Everything About AI
NextFin News - Bitcoin miners that spent years monetizing cheap, interruptible electricity are now selling something more valuable: long-term access to powered data centers for artificial-intelligence customers. That pitch has survived the collapse of crypto-mining economics, but the latest AI rout has exposed its contradiction. The same sites that can command multi-billion-dollar AI contracts also inherit the sector’s most expensive risks—long construction schedules, heavy GPU financing, customer concentration and valuations tied to future power demand rather than current cash flow.
Companies including Core Scientific, IREN, Applied Digital and TeraWulf have used grid connections, land and operating expertise developed for Bitcoin mining to pursue a different business model. Their contracts show why investors rewarded the pivot: Core Scientific says its relationship with CoreWeave covers approximately 590 megawatts of critical IT load and about $10.2 billion of contracted revenue over 12-year terms. IREN disclosed an approximately $9.7 billion Microsoft agreement through 2031, with an average five-year term and 20% prepayments before the applicable delivery dates. Applied Digital has disclosed 400 megawatts contracted to CoreWeave at Polaris Forge 1 for about $11 billion of prospective lease revenue over a roughly 15-year base term.
Those figures are not interchangeable with realized revenue. They are commitments that require equipment, financing, construction and delivery. That distinction matters more when investors are selling AI-linked assets. A Bitcoin mine can curtail machines when economics deteriorate; an AI data center must meet a hyperscaler’s performance, cooling and uptime requirements while carrying a much larger capital burden. The market is therefore testing whether the miners’ old advantage—time-to-power—can outweigh their new exposure to the AI financing cycle.
The Pivot Is Real, but “Power Landlord” Is Not Yet a Low-Risk Business
The first-order story is straightforward: AI customers need large quantities of electricity and data-center capacity, while former miners already control sites connected to the grid. The second-order story is less comfortable. A powered site is valuable only if it can be converted into a facility that satisfies a customer’s technical requirements on schedule and at an acceptable cost. Old mining infrastructure solves the earliest bottleneck, not the entire project.
Core Scientific’s filings illustrate both sides. Its February 2025 expansion with CoreWeave added $1.2 billion of contracted revenue at the Denton, Texas, site and lifted total contracted HPC infrastructure with that customer to approximately 590 megawatts. The company also said it would fund $104 million of capital expenditure for the 70-megawatt expansion, while CoreWeave would fund additional customer-specific costs. The contract is valuable because it turns an energized mining campus into a long-duration lease stream. It is also a construction obligation: the landlord must deliver powered core-and-shell infrastructure, and the tenant’s economics depend on the broader AI cloud chain remaining solvent and expanding.
IREN’s disclosures make the capital intensity clearer. The Microsoft agreement is approximately $9.7 billion over its term, but IREN expects roughly $5.8 billion of GPU-related capital expenditure associated with the contract. The company’s 750-megawatt Childress campus gives it a large physical base, yet the contract requires four deployment tranches and a shift from mining hardware toward liquid-cooled AI infrastructure. A headline contract value is therefore best read as the top line of a project-finance model, not as immediate free cash flow.
Applied Digital offers a similar conversion from mining-era site ownership to AI leasing. Its filings describe 400 megawatts under contract with CoreWeave at Polaris Forge 1, approximately $11 billion of prospective lease revenue and a roughly 15-year base term, with three additional five-year options. A second campus adds another 200 megawatts and approximately $5 billion of prospective lease revenue with a US investment-grade hyperscaler. The scale is substantial, but so is the duration. The longer the lease, the more the project depends on future GPU economics, tenant credit and the cost of capital used to build the facility.
This is why the label “power landlord” both captures the opportunity and hides the risk. Traditional landlords generally own an asset that can serve multiple tenants. These companies are often building highly specialized facilities around a small number of customers and particular cooling, networking and power specifications. Their bargaining position improves when power is scarce, but concentration can make a single delayed deployment or renegotiated contract disproportionately important.
