The AI infrastructure “third-party capacity” trade still has legs

Overview

This report discusses “Emerging AI Infra stocks,” namely companies transitioning from Bitcoin mining to AI data centers, AI colocation and cloud computing. The core judgment is that although the market is concerned that hyperscalers and AI labs building their own capacity could weaken demand for third-party capacity, miners that already control GW-scale power resources and have delivery experience retain structural value amid lengthening grid-connection timelines, political and regulatory constraints on data center construction, and continued growth in AI computing demand.

Core views

The report’s core views include: first, demand for third-party AI capacity remains resilient because “time to power” and “time to compute” have become scarce capabilities; second, miners’ large contracted orderbooks improve revenue visibility, with industry tracked orders reaching approximately 7.5GW and more than $150Bn in multiyear contracts; third, recent transactions show longer lease terms, improved yields and greater use of triple net structures, while project financing conditions are also improving; fourth, execution capability is becoming the key differentiator, and management teams that can deliver on time, maintain customer relationships and continue expanding their power pipelines will receive greater recognition.

Analysis framework

The report combines industry deal tracking, transaction-level contract economics analysis, project delivery progress monitoring, power-capacity pipeline assessment and comparative ratings analysis of covered companies. The focus is not the traditional mining-coin cycle, but rather the orderbook, ARR, NOI, IRR, capex per IT MW, contract duration and customer quality following miners’ conversion of power assets into AI colocation, HPC and cloud-computing infrastructure.

Methodology notes

• Industry supply-demand analysisTime-to-power / time-to-compute framework

  Measures AI infrastructure scarcity through the speed of grid connection and delivery of available computing capacity.

  The report believes that when grid-connection timelines in key data center markets can extend to approximately four years and regulatory and political resistance increases, miners with existing power, sites and construction experience can shorten the time required for customers to obtain computing capacity.

• Project economics analysisAI colocation deal economics

  Assesses AI colocation project quality using lease yields, NOI margin, capex per IT MW and unlevered IRR.

  Recent AI colocation transactions generally indicate capex of $9-12Mn/IT MW. Some triple net leases can convert nearly all revenue into NOI, and the report estimates unlevered IRRs of approximately 8-13% for different transactions.

• Orderbook and revenue visibilityContracted orderbook / ARR ramp

  Assesses future revenue certainty through contracted capacity, total contract value and annual recurring revenue.

  The report closely tracks miners’ contracted AI capacity, multiyear contract value, IREN’s CY26e ARR target and the mitigating effect of customer prepayments on financing and capex risks.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• WULF

  Core beneficiary of AI colocation and power pipeline

  Strengths

  Signed an approximately $19Bn contract with Anthropic, with an AI orderbook of approximately $27Bn; controls approximately 3.6GW of power assets, with sites across key markets including Kentucky, Maryland and New York.

  Weaknesses

  The Anthropic transaction has not disclosed its complete margin structure and capex details; long-term delivery remains to be validated.

  Comparison

  The WULF-Anthropic transaction is approximately $2.4Mn per IT MW, above the industry average of approximately $1.8Mn per IT MW and above the economics of CLSK’s and HUT’s recent transactions.

  Risks

  Project delivery delays, capex increases, changes in financing conditions, customer concentration and state-level regulatory changes.

• CLSK

  New entrant into AI colocation from Bitcoin mining

  Strengths

  Its first AI colocation contract totals 175 IT MW and $6.6Bn. The 20-year triple net structure can generate a nearly 100% NOI margin, while preserving further commercialization opportunities for its Texas portfolio.

  Weaknesses

  Pricing for the first contract is below the WULF-Anthropic transaction, and its execution track record still needs to be established.

  Comparison

  CLSK generates approximately $1.9Mn per IT MW, below WULF’s approximately $2.4Mn per IT MW, but the report believes economics have room to improve as its execution track record develops.

  Risks

  First AI colocation delivery risk, project financing execution risk, and regulatory and construction risks at the Georgia and Texas sites.

• HUT

  AI colocation repeat-contract and customer-relationship validation play

  Strengths

  The second Beacon Point lease doubles contracted capacity to 704 IT MW; the triple net structure and renewal with the same highly investment-grade customer demonstrate customer acceptance.

  Weaknesses

  The report does not cover HUT, so investment-rating information is limited; yields are below those of the WULF-Anthropic transaction.

  Comparison

  The HUT phase-2 transaction is approximately $1.86Mn per IT MW, below WULF’s approximately $2.4Mn, but the repeat contract strengthens execution credibility.

  Risks

  Large data center construction execution, debt financing, customer concentration and risks related to AEP Texas interconnection.

• IREN

  Beneficiary of vertically integrated neocloud and AI cloud computing

  Strengths

  Raised its CY26e ARR target to above $4Bn, approximately 85% of which is contracted; customers include Microsoft, NVIDIA, Perplexity, Fluidstack and Figure AI; its global 5.8GW power portfolio provides room for long-term expansion.

