IREN Ltd · Moat analysis

Report previewIREN.USReport date Last updated
← Back to reports
Report contents
01

Industry landscape

Where the company competes: power sites are the scarce asset

IREN Limited sells computing power out of data centers it builds and owns. Most investors file it under bitcoin mining, and that is still the largest use of its machines, but the arena that actually sets its returns is narrower and more physical: a small number of large grid interconnections in low-cost power markets, where a company can put hundreds of megawatts of load and turn the electricity into either SHA-256 hashing or rented GPU capacity. The company is incorporated in Australia, keeps its principal office in Sydney and trades on Nasdaq under IREN; its fiscal year ends June 30, and it reported 394 million ordinary shares outstanding as of August 14, 2026 on the cover of its fiscal 2026 10-K.

The value chain decides where the margin sits. Upstream, IREN buys ASICs from Bitmain and MicroBT and GPUs from Nvidia, and it buys firm power from utilities and grid operators. Downstream it faces two very different customers: the bitcoin network, which is not a customer at all but a protocol that pays whatever the block subsidy and fees happen to be, and AI developers that sign contracts for GPU hours at negotiated rates. Neither hardware vendor gives IREN a durable cost edge, because both sell the same catalog products to every buyer in the market. The utility earns a regulated return. What remains is the spread between the all-in cost of power and interconnection at a given site and the market price of the output. That spread is IREN's business.

Market structure: many miners, few cheap interconnections

Bitcoin mining at the network level is global and location-blind. Hashrate migrates toward the cheapest electricity on earth, no single operator can move the price of bitcoin, and the product is perfectly fungible. The listed self-miners alone — Marathon Digital, Riot Platforms, CleanSpark, Cipher Mining and TeraWulf among them — plus hosting and colocation operators such as Core Scientific and Applied Digital, plus a long tail of private farms, put more than a dozen credible competitors in the field. That count is the practical test: when meaningful players cannot be counted on one hand, entry barriers at that layer are weak. The machines are catalog items, the mining software is largely open, and a leased warehouse with a power contract is enough to start.

The site layer looks nothing like that. Large-load interconnections in ERCOT and in BC Hydro's service territory are limited, queue positions take years to secure, and long-lead electrical equipment remains a bottleneck for anyone starting from scratch. In any one power market the number of sites that combine cheap energy, existing capacity and permitted land is small. So the competitive boundary is split: IREN sells hashrate and GPU hours into a global market at prices it cannot influence, but it competes for the inputs — megawatts, transformers, land — in local markets where the rival list is short.

That local list is changing. Hyperscalers and AI data center developers now bid for the same power contracts and the same electrical equipment that miners do. Their entry raises the opportunity cost of every megawatt IREN holds, which cuts both ways for the company: it lifts what an AI contract at an owned site is worth, and it raises the price of buying or building the next one. The rival set for a West Texas interconnection today is no longer just other miners.

Barriers: high at the site level, low at the hashing level

The two layers deserve separate verdicts. At the hashing layer, barriers to entry are low and likely to stay low; nothing about buying ASICs and renting space protects an incumbent's margin. At the site layer, barriers are medium to high, because the constraint is time and permits rather than capital alone. Any moat IREN builds has to come from the second layer, and it has to survive the fact that its two revenue streams have different economics: bitcoin revenue resets with the price and with each halving, while GPU cloud revenue depends on multi-year contracts with a small number of large buyers who can multi-source across neoclouds, hyperscalers and rival chip vendors.

Whether IREN is fundamentally a power-site owner that happens to mine bitcoin, or a miner that caught a favorable window, is the question the market structure leaves open. The evidence so far supports the first reading: the scarce asset in this business is an energized, permitted site, and IREN owns its sites rather than leasing them. What would settle it is contract duration and customer concentration in the GPU business — how much of that revenue is locked in for several years, and how much reprices at the next negotiation.

02

Competitive advantage

Does the company have a moat? Moderate, and it sits in the interconnection

Moat rating: Moderate. IREN's protected asset is the energized land and grid position under its machines, not the machines. The company reports about 5GW of secured power, and its sites are freehold with self-built electrical infrastructure; that takes years and permits to replicate, and capital alone cannot shorten an interconnection queue. The layer above, ASICs hashing on a global network and Nvidia GPUs rented by the hour, is a commodity that any funded entrant buys at the same catalog price IREN pays. That split explains how the company earned a 79.8% gross margin in the quarter ended September 30, 2026 while running a $0.19 billion operating loss.

AI data-center capacity is consolidating around the operators who can deliver hundreds of megawatts on schedule. Substations, transformers, shells and cooling consume cash years before the first dollar of revenue, so a site needs a large committed load to earn its cost of capital, and a small one cannot. Excess returns in this sub-industry would show up as returns above the cost of capital across a full contract cycle. IREN's record is one profitable year in five.

