If you wait for earnings you're 4-5 months too late. Remember earnings are for the previous full quarter. H2-4 delivery are the real signal.

H1 revenue will partially show in Nov earnings because it only covers last quarter (Jul-Sep) and H1 came online in August.

We should only expect to see partial revenue for H2-4 at Feb earnings (covering Oct-Dec), then a full quarter for all 4 Horizons at May earnings.

So the guided $4b ARR by end of 2026 will only be partially prorated for this quarter in the Feb earnings report, and then should fully apply for the Jan-Mar quarter, which will show in the May earnings report.

That doesn't mean they're not earning it, it just means that earnings reports are on a very long delay after-the-fact. The information in them is up to 4.5 months old.

SUMMARY:

2026 to date:

• IREN had a significant expansion of its AI customer base, including NVIDIA internal workloads, Perplexity, Cohere, Prometheus, Figure AI, Higgsfield AI and an unnamed leading frontier AI lab, alongside renewals and expansions with existing customers.
• Horizon 1 was delivered to and accepted by Microsoft, with the remaining Horizon phases based on the same design.
• IREN achieved NVIDIA Exemplar Cloud status for both B300 and GB300 deployments.
• The company expanded its software and European development capabilities through acquisitions and now reports more than 5 GW of secured power, backed by executed grid-connection agreements.

Pricing

• IREN's most recent three-year AI contracts have been signed at approximately $23 million per MW annually, with current discussions occurring above that level.
• Those contracts have included significant customer prepayments, with IREN describing GPU-capex payback periods of less than two years.
• Short-term capacity of around three to six months is attracting approximately $40–50 million per MW, particularly where capacity is available for immediate deployment.
• IREN continues to see upward pricing pressure as demand exceeds available AI infrastructure.

Power Allocation

• IREN is not allocating capacity based solely on the highest headline price. Customer growth prospects, strategic fit, contract duration and prepayments are also being considered.
• The company is deliberately seeking a diversified base of direct AI customers rather than allowing a small number of hyperscalers to control the end-user relationship.
• Microsoft has no right of first refusal or exclusivity over future Childress capacity. IREN remains free to allocate additional capacity between Microsoft, other customers, short-term contracts, on-demand services and inference workloads.

Revenue

• IREN has confirmed $4 billion of contracted ARR for its 2026 capacity, expected to be operating at that level exiting calendar 2026.
• Approximately $1 billion of ARR is currently operating.
• The NVIDIA internal-workload contract is expected to contribute approximately $700 million ARR from early 2027.
• IREN expects approximately 500 MW of IT load to come online during 2027 but has not provided a specific 2027 ARR target.
• Kent noted that the $4 billion 2026 ARR target represents only around 10% of its secured power capacity, and that some of that revenue was contracted at lower historical pricing.

Capex

• IREN's $25–30 billion capex programme covers delivery of its 2026 contracted capacity, including Microsoft, as well as GPU and data-centre spending through June 2027 and early expenditure on later projects.
• As a rough guide, IREN expects around one-third of the expenditure to go toward data centres and two-thirds toward GPUs.
• GPU payment terms are generally around 30 days after delivery, meaning much of the GPU expenditure occurs close to when the equipment begins generating revenue.
• Kent stated that $14 billion is already funded through existing cash and committed debt financing and is targeting another $8 billion through GPU debt financing and customer prepayments.
• Remaining funding may come from data-centre financing, convertible notes, secured or unsecured corporate debt and the ATM.

ATM

• Approximately $3.5 billion of capacity remains under IREN's existing ATM programme.
• Kent said IREN will continue comparing the ATM with alternative financing sources depending on market conditions.
• The company expects the vast majority of its remaining funding requirements to be raised through instruments other than the ATM.

