AI compute power race heats up: who can become the data center winner—IREN, Hut 8, or Core Scientific?

On July 20, 2026, Eastern Time, the capital markets bore witness to a collective celebration belonging to Bitcoin mining companies.

IREN announced that it has signed multi-year cloud service contracts totaling $2.8 billion with multiple AI developers, and raised its 2026 AI cloud business annualized revenue run-rate target from $3.7 billion to more than $4.0 billion. On the same day, Hut 8 said that the Beacon Point data center campus in Nueces County, Texas, completed full commercialization, with its second 15-year lease valued at $9.8 billion. On that day, the two companies’ stock prices rose by 19.57% and 10.37%, respectively. Core Scientific also rose 6.97%, closing at $22.40.

This is not an accidental linkage between sectors. It marks a deep identity restructuring of publicly listed Bitcoin miners in North America—from highly volatile, strong-cycle “mining companies” tightly bound to Bitcoin prices, to repositioning as energy infrastructure operators and AI data center platform providers. According to industry data, the total amount of AI and high-performance computing contracts already announced by publicly listed miners has exceeded $70 billion. Some industry forecasts believe that by the end of 2026, the share of revenue from AI business for publicly listed miners could rise to 70%, from about 30% currently.

The core logic behind this transformation is not complicated: the power capacity, land, cooling systems, grid interconnection approvals, and operational capabilities held by Bitcoin miners are precisely the most scarce and time-consuming resources in building AI data centers. When AI training and inference workloads drive exponential growth in demand for power and infrastructure, miners find that what they have on hand is exactly what large-scale cloud vendors need most.

But not all transformers take the same path. IREN, Hut 8, and Core Scientific represent three distinctly different AI transition routes. Who will win in this round of the compute power race depends on three key variables: power capacity, construction speed, and long-term customer contracts.

Power capacity: the biggest bottleneck for AI data centers is not the GPU, but the power

Global data centers are experiencing an electricity crisis.

Gartner predicts that global data center electricity consumption will reach 565 TWh in 2026, up 26% year over year. Global data center electricity consumption has already reached 68 GW, up 17% year over year, and at least 12 major markets are facing grid bottlenecks. By 2027, AI servers’ electricity usage will officially exceed that of traditional servers. The International Energy Agency defines 2026 as the “bottleneck year”—tech giants are enduring transformer shortages, interconnection queues, and regulatory constraints.

A sharp conflict has emerged between constrained power supply and expanding demand for AI compute capacity. For AI data centers, the scarcest resources are no longer GPUs, but grid interconnection points, substation capacity, and long-term, stable electricity contracts.

This is precisely where the core value of Bitcoin miners lies.

Bitcoin mining itself is an industry that is extremely dependent on electricity. Over many years of operations, miners accumulate large-scale power contracts, substation access, and grid interconnection agreements. These assets may be cost centers when mining economics are under pressure, but when AI compute demand surges, they become non-replicable competitive barriers.

IREN has secured approximately 5 GW of grid-connected power capacity. The company plans to deliver 480 MW of self-built AI cloud compute capacity in 2026—whereas 12 months earlier the figure was only 3 MW. By 2027, IREN aims to expand to 1.2 GW. In addition, IREN completed its acquisition of Spain’s Nostrum Group in June 2026, gaining about 490 MW of grid-connected power capacity on the European mainland.

Hut 8’s Beacon Point campus has 1 GW of utility capacity, ensured through its interconnection agreements with AEP Texas. The total contracted IT capacity of the company’s AI data center portfolio has reached 949 MW, supported by 1,330 MW of power capacity.

Core Scientific shows an even more aggressive expansion blueprint. The company plans to expand its AI infrastructure hosting business from its current rentable capacity of about 520 MW to a total pipeline size of about 3.0 GW. In the first quarter of 2026, Core Scientific completed project bond financing of $3.3 billion, netting $2.9 billion, to strengthen its capital structure and accelerate development across multiple locations.

The gap in power capacity is becoming the first watershed that differentiates how deeply these three companies have transitioned. Core Scientific’s most aggressive 3 GW forward target is the boldest; Hut 8’s 1.33 GW already-contracted power is the most certain; and IREN’s 5 GW of grid-connected capacity provides the greatest forward flexibility.

Construction speed: AI enterprises need rapid deployment, and miners’ “ready assets” are the core advantage

AI enterprises cannot wait. The data center industry expects to invest $750 billion in capital expenditures in 2026, and more than 23 GW of dedicated AI IT capacity is already under construction. But the reality is that in the first quarter of 2026, due to power and equipment issues, more than $130 billion worth of U.S. projects were delayed.

Bitcoin miners’ “ready-made infrastructure” addresses this pain point perfectly.

