Bitcoin miners are repurposing grid-connected sites for AI data centers, but access to power does not guarantee a share of AI revenue. Conversions require substantial investment, and customer funding for construction may be recovered through rent credits while operators remain responsible for financing, timelines and delivery. Contracts between Core Scientific and CoreWeave show why headline contract values can differ significantly from the cash operators actually receive.
Mining machines themselves have little use in training AI models. The value of a mining site lies mainly in its land, substations and grid connections—assets that may help data center developers avoid long waits for power. But they generate AI revenue only once facilities are upgraded, usable electricity is available and customers are in place.
SWI expands into AI infrastructure, but power figures need context
SWI Group gained control of bitcoin miner Genesis Digital Assets this year and announced an €800 million capital increase on October 12 to expand its AI infrastructure business. Two weeks later, it launched SWI Compute, disclosing an approximately $1 billion order for Nvidia equipment and its first customers. Some contracts were still being finalized at the time.
SWI previously said Genesis had more than 1.3 gigawatts of energized and approved capacity across 15 U.S. facilities. Its September half-year results cited approximately 1.2 gigawatts of secured grid access. The company has not explained the difference site by site, or distinguished capacity currently available from capacity requiring further work.
These power figures are not interchangeable. Electricity already delivered to a site, future capacity that has been approved and projects still waiting in the interconnection queue are at different stages of readiness for AI computing. In Texas, regulators approved ERCOT’s Batch Zero review process in June, while the state government ordered a review of about 474 gigawatts of large-load interconnection requests. Assessments of grid capacity, transmission upgrades and the allocation of associated costs remain part of bringing projects online.
SWI initially invested $500 million for an approximately 38.3% stake in Genesis and a 77.2% interest in the value of specified preferred-share liquidation preferences. A liquidation preference affects how proceeds are distributed in a sale or liquidation; it is not the same as ownership of ordinary shares. SWI has since taken control, but dividing the initial investment by 1.3 gigawatts would not show what it paid for deliverable power. Genesis also includes mining operations, equipment, land, electrical assets and future rights related to power.
Mining assets lose value as AI conversions raise costs
Bitcoin miners have long relied on specialized ASIC machines to compete for block rewards while paying ongoing electricity bills. Those bills do not fall in step with bitcoin prices, whether the market is rising or declining. Core Scientific recorded $266.5 million in impairment charges on mining equipment and infrastructure in the first quarter of 2026: $151.6 million for mining machines and $114.9 million for mining infrastructure. The company cited weaker bitcoin prices, lower mining revenue per unit of computing power and declining resale values for used equipment. Its newer AI hosting business, meanwhile, had begun generating revenue.
Mining machines depreciate with use, while grid access can be harder to replicate. Even developers with the capital to buy GPUs must verify transmission capacity, determine what upgrades are needed and establish who will pay for them. Power access is therefore a key asset in a mining-site conversion, but a single capacity figure can be misleading if availability and project timing are unclear.
Mining facilities and AI data centers also have different operating requirements. A mining site typically contains many machines that can be switched off independently. An AI data center needs cooling for high-density hardware, high-speed networking, redundant power and stricter uptime standards. Buying an existing mining site provides a starting point, not a substitute for engineering work or customer delivery.
Not all electricity drawn from the grid reaches computing equipment. At a facility with a power usage effectiveness (PUE) of 1.25, 100 megawatts of grid supply would leave about 80 megawatts for IT equipment, with roughly 20 megawatts used for cooling, power distribution and other facility operations. A 100-megawatt grid connection is therefore not equivalent to 100 megawatts of usable IT capacity.
Nvidia’s GB300 NVL72 reference design uses liquid-cooled racks, each containing 72 GPUs, with power consumption of up to 142 kilowatts per GPU. In a September mining report, CoinShares estimated that converting mining infrastructure to AI-grade facilities costs about $8 million to $15 million per megawatt, compared with roughly $700,000 to $1 million per megawatt for traditional mining infrastructure. That cost gap helps explain why a site with power access is not valued the same as a completed AI data center.