The industry’s own filings describe secured power as a central enabler of AI infrastructure. That is the opportunity. It does not eliminate the question of who bears the cost of turning that asset into dependable AI capacity.
Why the Bitcoin-Mining Roots Matter in an AI Selloff
The market’s current discomfort is not simply that AI stocks are volatile. It is that former miners are now correlated with two cycles instead of one. Bitcoin mining exposed them to the coin price, network difficulty and electricity costs. AI infrastructure adds long-duration growth expectations, customer concentration and a capital-spending cycle that can reverse before the power asset itself becomes obsolete.
Mining’s flexibility was once a competitive advantage. Machines can be switched off when power prices rise, and miners can relocate or delay hardware purchases. AI workloads are less forgiving. Training and inference clusters require high availability, complex networking and advanced cooling. A site that can interrupt mining load is not automatically ready for a hyperscaler. The conversion process creates a mismatch between flexible supply and inflexible demand.
That mismatch changes the transmission mechanism through which an AI rout affects miners. The direct effect is a lower valuation for companies associated with AI. The cross-asset effect is higher sensitivity to interest rates and credit spreads because the value of a 15-year or 25-year lease depends on discount rates. The cross-industry effect is pressure on GPU suppliers, cloud customers and power developers to prove that demand will justify the infrastructure already being financed. The third-order expectation gap is that investors may stop valuing every megawatt of secured power as if it will become high-margin AI revenue on schedule.
TeraWulf’s Abernathy joint venture shows the reward and the exposure. The company disclosed a 168-megawatt critical IT-load project under a 25-year lease structure, approximately $9.5 billion of contracted revenue to the joint venture and a 51% TeraWulf stake. A credit enhancement backed by Google supports $1.3 billion of Fluidstack’s long-term lease obligations for project debt. TeraWulf’s investor materials describe approximately $260 million of average annual net operating income and an estimated margin near 70% for the campus.
Those terms are attractive because they resemble infrastructure finance rather than spot crypto revenue. They are also long-dated enough to create duration risk. A 25-year commitment can stabilize cash flows if the customer remains healthy and the facility performs. It can magnify the equity reaction if investors begin to question AI utilization, refinancing or construction schedules. The market does not need to conclude that AI demand is disappearing; it only needs to demand a higher return for waiting.
The pivot is therefore partly cyclical and partly structural. The cyclical component is the current AI valuation and financing wave. It can mean-revert if customer demand, GPU utilization and credit conditions stabilize. The structural component is the scarcity of grid-connected, developable power in major data-center markets. That scarcity will not disappear because AI-linked equities sell off. What can disappear is the premium assigned to a site before its conversion is complete.
The distinction is essential. If the problem were only a cyclical decline in Bitcoin, the AI pivot would be a clean hedge. If it were only an AI boom, the mining background would be a source of cheap power. In reality, the companies are moving from one volatile compute market into another market with more demanding technical and financial requirements.
The Strongest Case Against the Pivot—and the Signal That Would Break It
The strongest counter-thesis is that investors are confusing temporary equity-market weakness with a durable infrastructure downturn. Hyperscalers have signed long-term capacity agreements because power procurement takes years, transmission upgrades are slow and AI workloads continue to require more compute. Under this view, former miners are not merely chasing a fashionable narrative; they own scarce, energized campuses that can reach customers faster than greenfield competitors. The rout could therefore be a valuation reset that leaves the physical bottleneck intact.
The contracts support part of that argument. Core Scientific’s approximately 590 megawatts of contracted HPC load, IREN’s 200-megawatt Microsoft deployment, Applied Digital’s 600 megawatts leased across two campuses and TeraWulf’s 168-megawatt Abernathy project represent signed commercial relationships rather than generic aspirations. Google’s credit enhancement for the TeraWulf venture and Microsoft’s prepayment structure with IREN also show that customers and financial counterparties are willing to support development when the site and tenant fit their needs.