Spain

• Spain sits within IREN's 2028 expansion pipeline.
• Its principal Spanish project currently has 300 MW of secured capacity and is being prioritised as the company's flagship development in the country.
• IREN sees European demand being supported by hyperscalers and AI labs seeking additional capacity, sovereign-AI requirements, European data-sovereignty rules and the region's growing AI ecosystem.
• IREN's European development acquisition also brought a roughly 60-person development, project-management and construction-management team, providing a base for further European expansion.

Australia

• Australia also forms part of IREN's 2028 expansion pipeline.
• The announced 800 MW Bundy project is only one of several development projects IREN has in Australia.
• Kent said IREN is seeing demand from hyperscalers and large frontier AI labs, as well as Asian customers that value Australia as a geopolitically stable location for infrastructure.

ClusterMAX Rating

• Kent said ClusterMAX evaluates managed Slurm and Kubernetes clusters, while IREN historically has not had a live managed-services environment that it could provide for testing.
• Managed-services clusters are now under development, including contracted infrastructure for NVIDIA's internal R&D workloads, and IREN expects to have an environment available for this type of testing in future.
• Prince George was originally a Bitcoin-mining facility and has been progressively retrofitted for AI. IREN brought GPU capacity online before backup-power redundancy was complete because customers wanted access to the GPUs as quickly as possible; customers were aware of this and wanted the capacity anyway and commissioning of the additional redundancy is underway.
• Kent said Prince George does have redundant fibre, although British Columbia wildfires temporarily affected some redundant routes.
• Newer developments including Horizon and Sweetwater are being designed from the outset with full power, mechanical and cooling redundancy and concurrent maintainability, rather than relying on the retrofit approach used at Prince George.

Courtesy and credit to the amazing Agrippa

(https://x.com/Agrippa_Inv)

----

$IREN Q2 (FYQ4): My Biggest Takeaway

Over the past couple of days, I’ve taken the time to rewatch the earnings call several times, go through the 10-K, and dissect every bit of new information we received.

Was it a perfect earnings result / call? No.

There are several things that temper my near-term expectations. But on the other hand, I believe the long-term outlook has never looked stronger.

The one thing that really stuck with me was the significantly improved funding picture.

Ever since $IREN leaned heavily into its cloud strategy, instead of becoming a colocation player, one of the most prominent bear cases has been:

How do they finance such a capital-intensive strategy?

$IREN's model is arguably the most capital-intensive in the sector, and in retrospect, I think that may have been one of the key reasons the stock was punished this year.

The likes of $CRWV and $NBIS lean heavily on colocation partners as they expand, meaning most of the capex they have to fund directly is concentrated around GPUs.

Meanwhile, pure-play colocation companies like $CIFR, $WULF and $APLD sit on the other side of the equation. When your customers are signing long-duration leases, it’s relatively straightforward to secure financing for a large portion of your infrastructure capex.

$IREN's situation is fundamentally more complex.

They pride themselves on owning both the GPUs and the data center infrastructure those GPUs sit inside.

I’d argue the benefits of that strategy are pretty obvious.

You avoid paying colocation fees, improve your underlying economics, retain greater control over the infrastructure, and own valuable long-duration assets rather than taking on long-duration lease liabilities.

That can also improve the long-term credit profile. Just look at $IREN's 1% coupon on seven-year notes versus $NBIS's 4.5% on notes of equal duration.

That’s all great, but the main drawback is that the strategy is MUCH more capital-intensive.

Funding the GPUs isn’t really the problem, particularly in today’s market environment. Customers are willing to make substantial prepayments, while lenders are increasingly comfortable underwriting GPU financing against contracted cash flows.

The harder part has always been funding the data center infrastructure itself without a 10-20 year lease backed by an investment-grade counterparty.

Until recently, $IREN's primary solution was effectively the public markets; raise equity and use dilution to fund a meaningful portion of the infrastructure buildout.

For a long time, I argued that this was a necessary evil in exchange for the long-term benefits of being your own landlord and enjoying the superior margins, asset ownership and control that come with it.

But clearly, the market didn’t see it that way this year. I also know plenty of $IREN bulls who have been highly skeptical of that approach.

