Keel’s all-in bet on AI power

Keel reported that the shutdown of its U.S. bitcoin-mining operations was complete. During the transition it sold 1,085 BTC for $75 million and retained 1,861 BTC. The company is now presenting those U.S. sites as infrastructure for AI and high-performance computing rather than as mines waiting for a better point in the bitcoin cycle.

That distinction is important. Companies can describe a diversification plan for years while the original operation remains the economic centre of gravity. A completed shutdown removes that ambiguity in the United States. Keel has given investors a clearer test: the converted sites must secure customers and produce revenue that can replace what mining once supplied.

The second-quarter figures show the cost of reaching that point. Keel reported $30 million in revenue and a $141 million operating loss. Those numbers do not reveal the ultimate economics of the AI plan, but they make the starting position visible. This is a transition carrying real cost before the proposed destination has been proven.

Bitcoin mining and AI computing begin with a shared constraint: both demand access to substantial electricity. Keel’s energised sites may therefore offer a head start that a developer without land, grid access or power agreements would have to spend time obtaining. That underlying power position is the strategic reason the pivot is plausible.

The similarity should not be stretched too far. AI and high-performance-computing customers can demand different buildings, cooling, networks, uptime standards and service arrangements. A site capable of running mining machines is not automatically ready to host the workloads of an external computing customer. Conversion is an engineering and commercial programme, not a new label on the same equipment.

The value of Keel’s sites will consequently depend on two linked deliveries. The company has to turn available power into customer-ready capacity, and it has to persuade customers to sign for that capacity on viable terms. One without the other produces either stranded infrastructure or contracts that cannot be served.

A mining business can be tracked through deployed computing power, bitcoin production, energy cost and the market price of the coins it creates. An infrastructure business needs a different scorecard. Signed megawatts, conversion expenditure, delivery dates, utilisation, recurring revenue and operating margin become the measures that show whether the pivot is advancing.

Announcements are not the same as contracted demand. A plan to serve AI customers has economic weight only when a named amount of capacity is committed, the site is ready and revenue begins. Investors should be wary of blending potential capacity with active capacity, or an expression of customer interest with a binding agreement.

The retained 1,861 BTC also deserves to remain separate in the analysis. It may provide liquidity or market exposure, but a change in bitcoin’s price cannot by itself validate the new operating model. Future reporting will be most useful when disposals of bitcoin, any remaining mining outside the U.S. and AI-related revenue are clearly distinguished.

Milestones should be comparable from one reporting period to the next. If Keel defines available, contracted and operational megawatts consistently, readers can see capacity progress through the conversion funnel. Changing labels or combining proposed capacity with revenue-producing sites would make a capital-intensive transition harder to judge at the moment transparency matters most.