered onto a cash-burning balance sheet.
• The asymmetry framing. Real Vision (conviction 90, 2026-06-21 — the freshest claim): "entered 2022-23 at margin-of-safety with 50-100x asymmetry, still below fundamental value" (real_vision--p3DhlY2TEk:9dc3f47b12). Brunell's undated "next Carvana, 100-bagger" and "largely played out" claims and Pompliano's "miners are the leveraged tip of AI capex — down 40% from peak, not time to sell" are consistent sector color, but untraceable — we do not weight them.

Honest composite note. Breadth is moderate (4 voices), skill is unexceptional (1.0 across the board), stance is unanimously bullish, and the claim set leans heavily on company-projected ramp figures. There is no cautionary voice in the KB, so the bear case in §3 is built from the reported fundamentals — which currently point the other way from the panel.

3. Synthos scores & the Bull / Base / Bear cases

The one-glance judgment — three scores, 0–10, each anchored to real metrics:

The three cases (our own scenario model — assumptions labeled; each target is a ~12–18-month fair value). We deliberately do not attach probabilities; the cases bound the range, the scores summarize them.

Synthos fair value = the base case, ~$65 (+48%) — deliberately below the Street's $81.86 consensus, because the Street is largely marking the company's own ramp targets to model while the last three prints all missed. The 4.5× bull-to-bear ratio is the honest signal: this is a venture-style distribution inside a $15.7B listed wrapper. A +48% base-case gap would normally scream Buy; here the price action (below both DMAs, MACD negative) and three straight misses say the market is actively re-underwriting the ramp — so we demand the trigger first. This is a tracked call — the Forecaster Scorecard grades it once it matures.

4. Exponential Potential

Synthos separates compounders from exponentials. IREN is a textbook prospective exponential — the steep part of the curve is entirely in front of (not behind) the reported numbers:

• Forward growth (consensus): revenue FY26E $730M → FY27E $2.89B (+295%) → FY28E $5.67B (+96%) → FY29E $8.44B (+49%) → FY30E $8.75B (+4%). EPS: −$0.49 (FY26E) → $0.23 → $2.63 → then thin single-analyst coverage ($1.20 FY29E, $2.22 FY30E — treat as noise).
• The second derivative is in the estimates, not the actuals: reported quarterly revenue decelerated then declined through FY26 ($240.3M → $184.7M → $144.8M, with Q4 FY26E ~$163M). The entire acceleration case rests on contracted AI-cloud capacity converting to recognized revenue from FY27 — per the panel, ~$850M/quarter exiting calendar 2026 (jensen_huang-Tj_LZgN0oWM:46df35ac1b).
• Room to run: $15.7B cap against an FY29E consensus revenue of $8.4B is ~1.9× forward-forward sales — if the ramp lands, the stock is cheap; the constraint is execution and funding, not TAM.
• Reinvestment wall (the anti-exponential feature): unlike an asset-light exponential, IREN must pour concrete first — capex ran 3.1× revenue TTM and FY25 FCF was −$1.13B. This is an exponential with a capital wall, which is why the risk score is 9.
• Estimate-quality caveat: the FMP out-year EBITDA/EBIT estimate rows are internally inconsistent (deeply negative EBITDA alongside positive net income, e.g. FY30E EBITDA −$1.79B vs net income +$496M) — we use the revenue and EPS lines and flag the rest as unreliable.

Exponential Potential: High (8/10) — the steepest consensus ramp in our pool, with the honest caveat that it is a contracted-but-unproven exponential: not one dollar of the FY27 step-up is in the reported numbers yet.

