Core Scientific is accelerating its transformation from Bitcoin miner into high-density AI infrastructure provider, with colocation now accounting for the overwhelming majority of its business. In its official third-quarter results, the company reported $215.4 million in colocation revenue during Q3, compared with just $12.3 million from self-mining and $5.2 million from hosted mining. Core Scientific had 485 MW of customer capacity billing at September 30, supporting more than $700 million in estimated average annualized colocation revenue.
The transition is advanced, but the mining exit requires a distinction. Core Scientific expects its remaining hosted mining business to end by December 31, while self-mining is being progressively wound down without a disclosed final shutdown date. The company still operated its own miners at two facilities at quarter-end, but says it is no longer making material investments to maintain or expand mining capacity.
AI Contracts Reshape Core Scientific’s Revenue Base
CoreWeave remains the foundation of the colocation business. Approximately 485 MW of the roughly 590 MW leased to CoreWeave were already billing at September 30, with the remaining capacity under construction and commissioning. Core Scientific’s investor presentation puts the CoreWeave contracts at more than $10 billion of potential revenue, with roughly $850 million in estimated average annualized GAAP revenue.
AMD adds a second major commercial relationship. Core Scientific signed 15-year agreements covering approximately 530 MW across five sites, including 377 MW leased directly to AMD and 152 MW to a neocloud counterparty supported through agreements with AMD. Reuters reported that capacity begins coming online in 2027 and could eventually expand to 2.5 GW. The initial 530 MW carries more than $14 billion of potential base contracted revenue, while the additional capacity remains an expansion opportunity rather than committed deployment.
The structure places Core Scientific within a broader migration of mining infrastructure toward AI. Hyperscale Data has already shut down Bitcoin mining at its Michigan site to convert it into an AI data center, while JPMorgan has highlighted the accelerating HPC transition among listed miners such as Cipher and CleanSpark. Power access, land and existing data-center infrastructure are increasingly being monetized through long-duration compute contracts instead of Bitcoin block rewards.
Mining Economics Continue to Deteriorate
The financial results show why that reallocation is occurring. Self-mining revenue fell 79% year over year to $12.3 million in Q3, while its direct cost of revenue reached $19.8 million, producing a segment-level gross loss. Core Scientific attributed the decline to 66% lower Bitcoin production and a 38% lower average Bitcoin price compared with Q3 2025. Mining had fallen to 12% of nine-month revenue by September, down from 78% during the comparable 2025 period.
The company has already incurred substantial costs while leaving that business. It recognized a $266.5 million impairment on mining-related property and equipment earlier in 2026 and another $41.9 million loss after terminating its equipment agreement with Block and Proto. Those charges reflect assets and commitments becoming less useful as Core Scientific redirects capital toward high-density colocation.
Financing has changed alongside the operating model. Core Scientific’s earlier $1 billion Morgan Stanley term loan was fully repaid and terminated in May, using proceeds from a $3.3 billion offering of 7.75% senior secured notes due 2031. It subsequently added $600 million of committed revolving and letter-of-credit facilities in August. The AI pivot therefore replaces Bitcoin-price and mining-difficulty exposure with substantial construction, financing and long-duration contract execution risk.
That capital intensity has become increasingly important as higher borrowing costs pressure infrastructure companies. Higher Treasury yields can affect crypto companies more directly than Bitcoin itself because data-center conversions require debt, construction spending and returns on invested capital. Core Scientific spent $985.1 million on capital expenditures in Q3 alone. Contracted AI demand improves revenue visibility, but realizing that value still depends on delivering megawatts on schedule and financing the buildout efficiently.
Customer concentration is another emerging risk. Core Scientific disclosed that one colocation customer generated 93% of total Q3 revenue, although the AMD and neocloud deployments should broaden the customer base as they begin contributing revenue. The business has already shifted economically away from Bitcoin mining, but its next test is whether the expanding AI portfolio can diversify revenue while maintaining the economics implied by more than $24 billion of contracted capacity.