Market position: shrinking in the old business, unproven in the new

Market position: Moderate. In bitcoin self-mining IREN is a mid-sized operator, and the operating metrics it reports show installed hash rate falling from about 50 EH/s to about 38 EH/s during fiscal 2026 as the company pulled miners to make room for GPUs. The trade was deliberate. It still means the position in the business paying today's bills shrank while replacement revenue was being installed: AI cloud revenue was $33.6 million in the March 2026 quarter, about 23% of the $144.8 million total.

Share across three or more years cannot be confirmed from the public figures, so stability has to be read from the company's own capacity and revenue. Revenue rose from $75.5 million in fiscal 2023 to $501.0 million in fiscal 2025, then fell 42% from the September 2025 quarter to $140 million in the June 2026 quarter, before rebounding 64% to $230 million in the September 2026 quarter. Stability: Weak. A line that moves that far inside five quarters belongs to a business being rebuilt, not one being defended.

Profitability: one year in five earned a return, and it was 6%

Profitability: Weak. Return on equity from fiscal 2022 through fiscal 2026 ran -209.8%, -46.3%, -4.1%, 5.97% and -23.4%. The one positive year also produced a 4.25% return on assets, below any plausible cost of equity. An estimated ROIC for fiscal 2025 is harsher: operating income of $0.02 billion against invested capital of about $2.2 billion (equity of $1.82 billion plus debt of $0.96 billion less cash of $0.57 billion) comes to roughly 1%. Fiscal 2026 reversed hard, with an operating loss of $1.05 billion on $710 million of revenue.

Gross margin tells a better story and a misleading one. It rose from 28.0% in fiscal 2023 to 43.9%, 68.3% and 68.9% in the next three years, then 79.8% in the September 2026 quarter, on cheap power and a rising bitcoin price. Everything below that line is excluded. In fiscal 2026, $0.49 billion of gross profit turned into a $1.05 billion operating loss, so about $1.54 billion of cost sat under the gross-profit line (estimated): GPU depreciation, site overhead and stock compensation for a workforce being retooled.

Leverage muddies ROE further. At September 30, 2026, liabilities were $15.5 billion against $20.61 billion of assets, a 75% liabilities ratio that includes deferred revenue and payables, and debt of about $10.1 billion stood against $5.12 billion of equity. With that much leverage and negative earnings, return on capital is the number that matters, and it is negative.

Margins and scale: a high gross margin over a negative net margin

Margins and scale: Moderate. Listed self-miners buy ASICs from the same few vendors and sell into the same hash price, so gross-margin differences come mainly from delivered power cost per megawatt-hour. The 68.9% fiscal 2026 gross margin is evidence of a low-cost site base, not of pricing power over customers. Scale is the weaker leg. Revenue divided by assets is under 5%, and the AI contracts have not yet spread fixed costs across enough output to reach net margin, which was -99.4% in fiscal 2026.

Overall rating: Moderate, on the sites and not yet on the returns

Overall rating: Moderate. The barrier is narrow but real: scarce interconnections and owned land that take years to replicate, plus two sophisticated buyers who have committed $13.1 billion over five years. The returns do not corroborate it. No year in the fiscal 2022-2026 span earned above a plausible cost of capital, and the largest capital program in the company's history is only now being deployed. Profitability sits below Fairly Strong, which caps the overall rating at Moderate; it is not Weak because the site layer keeps entrants out on a multi-year timeline and the contracts show buyers will pay for access.

Where the advantage comes from and how long it can last: one contract cycle, already sold

Three things separate operators in this sub-industry: the delivered cost of power, the megawatts actually energized, and the price and duration of the contracts signed against them. IREN's fiscal 2026 annual report and its announced agreements carry the evidence on all three. The metric that tests the moat is contracted revenue per dollar of capital deployed. IREN has announced $9.7 billion over five years with Microsoft, 20% prepaid, and $3.4 billion over five years with Nvidia, which also committed up to $2.1 billion of equity. Both running their course is about $2.6 billion a year (estimated), roughly 13% of the $20.61 billion of assets on the books at September 30, 2026, against a fiscal 2026 revenue yield on assets of under 5%. The September 2026 quarter shows what delivering that takes: $5.15 billion of investing outflow against $230 million of revenue.

Supply-side advantage: cheap power, real but thinner than the margin implies

Supply-side advantage: Moderate. IREN's cost edge comes from where its sites sit and who owns them. It buys power in ERCOT and in BC Hydro's territory, builds its own substations and shells, and holds the land freehold, so it pays no landlord rent and no colocation operator's markup on the electricity behind its machines. Against a miner leasing warehouse space, that is structural. On the same ASIC catalog and the same hash price, the operator with the lower all-in cost per megawatt-hour keeps more of every block reward and can still mine at a bitcoin price where a leased-site rival shuts down. Against another site owner the edge narrows to a cost ranking: the rival buys the same machines at the same price and pays a similar power rate, and the gap is a few dollars per megawatt-hour, not a wall.

