Author: Zen, PANews
How much can a company with an annual revenue of $707 million spend in a year? The answer given by IREN is at least $4.7 billion.
On August 27, the AI Cloud company IREN, which transformed from a Bitcoin mining enterprise, announced its financial report for fiscal year 2026. Its annual revenue grew to $707 million, but net cash outflow from investment activities had reached $4.723 billion, more than three times that of the previous fiscal year.
In addition to the $4.7 billion already spent, as of June 30, IREN also had $13.81 billion in capital commitments, mainly related to AI hardware and data center construction. A year ago, this figure was only $369 million.
Apart from GPUs, IREN spent $148 million to acquire data center developer Nostrum and secured cloud software company Mirantis for approximately $544 million. In June of this year, it even placed its name on the Golden State Warriors' jerseys, with collaboration fees averaging over $50 million a year.
From a Bitcoin mining enterprise to AI Cloud, IREN is clearly not opting for a gradual transformation, but is instead trading capital for time, trying to complete its identity switch before the computing power window narrows.
GPUs haven't returned their cost, and the next generation is already on the way
In IREN's AI gamble, it is undoubtedly the GPUs that have consumed a large amount of capital first.
In November 2025, to fulfill a five-year $9.7 billion AI Cloud contract with Microsoft, IREN signed a procurement agreement worth about $5.8 billion with Dell, primarily to purchase NVIDIA GB300 NVL72 systems and servers, storage, networks, and other supporting equipment. These computing powers will be deployed in the Horizon 1 to 4 at the Childress site in Texas, with a total IT load reaching 200MW. Microsoft will pay a 20% advance for each batch as per the contract.
This bulk procurement has not yet been completed, and IREN has already started to place orders for the next batch.
This March, it ordered over 50,000 NVIDIA B300 GPUs, planning to deploy them in Mackenzie, Canada, and Childress, USA, with Dell orders totaling approximately $2.3 billion and $1.2 billion, respectively, a total of $3.5 billion.
In May, NVIDIA also became a major client of IREN, with the two parties signing a five-year $3.4 billion AI Cloud contract, while also collaborating on AI infrastructure of up to 5GW. To fulfill this contract, IREN then signed an order for Blackwell systems with Dell worth approximately $1.6 billion. NVIDIA also received the right to purchase up to 30 million shares of IREN stock at $70 per share over the next five years, corresponding to a potential investment amount of up to $2.1 billion, and the exercise of these rights is linked to the scale of GPU deployment by IREN.
Just the three batches of disclosed Dell procurement agreements have a nominal value of about $10.9 billion. Of course, IREN's massive GPU orders are not a one-time capital expenditure that can be "paid in full" at once, and a large part has already been included in its disclosed capital commitments.
As of the end of June this year, the installed or ordered products span NVIDIA H100, H200, B200, B300, GB300 as well as AMD MI350X, with the latest procurement plans even extending to the next-generation Vera Rubin system, the VR200. Hopper is still providing services, Blackwell is being deployed on a large scale, and Blackwell Ultra has not yet been fully rolled out, yet IREN has already begun to allocate budget and infrastructure for Rubin.
This means that the previous batch of GPUs has not yet completed depreciation nor generated all revenue, while the next round of capital expenditures has already begun.
This is because the core asset of AI Cloud is rapidly evolving computing devices, and the next generation of GPUs usually means higher computing power, memory capacity, and energy efficiency, all of which will directly impact the cost per unit of computing power. Once competitors deploy newer hardware first, the leasing prices and customer appeal of old GPUs may decline accordingly.
IREN has also identified this technological iteration as a clear risk: GPUs, networks, storage, cluster architecture, and software are continually evolving rapidly, and existing equipment may become obsolete quickly due to the emergence of new technologies.
Thus, from the moment IREN decided to enter this business, it essentially joined a perpetual hardware arms race. While it needs to ensure that the GPUs it has already purchased generate enough cash flow quickly, it also must prepare new funds for the next generation of hardware. This makes IREN's expansion logic even more aggressive.
After buying GPUs, IREN has also started buying software and brands
Merely possessing land, power, data centers, and GPUs is not enough to have a truly competitive AI Cloud company, so IREN continues to place orders.
In June, IREN acquired Spanish data center developer Nostrum for approximately $148 million, of which about $94.8 million was paid in cash. What Nostrum brings is not just ready-made AI clients, but data center development, engineering, and construction teams, as well as the local development capabilities IREN needs to enter Europe.
In August, it completed the acquisition of Mirantis for approximately $544 million, with a significant portion paid through the issuance of IREN stock, as well as around $40 million in cash, restricted stock, and other considerations.
Mirantis has long operated Kubernetes, cloud infrastructure, and enterprise cloud services. After the acquisition, IREN gained not merely a software product but a set of capabilities for managing GPU clusters, deploying client environments, orchestrating AI workloads, monitoring resource usage, and providing enterprise support. IREN even directly broke down its AI Cloud into three layers in the latest 10-K: Data Centers, Compute, and Software. **Mirantis fills in the last piece.
