Billion-dollar AI contracts. Can a former Bitcoin miner profit from the AI boom?
Artificial intelligence has transformed over the past two years from a technological promise into the most capital-intensive sector on the planet. Training and operating large language models consume computing power in volumes that no one planned for five years ago and that today's data centers handle only with difficulty. Hyperscalers such as Microsoft, Google, and Meta, as well as companies focused exclusively on AI like OpenAI or Anthropic, compete for every available graphics processor and every megawatt of free capacity.
Key points
AI Cloud Services revenue grew 94.2 percent quarter-over-quarter in the third fiscal quarter of 2026, while bitcoin mining revenue fell 33.6 percent in the same period.
Contracted annual recurring revenue (ARR) for year-end 2026 exceeds $4 billion, of which 85 percent is already covered by signed contracts with Microsoft, NVIDIA, and other customers.
The number of shares outstanding has grown seventeenfold since the IPO in November 2021 to 357.4 million, with the majority of the dilution occurring in the last twelve months to finance the expansion.
Free cash flow was negative by $2.3 billion over the last twelve months, consistent with the massive construction phase of the Sweetwater and Childress data centers.
A proprietary DCF model shows intrinsic value around $40 per share, well below the average analyst consensus of $75 to $82.
The problem is not just the chips themselves. The most valuable commodity of the AI revolution is paradoxically becoming electricity. New data centers need hundreds of megawatts of stable power, access to the transmission grid, and land where such infrastructure can be built at all. According to analyst estimates, demand for computing capacity exceeds available supply by tens of percent, and this imbalance is likely to widen in the coming years because building new power plants and transmission lines takes years, while demand for GPU clusters grows quarter after quarter.
In this environment, companies that already own what the AI industry most desires—energy, land, and experience in building large-scale data centers—gain an unexpected competitive advantage. This includes players that originally built their infrastructure for a completely different purpose.
That is precisely the case of IREN Limited $IREN, an Australian company founded in 2018 by brothers Daniel and Will Roberts, which for years built a reputation as one of the world's most efficient bitcoin miners. Today, IREN is transforming into a vertically integrated provider of AI cloud infrastructure with customers that include Microsoft $MSFT and Nvidia $NVDA.
The key question is whether this is a well-thought-out strategic transformation that creates long-term value for shareholders, or just an opportunistic exploitation of the current hype around
How the company makes money
Bitcoin mining: a high-margin business with zero predictability
Bitcoin mining works on the principle that specialized devices solve cryptographic puzzles in exchange for newly mined bitcoins and transaction fees. Thanks to cheap renewable energy and one of the most efficient machine fleets in the industry, IREN achieved total costs per mined bitcoin of around $34,000 to $41,000, placing it among the lowest-cost producers in the market. With a bitcoin price around $64,000, where the cryptocurrency was at the beginning of August 2026, this means a solid margin.
The problem with this business lies elsewhere. Mining revenue depends on three variables the company cannot control:
the price of bitcoin, i.e., the market value of the cryptocurrency at any given time,
network difficulty, which automatically rises as competitive computing power increases,
the block reward size, which halves every four years through the halving mechanism.
In October 2025, bitcoin reached an all-time high of $126,198 and has since lost about half its value. The total network hashrate meanwhile rose to approximately 890 EH/s, so even a company with a growing number of machines can see a declining market share. This is exactly what happened to IREN in recent quarters, albeit largely by design.
AI Cloud Services: leasing computing power with long-term contracts
The second segment works completely differently. IREN builds and operates its own data centers optimized for the latest generations of NVIDIA graphics processors, which it leases to AI developers and hyperscalers through multi-year contracts. The customer effectively rents GPU cluster capacity from IREN for months to years in advance, with the contract containing a guaranteed minimum payment regardless of how intensively the capacity is used.
This model is more akin to renting real estate than a traditional cloud business. Once the data center is built and the customer signs the contract, cash flow becomes relatively predictable for several years ahead. For some new contracts, customers also prepay up to 45% of the GPU procurement costs, reducing IREN's need for external financing and confirming their commitment to long-term cooperation.
The two largest contracts to date illustrate this:
Counterparty | Contract value | Duration | Scope of contract |
|---|---|---|---|
Microsoft | $9.7 billion (signed 11/2025) | 5 years, averaging $1.94 billion annually | Dedicated GPU capacity at the Horizon campus in Childress, Texas |
NVIDIA | $3.4 billion | 5 years | Air-cooled Blackwell GPU capacity |
Both contracts together create a revenue base that does not depend on bitcoin's price or mining difficulty.