Well, after this earnings call, I’ve changed my stance on the funding question. And that brings me to my biggest takeaway:

$IREN has found ways to pull forward the cash flows from cloud contracts and use that capital to help fund the underlying data center buildout.

There appear to be 2 primary mechanisms.

For shorter-duration, non-investment-grade enterprise contracts, $IREN can receive very large customer prepayments equivalent to 45-55% of GPU capex, or around 25-30% of total contract value upfront.

The key is that $IREN can then layer financing of ~90% of GPU capex on top of that.

The result is effectively excess upfront liquidity equivalent to ~35-45% of GPU capex, which can then be redeployed toward funding the data center infrastructure.

This degree of “over-financing” is only really possible because of the combination of very large customer prepayments and extremely strong underlying contract economics, which $IREN now guides at $20-$25M per MW.

For longer-duration, lower-yielding contracts with investment-grade customers, $IREN can pull cash flows forward in a different way: Post-stabilization data center financing.

Once a facility is fully operational, the customer is deployed and billing has commenced, the project is substantially de-risked. At that point, $IREN can raise debt against the stabilized data center asset and its associated contracted cash flows, effectively recycling capital that was originally tied up in construction.

We still don’t know the exact terms $IREN will receive on these financings, but I’d expect the debt to be structured around the duration and cash flows of the underlying customer agreement.

This approach becomes particularly powerful for more capex-intensive facilities because there is more invested capital that can potentially be pulled out once the asset reaches stabilization.

In essence, both approaches allow $IREN to pull forward contracted future cash flows and immediately reinvest that capital into expanding its data center footprint.

That directly reduces the amount of equity capital required to fund the ramp.

There is obviously quite a bit of nuance here, and there are drawbacks and risks that need to be considered. But if executed prudently, the long-term implications are extremely significant.

Less dilution today means materially greater earnings per share economics later, once $IREN reaches sufficient scale and operating leverage to generate substantial profits.

And this is why I think much of the backlash surrounding the earnings call, particularly around the headline $25-$30B capex guidance, misses one of the most important changes in $IREN's strategy.

The funding model itself is evolving, with $IREN increasingly finding ways to make each contracted deployment help finance the next one.

That is a massive development, and I think many retail investors are overlooking it.

Given this fundamental change in the funding model, I firmly believe $IREN is a more valuable company today than it was 4 days ago.

I’m currently writing a deep dive on all the new information and nuances from the earnings release, including this financing strategy, but also several of the negative aspects and near-term headwinds I see.

This is the first of four tranches in a 5 year, $10 billion contract. The remaining 3 tranches are due by end of year.

Why is H1 acceptance so important? Because it proves their execution and their moat.

While many data centre buildouts are facing supply constraints, long grid approvals, governor audits and heavy community opposition (and some are outright being cancelled), IREN's 700MW Childress and 1.4GW Sweetwater 1 facilities are already years past these hurdles.

They also have Nvidia onboard. Sweetwater 1 has already been designated as Nvidia's DSX reference implementation and Nvidia have already contracted 60MW from IREN for its own internal AI workload.

IREN fully own the entire vertical datacentre stack from GPUs and buildings right down to construction pipelines, land and grid power contracts that were approved years ago. No other neocloud has this, and it means they (a) their bottom line is unaffected by land, power and building fees and (b) have complete control over their design and construcion pipelines and can optimize them.

The Childress and Sweetwater facilities are situated in remote communities at a crossroads between major fibre backbones and the renewable ERCOT grid where their usual approach has been to relieve the grid of surplus load during periods of oversupply.

Their debt structure is generally considered to be very healthy.

Yes, they do have an outstanding ATM and some may not like this, but they need to fund a massive pipeline and they are generally considered to have a very favourable debt structure and asymmetric upside if you believe in the AI buildout.

DYOR.

This is the first of four tranches from the $10B 5 year contract. The remaining 3 Horizon facilities are due to be completed by end of year.

