Bitcoin miners chase AI deals, but converted data centers are costly and slow to earn
Bitcoin mining companies have announced more than $100 billion in AI hosting contracts, but very little of that capacity was actually earning revenue in September. A review of the deals found only about 550 megawatts was billing customers out of more than 4 gigawatts under contract.
The reason is simple: a Bitcoin mine is not an AI data center. The buildings, power connections, and land are valuable, but converting them takes years of construction and large amounts of borrowed money before any promised revenue arrives.
Key points at a glance
- More than $100 billion in AI contracts have been announced by Bitcoin miners, but only around 550 MW was billing out of more than 4 GW under contract as of September.
- CoinShares estimated converting mining infrastructure to AI-grade facilities costs roughly $8 million to $15 million per MW, versus $700,000 to $1 million per MW for mining builds.
- Core Scientific reported $136.7 million in AI colocation revenue for the quarter ended June 30, 2026, with 437 MW billing by mid-July.
- SWI Group took control of miner Genesis Digital Assets and launched SWI Compute, disclosing roughly $1 billion in NVIDIA equipment orders on Oct. 26.
- Core Scientific issued $3.3 billion of secured notes at 7.75% in May, adding about $255.8 million in annual interest costs.
Why Bitcoin mines attract AI companies
Mining companies spent years securing cheap electricity, building substations, and negotiating grid connections. AI companies want those power connections badly, because new ones are hard to get. Utilities must assess transmission capacity, identify upgrades, and decide who pays for them.
The mining computers themselves, known as ASICs, are nearly useless for running AI models. What matters is the electricity access around them.
However, Bitcoin mining hardware loses value as it ages, while a working grid connection does not. Core Scientific's second-quarter filing showed the strain on the old business: the company wrote down $266.5 million of mining equipment and infrastructure during the first quarter, including $151.6 million for mining machines and $114.9 million for infrastructure. Its mining operation lost money at the gross-profit level, while its newer AI hosting business earned a positive gross margin.
Converting a mine into an AI data center is expensive
Bitcoin mines are built to run thousands of independent machines that can be switched off. AI data centers need much more: dense liquid-cooled GPU racks, advanced cooling, high-speed networking, redundant power, and strict uptime guarantees.
There is also a difference between a site's total grid connection and the power actually available for computing. At a power usage effectiveness ratio of 1.25, a site drawing 100 MW from the grid would deliver only about 80 MW to computing equipment, using the rest for cooling and power distribution.
CoinShares' September mining report estimated the conversion cost at roughly $8 million to $15 million per MW, compared with $700,000 to $1 million per MW for conventional mining infrastructure. That gap explains why owning a powered site and owning a finished AI facility command very different values.
How Core Scientific's contracts divide the money
Core Scientific's agreements with AI cloud provider CoreWeave show where the cash goes. Under terms disclosed in April, most sites start at a base license fee of $100 per kilowatt per month, rising 3.5% each year, with CoreWeave responsible for utility costs.
CoreWeave also funds certain approved construction expenses and receives rent credits worth up to $1.5 million per MW for specified landlord-owned equipment. Those credits can cut the monthly rent payment in half until the allowance is used up. In effect, the customer helps pay for the data center, then recovers that money through lower rent over time. One megawatt at the initial rate carries a base fee of $100,000 a month, but a 50% credit reduces the payment to $50,000.
Core Scientific reported $136.7 million of colocation revenue in the quarter ended June 30, 2026. Of that, $98.8 million came from license fees and $35.1 million from power charges passed through to customers. The electricity reimbursement also appears as an equal expense, so it adds to revenue without creating gross profit. More than a quarter of the colocation revenue was therefore a utility bill passed along rather than earned income.
The company also carries the risk. The agreements provide for daily late-delivery credits of $20,000, rising to $30,000 in specified circumstances, and some delays can give CoreWeave the right to terminate an order. Core Scientific provided an uncapped completion guarantee for specified financed projects if construction funds prove insufficient.
The financing bill keeps growing
Core Scientific recorded $954.2 million in capital expenditure during the first half of 2026, including $180.9 million funded by CoreWeave. In May, a financing subsidiary issued $3.3 billion of secured notes at 7.75%, which works out to roughly $255.8 million in annual interest on the full original principal.
Principal repayments begin only after rent starts and tenant credits run their course, but interest payments continue in the meantime.
Markets have started pricing in that risk. On Oct. 22, Reuters reported that short sellers held about $6 billion in unrealized gains across five AI infrastructure companies in October, including CoreWeave and Core Scientific. A further Reuters report on Oct. 26 described premarket declines of 4.5% for CoreWeave and 1.7% for Core Scientific after Elon Musk warned of lenders' reluctance to fund specialized AI cloud operators. Two days of price movement do not determine whether a business can repay its debt, but they show how quickly enthusiasm for AI contracts can meet anxiety about the financing behind them.
SWI's faster, earlier-stage bet
SWI Group took control of Bitcoin miner Genesis Digital Assets earlier this year and announced an €800 million capital increase on Oct. 12 to expand its AI infrastructure business. Two weeks later, it launched SWI Compute, disclosing roughly $1 billion in NVIDIA equipment orders and its first customers, with some contracts still being finalized. SWI also announced a 13 MW infrastructure partnership in Spain on Oct. 26.
SWI reported €631.6 million in first-half profit, largely reflecting the value recognized on its Genesis investment.
The capacity figures are harder to read. SWI's June announcement described Genesis land with more than 1.3 GW of energized and approved connections across 15 facilities. By September, its half-year results described approximately 1.2 GW of secured grid connections. The disclosures do not reconcile the two figures or break down which capacity is usable now versus what needs more work.
What is confirmed and what is not
Confirmed figures include Core Scientific's $136.7 million colocation revenue, its $266.5 million write-down, the $3.3 billion note issuance at 7.75%, and SWI's disclosed orders and capital raise. Reuters' reporting on short sellers and the October premarket declines is reported information.
What remains unclear is how much of the headline power capacity at companies like SWI is deliverable today, and when the rest of the industry's contracted megawatts will start billing. Headline contract values do not show how the revenue is divided among builders, lenders, and customers.
Why this matters
Former mining sites are becoming some of the most sought-after real estate in the AI buildout, because electricity access cannot simply be bought. But the contracts determine who actually gets paid. CoreWeave gets facilities without developing all the real estate itself. Lenders collect interest backed by project assets. Core Scientific keeps ownership and a potential long-term income stream, along with the responsibility and debt for building the facilities. That division is not obvious from the advertised contract value.
Sources
Newisty Editorial Team covers technology, cryptocurrency, digital products, online platforms, developer tools and the wider digital economy. Our content is researched from official sources, company announcements, public documentation, market data and other primary or reputable sources. Articles are reviewed and edited before publication for clarity, accuracy and useful context.