Miners already have ready land, power access, cooling systems, and operations and maintenance teams. Retrofitting these facilities to convert ASIC mining rigs into GPU compute clusters is far faster than building a brand-new data center from scratch. This is exactly why Bernstein views Core Scientific as a “capital expenditure advantage player”—the company can begin hosting AI workloads without having to spend tens of billions of dollars to build new infrastructure.

IREN chooses the path of building its own AI cloud services. The company is actively procuring high-performance GPUs and plans to deploy 140,000 units by the end of 2026. Its core commercial offering provides enterprise-level AI cloud computing leasing services for both training and inference workloads. The one-year expansion pace from 3 MW to 480 MW reflects IREN’s execution capability in both retrofitting and new construction.

Hut 8 follows an “electricity-first” development model. The company locks in power capacity and sites first, then designs and builds, and ultimately delivers the complete AI factories to tenants in the form of long-term leases. Beacon Point campus was originally designed as a flagship base for Bitcoin mining, and now it has fully pivoted to AI under 15-year long-term lease arrangements. Hut 8 redesigned the campus’s first data hall, using Nvidia AI data center architecture, increasing capacity by 57% with the same land and power configuration.

Core Scientific relies on its partnership with CoreWeave and adopts the hosting model. As of the first quarter of 2026, the company’s hosting business revenue was $77.5 million, with a gross margin of 57%. The company expects to deliver more than 450 MW of billable capacity by late summer 2026, and reach 590 MW—CoreWeave’s full commitment—by early 2027. Full-year 2026 capital expenditures are expected to be about $200 million.

Differences in construction speed essentially reflect choices in business models. IREN pursues revenue elasticity from operating its own cloud services; Hut 8 pursues cash flow certainty from long-term leases; Core Scientific aims to maximize asset return efficiency in the hosting model.

Long-term customer contracts: valuation leap from “miners” to “AI infrastructure operators”

The market is assigning new valuations to “AI infrastructure operators,” not “miners.”

The core basis for this valuation reshaping is long-term customer contracts. Bitcoin mining revenue is highly dependent on coin prices, the network hash rate, and electricity costs—none of which miners can control. In contrast, long-term contracts for AI data centers provide a completely different revenue structure: longer lock-in periods, clear pricing, and predictable cash flows.

VanEck points out that the clearest valuation metric at present is available operational electric power infrastructure: companies with AI leases have valuation multiples more than 10 times those of operational electric power. For miners that own hosted or leased facilities, enterprise value relative to earnings before interest, taxes, depreciation, and amortization (EBITDA) can reach 10 times or even higher. By comparison, miners that remain tightly linked to Bitcoin prices face significant valuation discounts.

IREN’s latest contract structure reflects this logic. The company has signed multi-year cloud service contracts totaling $2.8 billion with leading AI developers, including Microsoft, Nvidia, Perplexity, and Figure AI. In the raised $4.0 billion annualized revenue target, about 85% is already supported by contracts. The contracts also include customer prepayment arrangements that can cover about 45% of GPU capital expenditures, helping ease funding pressure caused by infrastructure expansion.

Hut 8 has achieved even more impressive results on long-term contracts. The two 15-year leases for the Beacon Point campus have a combined basic value of $19.6 billion; if all renewal options are fully exercised, the potential contract value could reach $50.2 billion. The company’s AI data center portfolio has cumulative 15-year base lease values totaling $26.6 billion, with average annual net operating income exceeding $1.75 billion. 100% of contracted capacity is leased to, or supported by, corporate customers rated as investment grade. Hut 8 currently has $26.6 billion in contracted base revenue.

Core Scientific’s contract structure is even more concentrated. The company signed a 12-year hosting agreement with CoreWeave worth more than $10 billion. Bernstein’s analysis shows that the agreement’s average asset return rate over the next five years is 75%, with a cost-benefit rate of 79%. In June 2026, Bernstein raised Core Scientific’s target price from $24 to $32, maintaining a “outperform the market” rating.

However, concentration also means risk. Core Scientific’s reliance on a single customer, CoreWeave, is significantly higher than the diversified customer structures of IREN and Hut 8. CoreWeave’s own debt level and capital intensity also create potential counterparty risk.

In terms of contract structure, Hut 8’s diversified, investment-grade tenant portfolio provides the highest income certainty; IREN’s diversified AI developer customer base provides the most balanced distribution of risk; Core Scientific’s single ultra-large contract offers the highest return per transaction, but concentration risk is also the most prominent.

Three paths, three valuation logics

IREN, Hut 8, and Core Scientific represent three typical routes for Bitcoin miners to transition to AI.

IREN chooses the “AI cloud service provider” path—selling AI cloud computing capacity directly to enterprise customers. The advantages of this model include high revenue elasticity, a diverse customer base, and strong brand independence; the challenges include the need for continuous GPU capital investment and building cloud service operations capabilities.