5. Financials (real numbers — FMP annual/quarterly)

• Revenue: FY25 (ended 2025-06-30) $501.0M, +167.7% (FY24 $187.2M, FY23 $75.5M). TTM revenue ~$757M.
• Quarterly trajectory (the uncomfortable part): Q1 FY26 (Sep-25) $240.3M → Q2 (Dec-25) $184.7M → Q3 (Mar-26) $144.8M — down two straight quarters, and Q3 missed the Street's ~$220M estimate by ~34%. Q4 FY26E is ~$163M. The data file does not say why revenue fell; the panel's framing is mining capacity being converted to AI cloud ahead of the revenue arriving — plausible, but from the statements alone this is simply a shrinking top line.
• Margins: gross 53.6% TTM; operating margin −29.2% TTM (operating income was negative in all three FY26 quarters: −$60.1M, −$84.7M, −$93.3M). FY25 operating income was barely positive at $22.1M.
• Earnings quality — read very carefully. FY26 GAAP swings are dominated by non-operating items: Q1 FY26 net income +$384.6M rode +$635.4M of other income; Q2 was −$155.4M; Q3 was −$247.8M (diluted EPS −$1.25). TTM GAAP EPS nets to ~$0.04 — an accounting artifact, not earnings power. (FMP's income-quality ratio of 3.13 reflects OCF far exceeding that near-zero net income.)
• Cash flow: TTM operating CF ~$533M — genuinely positive and growing — but TTM capex ~$2.34B (3.1× revenue), so TTM FCF ≈ −$1.8B (FY25: OCF $245.9M, capex −$1.373B, FCF −$1.127B). FY25 funding: +$701M net debt, +$603M equity issuance.
• Balance sheet (stale — flagged): the latest annual balance sheet in our file is FY25 (2025-06-30): cash $564.5M, total debt $964.2M, net debt $399.7M, PP&E $1.94B, equity $1.82B. The TTM ratio set implies the current picture post-convert: roughly $3.7B total debt, ~$2.2B cash, net debt ~$1.75B, net-debt/EBITDA 3.67× — consistent with the $2.96B 1%-coupon convert the panel cites. No quarterly balance sheet is in the file; we say so rather than guess.
• Dilution: weighted shares FY21 52.9M → FY25 214.6M → 216.6M now — ~4× in four years, before the convert. Stock comp is 17.3% of TTM revenue.

6. Valuation — priced in or room?

Trailing multiples are useless here by construction: ~1,098× trailing GAAP EPS (on $0.04 of artifact earnings), 23.0× EV/sales, 36.5× EV/EBITDA, 3.6× book, FCF yield −11.5%. FMP's letter rating is C (overall 2/5; DCF, P/E and P/B scores 1/5). The whole case is the forward compression: at consensus, today's $43.91 is ~193× FY27E EPS ($0.23) → ~17× FY28E EPS ($2.63) → single-digit× on FY30E if the thin out-year coverage is right. A reverse read: the price only makes sense if FY28 arrives roughly on consensus — i.e., you are underwriting a ~4× revenue step in FY27 that management has projected, analysts have modeled, and the income statement has not yet shown. Street targets (context): consensus $81.86, high $99, low $50, median $80 — even the low target is above spot, which tells you the sell side has fully adopted the ramp; the market (−42.5% off the high) has not. That disagreement is exactly what a Watch verdict is for.

7. Technicals (from the tech block)

• Trend: broken. $43.91 sits below both the 50-DMA ($54.4) and the 200-DMA ($49.1), and the 50 has bent below the 200 — distribution posture, not an uptrend. MACD −3.68 (negative).
• Location: −42.5% off the 52-week high ($76.41) — also the max drawdown from peak — and +185% off the 52-week low ($15.40).
• Momentum: RSI(14) 25.1 — oversold, and today's +13.1% pop on ~53M shares (vs 45.6M average) is the kind of violent mean-reversion a beta-4.28 name produces; it is not, by itself, a trend change.
• Relative strength: +180% over 12 months vs SPY +21% / QQQ +31% — enormous banked outperformance; but the 3-month picture (+26.3% vs QQQ +23.6%) shows leadership fading toward the pack, and the price is fighting its own moving averages the whole way.
• Read: oversold-in-a-downtrend. The lower-risk entries are (a) a reclaim and hold of the 200-DMA (~$49) confirming the price action has re-accepted the ramp, or (b) a capitulation flush into the low-$30s where the bear-case math starts to floor it. Chasing +13% days in between is how beta-4.28 names hurt people.

8. Moat & competitive position

The moat claim is owned power + vertical integration: land, ~3GW of secured energy (panel claim, jensen_huang-Tj_LZgN0oWM:4e470dfe86), self-built facilities and NVIDIA-preferred-partner access — in a world where grid connections, not GPUs, are the binding constraint. If true, that is a real, multi-year barrier: power queues cannot be short-circuited with capital alone. The limits are equally real: hyperscalers can build or buy elsewhere; rival miners (Cipher, TeraWolf, Galaxy) are executing the same pivot; and a one-customer-heavy AI-cloud book (the Microsoft-anchored ramp) is concentration, not a moat. Negative ROIC says whatever moat exists has yet to earn anything.

