Bernstein analysts have identified Bitcoin miners as emerging infrastructure suppliers for artificial intelligence operations. The research finds miners control 27 gigawatts of planned power capacity and have locked in $90 billion in AI-related deals. This positioning gives mining operations significant leverage as electricity becomes the primary bottleneck for data center expansion.

The shift reflects a fundamental market reality. AI model training and inference demand massive computational power. Miners already possess the engineering expertise, power procurement relationships, and operational infrastructure built for hash rate optimization. Those same assets translate directly to serving GPU clusters and inference servers.

Miners control stranded power sources that traditional data center operators struggle to access. Remote hydroelectric facilities, flare gas operations, and renewable energy sites often lack connection to conventional grid infrastructure. Mining operations have pioneered direct-connect arrangements with these power sources. AI companies increasingly purchase electricity through similar channels.

The $90 billion in AI deals signals real capital flows. Companies like Core Scientific, Marathon Digital, and Hut 8 have announced partnerships with AI workload providers. Some miners now allocate spare GPU capacity to inference tasks. Others sell power directly to AI infrastructure firms at premium rates. These revenue streams diversify income beyond Bitcoin block rewards.

Electricity constraints reshape AI expansion timelines. Hyperscalers compete intensely for available megawatts. Mining operations converted this competition into a new business model. By aggregating power supply options, miners function as critical intermediaries between renewable energy sources and AI compute providers.

This transition doesn't eliminate mining fundamentals. Bitcoin networks still require computational security. Miners still chase profitability through hash rate optimization. The change reflects revenue diversification under cost pressure from hardware competition and network difficulty.

The strategic advantage holds constraints. AI workload patterns differ from mining. Inference tasks require sustained power rather than flexible load shifting. Long-term power purchase agreements for AI differ structurally from mining arrangements