The figures show how thin it gets once everything is counted. Gross margin reached 79.8% in the September 2026 quarter, yet in fiscal 2026 roughly $1.54 billion of cost sat below the gross-profit line (estimated) and turned $0.49 billion of gross profit into a $1.05 billion operating loss. GPU depreciation, site labor and stock compensation consume most of what cheap electricity saves.

Power price is what breaks it. ERCOT's wholesale market clears on gas, wind and solar, and a sustained rise in the delivered cost at IREN's own sites compresses the spread dollar for dollar. The adverse outcome to follow is the gap between gross margin and operating margin: as the AI contracts ramp, that gap should narrow if cheap power is doing real work.

Demand-side advantage: two big buyers for five years, then a renegotiation

Demand-side advantage: Moderate. Captivity here is contractual and dated. Microsoft committed $9.7 billion over five years with a fifth prepaid, and Nvidia committed $3.4 billion over the same term plus an equity stake of up to $2.1 billion. The prepayment is the strongest evidence in the file that IREN had pricing power at signing: a buyer that pays before delivery is buying certainty of capacity, and capacity that must be energized by a date is scarce. Mid-build, moving a large commitment to another provider costs a year and the deposit, so within the term the customer stays.

None of that survives the term by itself. Microsoft's engineers have no data format, no retrained workflow and no software lock-in that makes them prefer IREN in 2031. At renewal the buyer can self-build, buy from another neocloud, or shift inference onto its own accelerators, and IREN would be bidding against all three. Concentration cuts the same way: with two counterparties behind most of the announced backlog, losing one would remove the majority of contracted revenue.

The pricing test is whether a rival's discount takes volume. On contracted capacity it cannot, for the term. On the rest it plainly can. AI cloud revenue of $33.6 million in the March 2026 quarter is a long way from the annual pace the contracts imply, and the uncontracted remainder clears at whatever the GPU rental market pays that month.

What breaks it is a delayed ramp or a renewal fought on price rather than scarcity. The number to follow is AI cloud revenue against the roughly $2.6 billion annual pace the two agreements imply.

Economies of scale: the fixed cost is capital, and capital does not queue

Economies of scale: Weak. Site fixed costs are large and front-loaded: substation, transformers, shell, cooling, then the GPUs. Minimum efficient scale runs to several hundred megawatts, so an entrant at 30MW cannot match unit costs. The barrier that creates is capital, and capital is the one input with no queue. Core Scientific, Applied Digital, Cipher Mining and TeraWulf are building against the same demand, and the hyperscalers are the largest builders of all. Scale becomes a barrier only when paired with captivity; here captivity is a five-year contract, so scale protects IREN for exactly that long.

The capital base also is not working hard yet. Revenue against assets yields under 5%, and the build is still accelerating. Until those assets produce, scale is a claim about the future.

Durability: the site holds, the price of its output is what moves

Substitutes threaten the output price before they threaten the site. Google's TPU and Amazon's Trainium give the largest buyers an in-house option for some inference workloads, and that reduces rental demand whether or not those chips are ever sold to third parties. A hyperscaler meeting its own demand internally competes with IREN as effectively as a rival neocloud signing the same customer. On the mining side, the halving schedule cuts the block subsidy on a fixed calendar regardless of what IREN does.

The closest historical parallel is merchant fiber in the late 1990s. Rights of way and conduits were genuinely scarce, capital flooded in, and the owners of the physical plant fared better than the operators who had locked in long-term capacity prices near the top. IREN is both: it owns the plant and it sells the capacity. The plant will still be there in 2031. The price at which the capacity renews is the open question.

Through the contract term, roughly to 2030, the moat looks likely to hold. The sites cannot be replicated faster than that, and the buyers have prepaid. Beyond it, two questions are unresolved. How much of the current capital program is being spent against signed revenue, and how much is speculative capacity built on the expectation that AI demand keeps outrunning supply? And once depreciation on those assets starts, what return on that capital will IREN earn?

Generated by DeepVal AI. Data and models may contain errors or be outdated. For research only; not personalized investment advice.

Continue the comparison

Continue reading the full report

  1. Comparison and summaryHow IREN Ltd compares with its closest rivals, and our moat rating

Free to sign up · 20 research points · unlocking this report uses 10

20 points cover 2 full reports. No card needed.

Research reflects its generation date. Preview the first two modules.

More on IREN Ltd