Riot repaid its $200 million Coinbase facility and released its collateral. IREN’s quarterly AI revenue has already overtaken Bitcoin mining. Long-term AI contracts are supporting billions in
Riot Platforms has eliminated a $200 million credit facility backed by financial assets including Bitcoin while committing capital to a business with a very different financing model: large-scale data centers.
On September 21, Riot voluntarily repaid the full principal and accrued interest on its Coinbase Credit facility, terminated the agreement and secured the release of the related security interests. There was no early termination fee.
The repayment arrives as a wider transformation among Bitcoin miners becomes measurable. IREN, TeraWulf, Hut 8, Cipher Mining and Core Scientific have turned power portfolios originally assembled for mining into AI and high-performance computing projects backed by multiyear customer contracts.
Some are already generating meaningful AI revenue. Others have billions of dollars of contracted capacity that still needs to be built.
Riot’s Coinbase facility began at $100 million in April 2025 and was expanded to $200 million the following month. The company ultimately drew the entire facility.
At June 30, Riot reported 11,380 BTC, including 5,821 BTC held as collateral. The September filing does not disclose the amount of Bitcoin pledged immediately before repayment, so 5,821 BTC should not be interpreted as the exact amount released this month.
The financing Riot is using for its data center expansion looks different.
In January, the company purchased approximately 200 acres at its Rockdale site for $96 million, financing the acquisition through the sale of roughly 1,080 BTC. The transaction accompanied Riot’s first major data center lease with AMD.
Riot has since arranged a $573 million senior secured delayed-draw facility to develop 191 MW of critical IT capacity at Rockdale. The associated 20-year lease is expected by Riot to generate approximately $9.1 billion in base rent.
Bitcoin has therefore played several roles in Riot’s expansion. It backed corporate borrowing and was later sold to acquire strategic land. The new data center business is increasingly capable of attracting financing against individual projects and their expected contractual cash flows.
Mining, however, remains Riot’s larger operating business. Second-quarter Bitcoin mining revenue reached $113.7 million, compared with $23.2 million from Data Centers.
Riot is building a second business rather than simply replacing the first.
IREN provides a clearer view of what the financial statements can look like further into the transition.
Its fiscal fourth-quarter AI Cloud Services revenue reached $70.5 million, exceeding the $66.7 million generated from Bitcoin mining. Full-year AI Cloud revenue climbed to $128.8 million from $16.4 million a year earlier.
The conversion was not cheap.
IREN recorded $638.8 million in non-cash impairments in FY2026, primarily associated with Bitcoin mining hardware being decommissioned as facilities were converted for AI workloads. It reported a $702.6 million net loss for the year.
That distinction is central to the sector’s transition. Existing access to land and electricity can transfer well from Bitcoin mining to AI. ASIC miners generally cannot.
IREN is replacing part of that hardware base with infrastructure designed around GPUs, higher-density compute and different cooling and networking requirements.
The company now has approximately $9.7 billion contracted with Microsoft over five years across four 50 MW deployments at Childress, alongside a separate five-year NVIDIA agreement valued at approximately $3.4 billion.
Its financing structure shows how those contracts can change the balance sheet.
IREN secured $3.65 billion of investment-grade GPU financing for the Microsoft deployments. Together with customer prepayments, the company says roughly 96% of the associated $5.81 billion GPU capital expenditure is financed.
The GPUs and contracted AI economics can now support financing that previously would have depended much more heavily on corporate equity, Bitcoin holdings or mining cash flow.
Not every former miner wants to operate the GPUs.
TeraWulf is developing infrastructure that customers can occupy under long-term agreements. It reported 81 MW of revenue-generating critical IT capacity at Lake Mariner at the end of June, increasing to 102 MW following another building delivery in July. A further 336 MW was under construction.
Its relationship with Fluidstack demonstrates another feature of AI infrastructure financing: the credit quality behind the tenant can matter almost as much as the tenant itself.
Completion of CB-3 activated $600 million of Google financial support for Fluidstack’s lease obligations.
Hut 8 is pursuing the model at considerably greater scale.
Its River Bend agreement with Fluidstack covers 245 MW for 15 years and carries approximately $7 billion of base-term contract value, with payments financially backstopped by Google. The infrastructure is intended to support Anthropic workloads.
By the second quarter, Hut 8 reported 949 MW of contracted IT capacity, approximately $26.6 billion of expected aggregate base-term contract value and $7.5 billion of investment-grade construction financing.