But the counter-thesis does not answer the execution problem. Secured power is scarce; completed, revenue-producing AI capacity is scarcer. IREN’s own filings say that it began decommissioning Bitcoin-mining hardware and reallocating capacity toward AI Cloud, with the transition targeted to be substantially complete by Dec. 31, 2026. That is a transformation still in progress. Applied Digital’s lease revenue is prospective over long terms. Core Scientific’s contracts involve customer-specific build-out. TeraWulf’s Abernathy delivery target is in the second half of 2026. The equity market is being asked to fund the bridge between contract signing and operating cash flow.
The falsifying signal for the cautious view is concrete: if the named projects reach their contracted delivery milestones through 2027, customers begin paying at the disclosed schedules, and utilization remains high enough to support the reported margins, then the argument that the pivot is mainly a financing narrative will be wrong. Conversely, repeated delays of more than two quarters across two or more major projects, combined with material contract renegotiations or a sustained decline in customer prepayments, would show that power ownership is not translating into dependable AI infrastructure economics.
That test is more useful than the daily share-price reaction. Stock prices can recover before projects operate, or fall while the underlying leases remain intact. The harder evidence is whether facilities are energized, accepted by customers and producing cash without repeated equity issuance or expensive refinancing.
Three Horizons for the New Infrastructure Trade
In the short term, sentiment and liquidity dominate. AI-linked miners can trade like high-beta technology stocks even when their power assets do not change in value. A broad retreat from long-duration growth can compress multiples, raise financing costs and make equity-funded construction more punitive. Bitcoin adds another source of correlation: if crypto and AI risk appetite weaken together, the companies lose the diversification benefit that the pivot was supposed to provide.
In the medium term, fundamentals separate the operators. The decisive variables are delivery timing, tenant prepayments, construction cost per megawatt, GPU availability, power utilization and the share of revenue that is contracted rather than exposed to spot mining economics. IREN’s approximately $5.8 billion GPU capital requirement, TeraWulf’s disclosed $8 million-to-$10 million project cost per critical megawatt and Core Scientific’s $104 million funding obligation for a 70-megawatt expansion show why capital discipline matters as much as contract size.
In the long term, the structural question is whether power becomes the durable bottleneck in AI. If grid-interconnection queues, transformer shortages and local opposition continue to delay greenfield capacity, existing mining sites can retain a time-to-power premium. Their advantage would be most durable where operators control land, transmission rights, cooling systems and expansion options, not just an electricity contract. If utilities and hyperscalers solve those constraints faster than expected, the premium will migrate away from legacy miners toward purpose-built data-center developers and power owners.
The base case is a selective repricing rather than a collapse of the thesis. Contracts with strong counterparties continue, but the market values operating milestones more than announced megawatts. The upside case arrives if AI demand remains firm, customer prepayments reduce funding pressure and former miners deliver facilities ahead of competing greenfield projects. The downside case is a capex digestion cycle: customers slow deployments, financing costs rise and the companies must carry partially converted sites while Bitcoin economics remain weak. Its trigger would be a combination of delayed delivery, lower customer commitments and a rise in project-level borrowing costs.
The beneficiaries of the structural trend are not necessarily every former miner. They are the operators that can convert scarce power into compliant, revenue-producing capacity while keeping leverage and tenant concentration within tolerable limits. The exposed companies are those whose valuation depends on the next contract announcement, whose projects require repeated equity financing or whose power cannot be upgraded to AI-grade reliability.
The market is now asking a better question than whether Bitcoin miners can become AI companies. It is asking whether their inherited power assets can produce durable cash flow after the AI premium is removed. That is a slower question, but it is the one that will decide whether the pivot is infrastructure or merely another cycle in disguise.
Data cutoff: Nov. 12, 2026, UTC. Company contract values and capacities are prospective or disclosed project figures and should not be read as realized revenue or completed operating capacity.
The Bitcoin-mining past is not a footnote to the AI pivot; it is the stress test. Power creates the option, but only delivered compute turns that option into infrastructure.
Explore more exclusive insights at nextfin.ai.