IREN's original assets only included power, land, and mining sites, and later continuously stacked data centers, GPUs, networks, storage, to finally cloud software and enterprise services. It is gradually piecing together a vertically integrated AI Cloud company from a Bitcoin mining enterprise.
If the aforementioned large expenditures still seem logical, what truly confounds the outside world and IREN shareholders is the spending on the NBA team Golden State Warriors.
On June 25, IREN signed a multi-year global partnership with the Golden State Warriors. Starting from the 2026-27 season, IREN will become the official AI Cloud partner and replace Rakuten as the jersey sponsor, with the brand logo appearing directly on Curry's left chest. Additionally, the partnership also includes the WNBA's Golden State Warriors team, the G League's Santa Cruz Warriors, and the Warriors' home arena, the Chase Center.
According to Sportico, this contract is worth over $50 million annually on average, setting a record for sponsorship in North American professional sports teams. Meanwhile, IREN's entire marketing expenditure for fiscal year 2026 was only $2.027 million, a mere $2.88 million in the previous fiscal year. Spending on just the Warriors for future years could potentially exceed IREN's entire annual marketing expenditures previously.
Some investors link this contract to IREN's ongoing use of ATM financing and issuance of convertible bonds, questioning why a company still in a heavy capital expansion phase would pay such costly sports marketing fees. Particularly because professional sports sponsorships are often used by businesses that need to continually reach massive consumers, where a jersey advertisement can directly translate into brand awareness and consumer decisions.
However, IREN is not a To C brand; its clientele consists of a very niche group of top decision-makers from AI labs and tech companies. This makes the annual sponsorship of over $50 million seem extraordinarily aggressive: IREN is spending money meant for mass brand exposure, yet the actual audience to be reached is a highly concentrated B2B customer base. Perhaps this expenditure is not enough to affect its billion-dollar data center plans, but such capital allocation does seem rather extravagant.
Nevertheless, IREN's calculation is that what it truly wants to buy may not be the brand recognition of tens of millions of ordinary NBA viewers, but the attention of the San Francisco Bay Area behind the Warriors. IREN believes that the Bay Area is home to many AI startups, tech talent, and potential clients, and the Warriors can help it quickly enter the vision of these individuals. IREN's latest 10-K has officially listed "establishing a global brand" in its strategy, including sponsorships, industry events, and marketing, for customer acquisition and strengthening relationships with tech partners.
Behind this is a very real issue brought on by the change in IREN's business model. During the Bitcoin mining era, it required almost no sales, whereas AI Cloud is different. It must negotiate contracts with AI labs, startups, and enterprises beyond Microsoft and NVIDIA, convincing these corporate decision-makers of cluster stability, securing acceptance of prices and service terms from procurement departments, and proving itself in a market already filled with AWS, Azure, Google Cloud, CoreWeave, and a host of new cloud vendors.
Thus, in a sense, IREN is now spending heavily on the issue of "who we are." Whether spending over $50 million a year for exposure in the manner of a consumer brand is akin to "spending gold to buy bones" is another matter.
Why does IREN dare to spend so much?
After reviewing IREN’s series of big investments, a question naturally arises: Where does its money come from? It should be noted that in fiscal year 2026, its net loss reached $703 million.
In fact, what truly supports IREN's expansion is a more aggressive capital structure: first locking in future revenue and then turning that future revenue into today's money to buy GPUs.
Among these, the collaboration between Microsoft and IREN is the most typical example. The $9.7 billion contract requires Microsoft to pay 20% of the contract amount in advance for each batch of computing power. This money enters IREN's account first, but it cannot be recognized as current income until the GPUs are actually delivered and services commence; it can only form deferred revenue first.
This is also the reason why IREN has a net loss of $703 million in fiscal year 2026 yet generated $2.1 billion in net operating cash inflow. The AI Cloud client prepayments have increased deferred revenue by about $1.842 billion. Additionally, the company's asset impairments of approximately $639 million and $418 million in depreciation and amortization are considered non-cash expenses; they depress profits but do not generate equivalent cash outflow during the same period. In simple terms, a considerable portion of the money IREN has not yet made in the income statement has already been sitting in its bank account ahead of time.
Aside from operating income, much more comes from financial institutions. IREN disclosed in its latest financial report that its Microsoft project has already secured $3.6 billion in investment-grade GPU financing, with a weighted cost of around 6%; the client prepayments combined with this financing can cover about 96% of the related GPU capital expenditures.
For other AI clients, the company has obtained $2.8 billion in GPU financing, of which $2.4 billion is led by Blue Owl and PIMCO-related investors, with a fixed rate of 9%, covering about 90% of the Mackenzie project's related GPU capital expenditures. Meanwhile, IREN states that recent client prepayments generally represent about 45% to 55% of the corresponding GPU capital expenditures.