The economics increasingly resemble infrastructure development: secure power, build the facility, sign a long-duration customer and finance construction against the expected contractual economics.
Cipher Mining has pushed contract duration even further.
Its Barber Lake development originally carried a 10-year Fluidstack lease. A September amendment added another 10-year commitment from a major AI company, taking expected contracted revenue from the facility above $9 billion over 20 years, according to Cipher.
That headline number comes with substantial near-term obligations.
Following design changes, Cipher agreed to fund the first $359.3 million of costs above the project’s original budget. Costs beyond that threshold are subject to a reimbursement arrangement with the tenant.
This is where comparisons based only on contract value become misleading.
Twenty years of expected payments can support financing and improve revenue visibility, but the data center must first be completed. Construction spending arrives considerably earlier than much of the revenue.
Core Scientific provides one of the more mature examples of the colocation strategy.
The company has approximately 590 MW contracted with CoreWeave across five sites, representing more than $10 billion of potential contract revenue and around $850 million of estimated average annual colocation GAAP revenue.
Its CoreWeave arrangements largely use take-or-pay structures and include annual escalators. Contracted revenue also feeds into a financing structure supporting $3.3 billion of senior secured notes due in 2031.
That is an important change from conventional mining economics.
Bitcoin mining revenue varies with BTC prices, network difficulty, fees, machine efficiency and electricity costs. A long-term data center lease creates a contractual revenue stream that can potentially be modeled years ahead.
For lenders, the same power infrastructure can consequently support a different amount and type of debt once a credible AI customer is attached to it.
Putting the companies side by side makes the divergence clearer.
How Bitcoin Miners Are Building Their AI Businesses The same power advantage is producing very different business models. Riot PlatformsDATA CENTERS 241 MW contracted critical IT capacity AMD + another AI tenant ~$9.1B expected base rent on 191 MW lease IRENAI CLOUD AI Cloud + infrastructure model Microsoft + NVIDIA ~$13.1B across two major agreements TeraWulfCOLOCATION 102 MW revenue-generating by July Fluidstack + Google credit support 336 MW additional capacity under construction Hut 8POWER + AI 949 MW contracted IT capacity Fluidstack + Google-backed payments ~$26.6B expected portfolio base-term value Cipher MiningLONG-TERM LEASE Barber Lake AI development Fluidstack + major AI company >$9B expected revenue over 20 years Core ScientificAI / HPC ~590 MW contracted across five sites CoreWeave >$10B potential contract revenue The common asset: power. The difference is how each former miner is turning that power into AI revenue, from cloud services and colocation to decades-long infrastructure leases. Contract figures use different definitions and durations and should not be compared as equivalent revenue metrics.The figures are not directly comparable. Some represent leases, some service agreements, some portfolio-level expected values, and the companies use different definitions and contract periods. They are better viewed as indicators of the scale and structure of each transition than as a ranking.
Long-term contracts address one weakness of Bitcoin mining: extreme revenue volatility. They introduce others.
AI data centers require sophisticated cooling, networking, redundancy and high-density electrical infrastructure. GPU deployments can require billions of dollars before revenue begins. Construction delays can push back rent commencement, while dependence on a small number of hyperscale customers concentrates counterparty exposure.
The assets are different too.
IREN’s $638.8 million impairment illustrates what happens when mining hardware no longer fits the company’s intended use of a site. Cipher’s $359.3 million initial exposure to additional Barber Lake costs demonstrates the construction risk sitting underneath enormous long-term contract values.
A miner therefore does not become an AI company simply because it owns cheap electricity.
Power gives it an unusually valuable starting position. Converting that advantage into a functioning AI data center requires another round of capital, equipment and execution.
Riot’s $200 million repayment captures one part of that transition without proving its outcome.
The more consequential numbers will appear as its contracted megawatts are energized and begin contributing revenue.
IREN has already crossed an important threshold, with AI Cloud Services surpassing Bitcoin mining revenue for a quarter. TeraWulf has more than 100 MW producing revenue. Core Scientific has moved hundreds of megawatts into long-term CoreWeave agreements. Hut 8 and Cipher are building against contracts stretching well into the next decade.
Their approaches differ, but the underlying asset that attracted AI customers is largely the same one that originally attracted Bitcoin miners: power that was already secured at scale.
The next phase will show whether these companies can turn that inherited advantage into durable cash flow before the cost of building for AI catches up with the size of the contracts they have signed.
The post Bitcoin Miners Found AI. Now Comes the Expensive Part appeared first on ETHNews.