Core Scientific-CoreWeave terms show how rent credits work
Some of Core Scientific’s facilities are already operating, and its contract with AI cloud provider CoreWeave offers a view of how the revenue is structured. The company reported $136.7 million in hosting revenue for the second quarter of 2026. By mid-July, capacity billed to customers had reached 437 megawatts. Its quarterly filing classified $98.8 million as license fees and $35.1 million as electricity costs passed through to customers, with the remainder from other services.
Electricity reimbursements are recorded alongside the corresponding expenses. They increase reported revenue but do not generate a matching amount of gross profit. In other words, a substantial share of hosting revenue is a pass-through of power costs rather than profit from providing facilities.
Under terms disclosed in April, base license fees for most CoreWeave projects start at $100 per kilowatt per month and rise by 3.5% annually, with CoreWeave paying for electricity. CoreWeave also funds or reimburses certain approved construction costs. For specified owner-owned equipment, the customer can receive rent credits of up to $1.5 million per megawatt. Those credits may temporarily cut monthly rent in half until the allowance is used.
At $100 per kilowatt per month, a one-megawatt facility generates a base monthly fee of $100,000. If a 50% construction credit applies, the operator would collect $50,000 in rent that month. Using the full $1.5 million credit at $50,000 per month would take 30 months, before accounting for rent increases and other contract terms. The customer helps finance construction, then recoups part of that investment over time through reduced rent.
The arrangement can reduce the amount Core Scientific must raise upfront, but the stated rent over a 12-year contract does not mean the full amount arrives in cash during the early years. Company filings say that advance payments, construction credits and straight-line revenue recognition can all create differences between accounting revenue and cash collected.
The contracts also provide for credits of $20,000 per day for certain delivery delays, rising to $30,000 per day later. Some delays may also give CoreWeave the right to terminate an order. One land lease is contingent on the related power agreement remaining in effect. Even after millions of dollars have been spent on construction, the power agreement remains a material part of a facility’s value.
Operators retain financing and delivery obligations
The April contract disclosures also said Core Scientific must provide an uncapped completion guarantee for certain projects if designated financing sources do not provide enough available construction funds. The company recorded $954.2 million in capital expenditures in the first half of 2026, of which CoreWeave funded $180.9 million.
In May, a financing subsidiary issued $3.3 billion of secured notes with a 7.75% coupon, backed by assets and rights tied to specific data center projects. Based on the original principal, annual coupon payments amount to about $255.8 million, before principal repayments. Scheduled principal payments do not begin until rent starts and the tenant credit period ends, but interest remains due during the credit period.
In this structure, CoreWeave gets access to data center facilities without having to develop all of the real estate itself and can recover part of its construction investment through rent credits. Lenders receive interest supported by project assets and expected customer payments. Core Scientific retains ownership of the facilities and the potential for long-term revenue, but remains responsible for construction, financing and delivery. The total contract value alone therefore does not show how much cash each party ultimately receives or what financing costs the operator bears.
Investor attention to AI infrastructure financing is also growing. On October 22, short positions in five AI infrastructure companies were reported to have generated about $6 billion in unrealized gains for the month, including positions in CoreWeave and Core Scientific. On October 26, CoreWeave shares fell 4.5% before the market opened and Core Scientific fell 1.7% after Elon Musk discussed concerns about loan financing for specialist AI cloud operators. Short-term share moves do not by themselves indicate whether a company can repay its debt, but they show that markets are weighing AI contract prospects alongside project financing pressure.
The central challenge in converting bitcoin mines is not simply securing more power, but turning grid access into operational AI facilities that can be billed to customers. Construction costs, customer credits, interest payments and delay obligations all affect how revenue is divided. Until projects are delivered, contract values are not the same as operating cash flow realized.