It can be seen that the most crucial element of IREN's capital play is GPUs and the long-term computing power contracts behind them, which themselves have begun to become financing assets. Financial institutions see the specific NVIDIA GPUs, specific data centers, and the expected cash flows from contracts with clients like Microsoft over the next three to five years. As such, IREN can build separate financing structures around a batch of devices without wholly relying on its own profits and cash.
This has led to the formation of a capital flywheel: clients first sign contracts and provide prepayments; IREN procures GPUs based on these contracts; financial institutions then provide financing based on the GPUs and contracts; the GPUs go live and produce long-term leasing and Cloud revenue; new contracts can support the next round of procurement.
Furthermore, IREN has extensively utilized common stock, convertible bonds, and other financing tools over the past year, and as of the end of June, the company's principal debt had risen from $990 million in the previous fiscal year to about $7.71 billion. Its 10-K also clearly states that future expansions may continue to rely on client prepayments, GPU asset financing, equipment financing, convertible bonds, and equity.
IREN's ability to continuously gain funding from the capital markets and financial institutions hinges on the fact that it already has a batch of long-term AI contracts that can be used for credit.
As of the end of August, the company stated that cash, committed GPU financing, and client prepayments had reached approximately $14 billion. Meanwhile, the capacity for 2026 is basically sold out, with signed contracts corresponding to about $4 billion in annual recurring revenue (Contracted ARR), of which the already online capacity accounts for approximately $1 billion ARR. This $4 billion is certainly not confirmed accounting income, but it signifies that IREN currently has at least a batch of long-term contracts supporting further financing.
Therefore, what IREN is really doing is leveraging client revenue commitments for the next few years to pry loose tens of billions of dollars from financial institutions and capital markets today. This is precisely why IREN has the confidence to spend so much.
The true gamble for IREN is a time window
After explaining IREN's financing logic, its nearly aggressive capital spending over the past year leaves just one last question: Why the rush?
Currently, the most scarce AI infrastructure resources are not limited to GPUs. Large-scale power supplies that have already come online, rapidly deployable data center parks, and the latest computing powers that can be quickly delivered after customer demand arises all determine whether an AI Cloud company can secure orders. IREN happens to have inherited two of the most difficult-to-replicate resources from the Bitcoin mining era—power and data center land.
The problem is that this first-mover advantage has obvious timeliness. On one hand, companies like Microsoft, Meta, Google, as well as CoreWeave, Nebius, Crusoe, and others are all rapidly building new AI data centers, with more power resources being redeveloped.
On the other hand, the supply of GPUs continues to expand, the hardware itself is driving all players to accelerate; for AI Cloud companies, as the GPUs at hand continue to depreciate, competitors may have already begun delivering products with better performance and energy efficiency. This makes the scarcity premium that mining enterprises once enjoyed likely to vanish quickly.
Consequently, the time-to-compute which IREN repeatedly emphasizes is at the core of this competition: when clients need computing power, whoever can quickly combine electricity, data centers, and GPUs into a truly operational model cluster will have an easier time securing contracts for the next few years.
This is also why IREN seems to be using capital to buy time. The urgency of this transformation arises from IREN's belief that it does not have much time left. From the costs the company is incurring, it is clear that management has already made a choice.
In fiscal year 2026, IREN still generated $578 million in revenue from Bitcoin mining, whereas AI Cloud revenue was only $129 million. In other words, looking at the annual revenue structure, mining remains the company's primary cash source at present. However, by the fourth quarter, the two curves have crossed for the first time: AI Cloud's quarterly revenue reached $70.5 million, while Bitcoin mining dropped to $66.7 million.
At the same time, IREN is actively phasing out many mining assets that still have book value. In fiscal year 2026, the company confirmed impairments of about $639 million, a significant portion of which comes from Bitcoin mining machines and data center equipment that were exited early for AI transformation. IREN has even proposed to complete the transition of existing data centers from Bitcoin mining to AI Cloud by December 31, 2026.
This also clarifies IREN's gamble; it is actually betting on three speeds simultaneously: AI computing demand grows fast enough to absorb the continuously increasing new supply; GPUs generate cash fast enough to outpace their own depreciation and financing costs; and IREN completes its business transformation quickly enough to establish a foothold before the power advantages held by mining enterprises are caught up by other AI infrastructure companies.
Thus, the greatest risks and opportunities now stem from speed. The faster it completes its transformation, the more likely today's massive capital expenditures will convert into future revenue. If it slows down, GPU depreciation, financing costs, and new competitors will catch up at the same speed.
IREN's management has recognized one thing: the AI infrastructure table is quickly filling up, and if they don't spend money now, they may not even have the qualifications or opportunities to burn money in the future.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