Peer set (FMP-supplied, market cap) — mostly useless, flagged: Banco de Chile $20B, Carlyle $16B, Galaxy Digital $8.4B, Houlihan Lokey $9.8B, KeyCorp $25B, Nomura $27B, Principal $24B, Tradeweb $22B. This is the "Financial – Capital Markets" misclassification leaking through — only Galaxy is even adjacent. The relevant cohort (Cipher, TeraWolf, Core Scientific-style miners-turned-neoclouds, CoreWeave-style GPU clouds) is absent from the file; judge IREN against that group, not this list.

9. Management, capital allocation & guidance

• Capital allocation: everything into the build — FY25 capex $1.373B (5.6× operating cash flow), funded by $701M of net debt and $603M of fresh equity; no dividend, no buyback (correctly — there is nothing to return). The panel reads the $2.96B 1%-coupon convert as cheap, in-demand capital (jensen_huang--dSTakWQaR8:e7de899f59); we agree on the coupon and note the eventual dilution.
• Insider activity (Form 4s, 2026-07-01): almost entirely awards at $0 — CFO Anthony Lewis +26,968 shares; directors +6,657 each; and very large indirect awards to each Co-CEO: 552,197 + 9,099,328 shares apiece (Daniel and William Roberts, who each report ~23.6M shares held indirectly). Two awards of ~9.1M shares each is ~8% of the 216.6M weighted share count in potential founder equity — worth an RIA's attention on dilution grounds even if performance-linked (the file does not show vesting terms). The only sale: director Christopher Guzowski, 11,958 shares at $36.32 (transaction dated 2025-09-16, filed 2026-07-01) — small, and below today's price.
• Guidance/track record: no management claims are in our KB for IREN (no IREN_mgmt voice), and the file has no earnings-call text. What it does have is the scoreboard: the last three prints missed EPS estimates (−0.34 vs +0.15 est; −0.44 vs −0.24; −0.25 vs −0.22) and Q3 FY26 missed revenue by ~34%. Whatever the ramp targets are, the near-term delivery record is poor — that is the single most important management datapoint we hold.

10. Catalysts & what to watch

• Next earnings: 2026-08-27 (Q4 FY26; Street EPS −$0.32, revenue ~$163M). The key line is not the quarter itself but AI-cloud revenue recognition and any FY27 revenue framing — consensus needs ~$2.89B in FY27; the exit rate must start showing immediately.
• AI-cloud contract conversion: evidence the Microsoft-anchored book (panel: $9.7B TCV, $3.4B ARR target exiting 2026) is turning into recognized revenue on schedule — the entire thesis in one line.
• GPU deployment milestones: the panel's markers — 23k GPUs by end Q1 2026, +40k in British Columbia by end-2026 (jensen_huang_ai-Tj_LZgN0oWM:245aa7778d) — any confirmed slippage is a direct hit to FY27 numbers.
• Funding events: another equity raise or convert before FY27 revenue arrives would confirm the bear-case funding treadmill.
• Bitcoin economics: still the incumbent business; a BTC downdraft pressures the cash flows funding the transition.
• Technical trigger: a reclaim and hold of the 200-DMA (~$49) — our tier trigger — or capitulation into the low-$30s.

Thesis tripwires (what would change the call): Q4 FY26 or Q1 FY27 showing a clear AI-cloud revenue step-up → upgrade path to Buy — Tactical; a Microsoft-ramp delay/renegotiation, a fourth straight miss, or a surprise dilutive raise → downgrade path to Avoid.