  Weaknesses

  Approximately 15% of CY26e ARR remains uncontracted; rapidly expanding cloud capacity from 3MW to 480MW presents execution challenges.

  Comparison

  IREN is more focused on an AI cloud/managed cloud model rather than pure colocation; GPU capex payback is approximately 2.6 years, and customer prepayments cover approximately 45% of related GPU capex.

  Risks

  GPU procurement and deployment, customer renewals, financing costs for short-term enterprise contracts, declines in cloud-service pricing and cross-regional project execution risks.

• Emerging AI Infra sector

  Third-party capacity theme amid constrained AI computing supply

  Strengths

  An approximately 30GW planned power portfolio, approximately 7.5GW of contracted capacity, more than $150Bn in contract value and improving lease structures support the sector investment theme.

  Weaknesses

  Company business models remain in transition from mining to data center infrastructure, with substantial differences in project quality and execution capability.

  Comparison

  Compared with traditional new-build data center developers, miners benefit from existing power, brownfield sites and faster grid connection; compared with hyperscaler self-build, third-party capacity provides faster external supply.

  Risks

  Substitution by hyperscaler self-build, political and regulatory constraints, grid bottlenecks, capex inflation, tightening financing markets and customer concentration.

Key data

• Industry contracted capacityApproximately 7.5GWThe scale of power capacity contracted by miners with hyperscalers, neoclouds and AI chipmakers over the past two years.
• Industry multiyear contract valueMore than $150BnThe report states that miners’ total contract value has exceeded $150Bn; under another definition, AI colocation orderbook is approximately $135Bn, excluding expansion options.
• New AI colocation contracts in July1.2GW gross power / $35Bn TCVMiners signed three AI colocation leases in July, showing that deal momentum remains strong.
• WULF-Anthropic contract401 IT MW / approximately $19Bn / 20 yearsLocated at the Kentucky Justified Data campus, with average ARR of approximately $950Mn; initial delivery is expected in H2’27e and full delivery in early 2028.
• CLSK’s first AI colocation contract175 IT MW / $6.6Bn / 20 yearsThe Sandersville, Georgia project has a triple net structure, average ARR of approximately $330Mn and approximately $1.9Mn per IT MW.
• HUT’s additional Beacon Point lease352 IT MW / $9.8Bn / 15 yearsAn additional lease with the same highly investment-grade tenant increases Beacon Point’s contracted capacity to 704 IT MW.
• IREN CY26e ARR targetAbove $4BnRaised from $3.7Bn; approximately 85% is contracted and approximately 15% remains uncontracted.
• IREN global power portfolio5.8GWIncludes projects in the United States, Canada, Spain and Australia; the current $4Bn ARR target uses only approximately 8% of the total power portfolio.

Impact & implications

The investment implication is that the valuation of former miners transitioning to AI infrastructure should not be based solely on near-term mining or cryptocurrency-price exposure. Their power assets, time-to-power advantage, contracted revenue visibility and AI customer relationships should be revalued. If orders continue to materialize, financing costs improve and delivery proceeds on schedule, companies such as WULF, IREN, CIFR, CORZ, CLSK and RIOT may continue to benefit. However, dispersion will increase, and the market will place greater emphasis on management execution, customer quality, project financing and regulatory risks.

Risks

• Hyperscalers and AI labs building their own capacity could weaken demand for third-party capacity.
• Data centers in some US states face constraints related to power prices, taxpayer protection, environmental issues, water resources, climate, tax incentives, subsidy reviews and planning and zoning.
• Lengthening grid-connection timelines could both reinforce scarcity and cause project delivery delays.
• Large-scale AI colocation projects face risks from capex overruns, long-lead equipment supply, labor and construction management.
• Long-term leases improve revenue visibility but may also limit future rent growth.
• Customer concentration, tenant credit, project financing costs and changes in spreads can affect equity IRR.
• Miners transitioning to AI infrastructure still need to consistently prove on-time delivery and the durability of customer relationships.

What to watch

• Progress on WULF’s initial capacity delivery to Anthropic and the target for full delivery in early 2028.
• Milestone deliveries by WULF, CORZ and CIFR in H2’26, including contracts related to Fluidstack, CoreWeave and AWS.
• Financing announcements, construction progress and further commercialization of CLSK’s Texas portfolio for its first AI colocation project.
• Energization, equipment procurement and delivery milestones for HUT Beacon Point Phase 1 and Phase 2.
• Delivery of IREN Horizon 1 in Q3’26, delivery of Horizons 2-4 by the end of CY26, and the ramp of the NVIDIA contract in early 2027.
• Whether the industry continues to maintain the pace of new deals each week, and the mix of new-customer versus repeat-customer contracts.
• Whether triple net structures, capex per IT MW, NOI margin, unlevered IRR and project debt spreads continue to improve.
• Changes in state-level US data center regulation, tax-incentive reviews, environmental constraints and grid-queue timelines.