Bitcoin miners’ move into AI infrastructure creates two separate questions: can a company earn an attractive return from its power assets, and what happens to the computing power securing Bitcoin? The answers can move in different directions.
The clearest way to understand the transition is to follow a site from electricity approval through construction to customer payments, then separately follow its mining machines. A large AI contract can improve a company’s prospects while revealing little about how much Bitcoin hashpower will disappear. Recent disclosures show why those distinctions matter.
A power announcement has several stages
Power access gives an operator a starting point. Turning it into dependable AI capacity requires a suitable facility, financing, construction and customer acceptance. Each stage has its own risks and timetable.
Core Scientific’s September 10 power update illustrates the first distinction. The company reported conditional approval for an additional 300 megawatts of electrical power at Pecos, Texas, but said the final allocation and load ramp would follow completion of ERCOT’s Batch Zero process. That was a dated statement about a grid-approval milestone, rather than confirmation that those additional megawatts were serving customers. Core Scientific’s September 10 report.
The units also matter. Electrical capacity at a site includes power supporting the facility; critical IT capacity describes power available for computing equipment. Readers need a company’s reconciliation before treating the figures as interchangeable. Core Scientific’s capacity definitions.
Riot’s August 10 results separated commitments from delivery. It reported 241 MW of contracted critical IT capacity across two customers and completion of AMD’s initial 25 MW deployment. Its newly announced 191 MW lease had phased delivery expected from December 2027 through June 2028. Those future dates are central to understanding when the contract could become an operating business. Riot’s second-quarter results.
Revenue quality matters as much as capacity
A contract headline compresses years of expected activity into one number. Evaluating the economics requires separating recurring service income, construction-related revenue, capital spending and financing costs.
For the quarter ended June 30, Riot reported $23.2 million of data-center revenue: $4.9 million from operating leases and $18.3 million from tenant fit-out services. Applying a recurring-rent interpretation to the full amount would misread the business mix. The distinction becomes especially important when comparing companies at different stages of construction. Riot’s revenue breakdown.
Core Scientific reported $136.7 million of colocation revenue and $797.5 million of capital expenditures for the same quarter. Its reported billing capacity rose from 395 MW at quarter-end to 437 MW by mid-July. Capital expenditure builds assets for future periods, so the spending figure is not a measure of that quarter’s operating loss. It does, however, show why funding the build-out belongs alongside revenue growth in any assessment. Core Scientific’s second-quarter results.
The business model matters too. Leasing a powered facility and selling GPU cloud services leave different equipment and utilization risks with the operator. IREN’s August 27 results highlighted $1 billion of operating annualized run-rate revenue as of August 26. Its notes define that metric using contracted GPU-hour pricing for commissioned GPUs annualized across a year, with storage and ancillary revenue, and warn that recognized revenue may be materially lower. A run-rate metric is useful context, but it cannot replace the income statement. IREN’s FY26 results and metric definitions.
The economic trade-off is therefore broader than comparing revenue per megawatt. Longer commitments may improve visibility, while construction overruns, borrowing costs, customer concentration and hardware replacement can consume the apparent advantage. The relevant comparison is what remains after delivering the promised service and paying for the assets behind it.
Follow the mining machines separately
For Bitcoin, the key question is whether displaced machines stop hashing, move elsewhere or are replaced by more efficient equipment. New AI construction on previously unused land need not remove any existing mining capacity. A conversion can remove machines locally without producing an equal reduction globally.
If mining capacity leaves and no other capacity replaces it, blocks initially arrive more slowly on average. Bitcoin adjusts mining difficulty every 2,016 blocks to move block production back toward its target pace. That adjustment helps the network continue operating; it does not recreate the departed computing power. Bitcoin’s proof-of-work documentation.
Remaining miners can benefit from lower difficulty. Hashprice measures expected mining revenue for a unit of hashpower over time, commonly dollars per PH/s per day. Holding Bitcoin’s price, subsidy and transaction fees constant, lower difficulty raises expected revenue per unit of hashpower. That mechanism can encourage other operators to add capacity, limiting the lasting effect of an exit. Luxor’s hashprice methodology.
Security conclusions therefore require evidence about net hashpower, who controls it and the resources available to a potential attacker. AI revenue itself contributes no Bitcoin proof of work. A healthier company balance sheet could support continued mining, or management could choose to deploy the money entirely outside Bitcoin.
As the transition develops, the revealing disclosures will pair delivered, billable capacity with cash requirements and explain what happened to retired mining fleets. AI hosting may reward an operator’s ability to secure and develop power. Bitcoin’s outcome will depend on where mining ultimately remains economical, and who continues doing it.