11. Key risks

• Execution / conversion risk (dominant): the FY27 consensus requires ~4× revenue growth in one fiscal year from contracts not yet visible in the P&L — after three straight quarters of declining revenue and three straight EPS misses.
• Customer concentration: the ramp is anchored on one hyperscaler relationship (per KB claims) — a renegotiation, delay or capacity re-scope is a thesis-breaking event, not a haircut.
• Funding & dilution: TTM FCF ≈ −$1.8B; ~4× share-count growth since FY21; ~$3B convert outstanding; 17.3% of revenue in stock comp; and fresh ~9.1M-share awards to each co-CEO. Owners are paying for the build twice — in capex and in dilution.
• Leverage: net-debt/EBITDA 3.67× on EBITDA that is itself swollen by non-operating items; interest coverage is negative on an EBIT basis.
• Volatility / de-rating: beta 4.28 and a −42.5% drawdown that is still in progress; a broken tape means the market is actively questioning the ramp.
• Bitcoin cyclicality: the legacy business funding the pivot is itself volatile and capital-hungry.
• Data gaps (honesty): no product-segment split (mining vs AI cloud), no quarterly balance sheet, stale annual balance sheet, inconsistent out-year EBITDA estimates, and a misclassified sector/peer set — conviction should be capped by what we cannot verify.

12. Verdict, position sizing & monitoring

Watch. The exponential is real on paper — a ~8× consensus revenue path, ~3GW of secured power, unanimous (if modest-skill) expert bullishness, and a Street that sees +86% to consensus and +14% even to its lowest target. But Synthos does not pay beta-4.28 prices for ramps that exist only in estimates: reported revenue has fallen three straight quarters, operations lose money, FCF is deeply negative, the last three prints all missed, and the price action is below both moving averages. Base-case fair value ~$65 (+48%) is genuinely attractive — which is exactly why this is a Watch with defined triggers rather than an Avoid.

• Trigger to act (either): (a) a reclaim and hold of the 200-DMA (~$49) plus the 2026-08-27 print showing AI-cloud revenue actually inflecting — upgrade candidate to Buy — Tactical; or (b) capitulation into the low-$30s, where entry approaches the bear-case floor math.
• Sizing if triggered: satellite, 0.5–1.5% maximum — sized so a 60% drawdown (well within a beta-4.28 name's range) is survivable and re-buyable.
• Monitoring: re-underwrite on the §10 tripwires; formal re-score at each print (next 2026-08-27). This verdict is logged as a tracked Synthos call as of 2026-07-06 at $43.91.
• Single biggest risk: the AI-cloud ramp slipping or repricing while the capex burn and dilution continue — the story dies faster than the balance sheet can adapt.

Provenance & disclosures

• Traceability: 13 KB claims across 4 voices (Jensen Huang-attributed ×7, Real Vision ×1, Anthony Pompliano ×3, Natalie Brunell ×2), all bullish, top skill 1.0, last dated claim 2026-06-21. 8 claims carry claim_ids and are cited inline; 5 (Pompliano/Brunell) lack claim IDs and dates and are used only as sector color. The KB contains no bearish IREN voice and no IREN_mgmt voice; the bear case is Synthos's own, built from reported fundamentals. kb_net_conviction is left null rather than invented.
• Data as-of: fundamentals 2026-03-31 (Q3 FY26, filed 2026-05-08) · estimates & prices 2026-07-06 (data-file pull) · expert claims through 2026-06-21. Forward figures are analyst consensus (FMP), labeled as estimates.
• Data-quality caveats (material): (1) latest balance sheet is annual FY25 (June 2025) — current debt/cash are inferred from FMP TTM ratios (~$3.7B debt / ~$2.2B cash) consistent with the convert the panel cites; (2) FMP out-year EBITDA/EBIT estimates are internally inconsistent and were not used; (3) FY29–FY30 EPS estimates rest on a single analyst; (4) no product-segment (mining vs AI cloud) split exists in the file; (5) FMP's earnings-calendar EPS actuals (e.g., −$0.25 for Q3 FY26) differ from GAAP diluted EPS in the income statement (−$1.25) — adjusted vs GAAP basis; both are shown where used; (6) sector is misclassified as Financial Services, contaminating the supplied peer list.
• Panel-figure caveat: the $9.7B Microsoft deal, $3.4B ARR target, ~3GW power and GPU counts come from KB expert claims (cited by claim_id), not from the financial data file — they are the panel relaying company projections, and none are yet visible in reported revenue.
• Not investment advice. Independent research, educational and informational only, never personalized. Hypothetical/forward figures are labeled; the only performance numbers Synthos will headline are the live, real-money Flagship's.
• Version: 2026-07-06. Prior versions available via the deep-dive version dropdown ("based on the info at the time").