Hut 8 chooses the “AI data center landlord” path—leasing the full AI data center capacity to investment-grade enterprise customers in the form of long-term leases. The advantages of this model include extremely high revenue certainty, predictable cash flows, and excellent tenant credit quality; the challenges include a long construction period, large upfront capital investment, and relatively fixed lease return rates.

Core Scientific chooses the “AI compute hosting provider” path—providing customized AI compute hosting services to a single ultra-large customer (CoreWeave). The advantages of this model include extremely high asset return efficiency and relatively lower operational complexity; the challenges include high customer concentration risk and uncertainty about contract replicability.

Bernstein’s analysis offers an intuitive reference: Core Scientific’s AI asset return rate reaches 75%, while TeraWulf and Cipher Mining, which adopt the “electricity landlord” model, are only 5% and 4%, respectively. The core reason for this huge gap is not the power capacity itself, but whether long-term, high-value hosting contracts are secured.

But Bernstein also makes it clear that Core Scientific’s 75% ROA largely stems from its unique capital expenditure structure: the company retained already-built infrastructure from its bankruptcy restructuring, meaning it does not need to pour massive capital into new builds. This “capital expenditure advantage” is almost impossible for later entrants to replicate.

Market pricing has already reflected this split. As of the close on July 21, 2026 (Beijing Time), IREN was at $40.20, up 19.57% on the day; Hut 8 was at $100.93, up 10.37% on the day; Core Scientific was at $22.40, up 6.97% on the day. Across the year-to-date period, all three companies’ stock prices have significantly outperformed the spot price of Bitcoin (Bitcoin traded around $65,500 on that day).

Conclusion

The transition of Bitcoin miners to AI data centers is, at its core, the reallocation of “power resources” from one computing use case (PoW mining) to another (AI training and inference). Traditional miners that have power capacity, grid access, land, and cooling facilities are becoming an increasingly important new force in the AI compute infrastructure space.

The core of this competition is not who owns the most mining rigs, but who has: power capacity that can be called on quickly, infrastructure build-out capability that can be deployed rapidly, and the ability to lock in long-term high-quality customer contracts.

With 5 GW of power capacity and a $4.0 billion annualized revenue target, IREN demonstrates scale ambitions; with $19.6 billion in contracted base and 100% investment-grade tenants, Hut 8 demonstrates revenue certainty; with a 75% asset return rate and a 3 GW expansion plan, Core Scientific demonstrates return efficiency.

Each path has its strengths and weaknesses, but they all point to one trend: the market is repricing these companies under the framework of “AI infrastructure operators” rather than “miners.” VanEck notes that the valuation multiple for miners with AI leases is more than 10 times that of operating power. Bernstein expects that the total AI revenue of the covered Bitcoin miners will grow from $1.2 billion in 2026 to $10.7 billion in 2030, achieving a 9x increase.

For investors, the key question in this sector may no longer be “how high the Bitcoin price will go,” but “who can secure the most irreplaceable niche in the long-term build-out of AI infrastructure.” The combination of power capacity, construction speed, and contract quality—these three variables—will determine which of IREN, Hut 8, and Core Scientific ultimately wins in this round of the compute power competition.

FAQ

Q: What is the fundamental driving force behind Bitcoin miners transitioning to AI data centers?

After the Bitcoin halving, mining profitability has remained under sustained pressure, and in the first quarter of 2026 the hash price fell to about $29 per PH/s. Meanwhile, AI compute demand has exploded higher, and the power capacity, land, and cooling facilities held by miners happen to meet the core needs of AI data centers. The total value of AI contracts already signed by publicly listed miners has exceeded $70 billion.

Q: What are the fundamental differences in the business models of IREN, Hut 8, and Core Scientific?

IREN takes the AI cloud services route, selling compute capacity directly to enterprises; Hut 8 takes the long-term leasing route, renting data center capacity to investment-grade tenants; Core Scientific takes the hosting route, providing customized compute hosting for single customers such as CoreWeave. Their revenue certainty, return rates, and customer concentration levels differ.

Q: Why has power become the biggest bottleneck for AI data centers?

AI compute expansion demand sharply conflicts with existing limits on grid capacity. Global data center electricity consumption in 2026 is expected to reach 565 TWh, up 26% year over year. Transformer shortages, interconnection queues, and regulatory constraints make power access a more scarce resource than GPUs.

Q: How is the market repricing these transitioning miners?

The market is pricing them using the framework of “AI infrastructure operators” rather than “miners.” The valuation multiples of companies with AI leases are more than 10 times that of operating power. Bernstein expects that total AI revenue for miners will increase from $1.2 billion in 2026 to $10.7 billion in 2030.

Q: What are the main risks facing the transitions of the three companies?

IREN faces challenges in sustained GPU capital investment and building cloud service operations capability; Hut 8 faces pressure from long construction cycles and heavy upfront investment; Core Scientific faces concentration risk because it is highly dependent on CoreWeave. In addition, VanEck points out that the industry’s AI transition faces a funding gap of about $50 billion.