I've been skeptical over the past few weeks and everyone jumped on the subcontractor facebook pic bandwagon as if "subcontractors finished = handover" which is simply not true. IMO fake hype is about as helpful as fake FUD.

However with both Frans and Jim aligned, I think my last traces of skepticism are gone. Jim is very controlled, methodical and measured with his approach, even if the signals point against his bullishness he'll objectively acknowledge them. Probably the most objective a bull could ever be.

https://x.com/jiahanjimliu/status/2085806131637793040?s=20

https://preview.redd.it/n703a8ajt1ih1.png?width=1312&format=png&auto=webp&s=0d50cdc0615b05b68233b5e3253042a1e98ce5b8

https://preview.redd.it/s908ps0tbchh1.png?width=1117&format=png&auto=webp&s=26af098441236ed3ed2de0aca06d9be7eb63c4a5

IREN Completes Acquisition of Mirantis

IREN Completes Acquisition of Mirantis

Strengthening the Software Layer of its Vertically Integrated AI Cloud Platform

NEW YORK, Aug. 04, 2026 (GLOBE NEWSWIRE) -- IREN Limited (NASDAQ: IREN) (“IREN”) today announced it has completed the acquisition of Mirantis, Inc. (“Mirantis”), a leading provider of cloud software and services, through the issuance of approximately 12.6m ordinary shares, fixed at signing, plus cash, restricted stock units and other consideration of approximately $40m as of closing.

The acquisition deepens IREN’s capabilities across AI workload orchestration, monitoring and customer support, further strengthening its vertically integrated AI Cloud platform spanning owned and operated data centers, compute and software.

The acquisition also supports IREN’s strategy to serve a large and diverse customer base over time, including hyperscalers, enterprises and AI developers across bare metal and managed cloud services, and has already facilitated several of IREN’s announced and prospective AI Cloud contracts.

Mirantis brings deep software engineering and technical expertise, and a track record of serving more than 1,500 enterprise customers globally. Mirantis is an inaugural partner of the NVIDIA AI Cloud Ready Initiative, and has integrated k0rdent AI with NVIDIA DSX OS software components, supporting current and next-generation NVIDIA architectures. The open-source k0rdent AI platform will continue to be developed and supported for Mirantis’ customers.

The combination brings together IREN’s owned and operated data centers and compute with Mirantis’ flexible, interoperable software layer, giving customers greater choice and control in how they deploy and scale AI workloads.

Daniel Roberts, Co-Founder and Co-CEO of IREN, commented:

“From the beginning our view has been simple: own the land and power, build the data centers, deliver the compute. Mirantis adds the software layer on top, turning infrastructure into a platform. That’s what lets us serve everyone from hyperscalers running bare metal to enterprises who want fully managed AI cloud.”

Alex Freedland, Founder and CEO of Mirantis, commented:

“For more than a decade, Mirantis has helped enterprises deploy and operate mission-critical cloud infrastructure software, and that commitment to our customers remains unchanged. Becoming part of IREN gives us the opportunity to bring those capabilities to an even larger infrastructure platform, accelerating innovation while continuing to invest in the open and infrastructure-agnostic k0rdent AI platform.”

https://x.com/i/status/2074999495990071692

‣ Horizon 1 handover confirmed 19th July

‣ Commisioning H2

‣ MSFT very happy with IREN so far, very hands off. Great partners.

‣ IREN on call with NVDA basically everyday

‣ Deals signed today are even better than 2 months ago.

‣ Pre payments sweetspot now between 25-40 roughly. Some even offering 100%

‣ IREN GB300 NVIDIA EXEMPLAR Cloud certified. Likely announced next week. Only THREE have this right now.

‣ IREN selected with small number for VERA rubin for testing VERY SOON. Unofficially late EOY mass production. Likely early 2027

‣ H5-H6, Sweetwater 1 Vera Rubin

‣ Mirantis deal closing this month is the aim

‣ Semi conductor shortage micron $MU seeing into 2031