MARA Sees Bitcoin Mining Assets as a Fast Track to AI Data Center Growth

Key Points

• Bitcoin miners are leveraging existing power, land and infrastructure to expand into AI data centers, colocation and cloud services. MARA views these assets as a fast track for hyperscalers and AI companies seeking additional computing capacity.
• AI data-center development requires far greater capital than Bitcoin mining, with direct-liquid-cooled construction costs rising to approximately $10 million–$10.5 million per megawatt. Labor, transformers, switchgear, financing and regulatory approvals remain major constraints.
• Companies are pursuing large AI infrastructure deals, but executives warned that speculative power requests could exceed realistic demand. MARA is working with Starwood to reduce construction risk and expects to sign two leases by year-end while evaluating both large campuses and smaller inference-focused sites.

Executives from Bitcoin mining and digital infrastructure companies said access to energized power, land and data-center development capabilities is creating an opportunity to serve growing artificial intelligence infrastructure demand, while also presenting significant construction, financing and regulatory challenges.

Speaking at an H.C. Wainwright panel moderated by Mike Colonnese, the companies described how Bitcoin miners are seeking to repurpose or complement mining operations with AI high-performance computing, colocation and cloud-service deployments. Colonnese said the companies represented on the panel have line of sight to more than 14 gigawatts of power capacity.

Bitcoin Mining Assets as an AI Infrastructure Advantage

Fred Thiel, chairman and CEO of MARA NASDAQ: MARA, said Bitcoin miners historically focused on acquiring low-cost power and deploying computing equipment as efficiently as possible. In contrast, AI data centers require substantially more capital per megawatt but can generate higher revenue and asset values per megawatt.

“Bitcoin miners have a lot of land and power that is turned on,” Thiel said, calling those assets a potentially fast route for hyperscalers, frontier-model providers and neocloud companies seeking to bring AI computing capacity online.

Russell Cann, co-founder and chief development officer of Core Scientific, said the company’s early AI deployments demonstrated the changing technical and cost requirements of the market. He said an air-cooled site was converted in roughly 45 days to support higher-density racks running H100 and H200 graphics processing units. However, direct-liquid-cooled facilities generally required new construction rather than conversion of legacy air-cooled data centers.

Cann said estimated direct-liquid-cooled construction costs increased from about $4.5 million per megawatt in late 2023 to approximately $10 million to $10.5 million per megawatt for sites turned on during the current year. He cited labor, transformers and switchgear as major constraints.

Colocation and Cloud Strategies

Panelists described different approaches to monetizing AI infrastructure. Sam Tabar, CEO of WhiteFiber, said the company operates both cloud and colocation businesses, which require separate technical capabilities. WhiteFiber was spun out from Bit Digital as a pure-play AI infrastructure company, he said, in part to broaden financing opportunities and attract institutional shareholders that may not have wanted crypto exposure.

Tabar said WhiteFiber acquired Enovum to add an experienced data-center development team. He cited projects including the conversion of a mattress factory into an AI campus for Cerebras and a North Carolina facility developed for Nscale under an $865 million contract.

Michael Potter, CFO of Bitdeer, said the company is pursuing both colocation and GPU-as-a-service. Bitdeer signed a colocation agreement in Tydal, Norway, for about 121 critical IT megawatts, with tenant Volta providing GPUs for its end customer. Bitdeer is also developing a neocloud business in Malaysia, beginning with a 10-megawatt site that Potter said has two customers.

• CleanSpark said it signed a 20-year, triple-net lease valued at $6.6 billion for a 250-megawatt powered site in Sandersville, Georgia.
• MARA said it expects to sign two leases by year-end and is evaluating both large campuses and smaller inference-focused sites.
• Core Scientific said it recently announced a roughly 500-megawatt agreement with AMD and granted AMD rights to the next 2,000 megawatts at sites in Texas and Oklahoma.
• Soluna said it is developing AI campuses alongside wind and solar power facilities, including the 350-megawatt Cotty 2 project and the approximately 300-megawatt Dorothy III project in Texas.

Financing, Execution and Power Constraints

John Belizaire, CEO of Soluna Holdings, estimated that a theoretical 100-megawatt project could cost $1.2 billion to $1.4 billion, or roughly $10 million to $12 million per megawatt. He said Soluna expects project financing to be principally debt-funded, using an approximately 80% debt and 20% equity structure.

Matt Schultz, chairman and CEO of CleanSpark, said delivery certainty has become a central issue for lenders and tenants. CleanSpark selected an engineering, procurement and construction management provider that manufactures many data-center components in a factory, he said, and entered into a fixed-price, lump-sum turnkey contract. CleanSpark expects its first Georgia data hall to be delivered in December 2027.

Thiel said MARA partnered with Starwood because the company wanted to reduce construction risk and leverage expertise in hyperscale data-center development. He added that contributing sites into project joint ventures could improve cash-on-cash returns and allow multiple projects to advance simultaneously.

Executives Debate AI Infrastructure “Bubble” Concerns

Most panelists rejected the idea that demand for AI infrastructure is a bubble, citing demand for both training and inference computing. However, Core Scientific’s Cann distinguished between real demand and speculative power requests.

Cann said roughly 5 gigawatts of rack capacity is being delivered annually against about 15 gigawatts of annual chip demand, but he questioned the validity of all proposed data-center projects. He said Texas had received requests totaling 474 gigawatts, a figure he described as far beyond realistic demand or available supply.

“The issue is not the guys delivering rack space or the guys delivering the chips,” Cann said. “It is every farmer and rancher” near gas or transmission infrastructure claiming potential data-center capacity.

Schultz said political and community considerations are increasingly important, while Tabar said retrofitting existing industrial facilities can reduce opposition associated with greenfield construction. Panelists also said Bitcoin mining remains strategically useful because miners can rapidly deploy load and monetize power before AI data centers are completed.

About MARA (NASDAQ:MARA)

MARA Holdings, Inc, formerly known as Marathon Digital Holdings, Inc, is a digital asset technology company focused primarily on mining bitcoin. The company operates and supports computing infrastructure designed to secure the Bitcoin network by processing transactions and earning bitcoin rewards.

MARA develops and deploys large-scale data center and energy infrastructure for digital asset mining. Its activities include operating mining facilities, managing specialized application-specific integrated circuit (ASIC) mining equipment, and pursuing energy and infrastructure solutions intended to support more efficient and scalable computing operations.

This instant news alert was generated by narrative science technology and financial data from MarketBeat in order to provide readers with the fastest reporting and unbiased coverage. Please send any questions or comments about this story to contact@marketbeat.com.