The Miner That Tripled Its AI Cloud Run-Rate in 60 Days

The story of AI in 2026 is really a story about electricity. Every major hyperscaler is racing to lock down gigawatts of grid-tied power for the next generation of training and inference clusters. Utilities are quoting interconnection queues that stretch three to seven years for new data center capacity in most major U.S. markets. The bottleneck on AI is no longer chip supply or capital. It is whether you can plug your servers in.

That bottleneck has quietly created an unusual class of beneficiary: companies that already own large amounts of energized, grid-tied capacity built for a different purpose. Bitdeer (NASDAQ: BTDR) is one of the cleanest examples. The company controls roughly 3 gigawatts of electrical capacity across four continents, with 1,744 megawatts already energized and operating. That power was originally built to run Bitcoin mining hardware, and it’s now increasingly being redirected toward AI workloads.

The most concrete read on their pivot is the AI cloud business, which is growing fast enough to register against the company’s larger mining segment. Bitdeer just reported approximately $69 million in annualized recurring revenue from its AI cloud at the end of April, up roughly 60% from $43 million in March and more than triple the $21 million figure from February. The GPU fleet almost doubled from 2,128 deployed at the end of March to 4,184 by the end of April, with utilization holding above 90% and roughly 80% of those GPUs under contract to external customers. 

Recent deliveries include NVIDIA’s latest-generation nodes in Malaysia and in the U.S. SemiAnalysis, the semiconductor research firm that publishes the closely-watched ClusterMAX rating of GPU cloud providers, included Bitdeer AI in its latest cohort at the Bronze tier, providing external validation of the platform's GPU cloud capabilities.

The colocation strategy is the larger, slower-moving piece. Of Bitdeer’s 1,259 megawatts of pipeline capacity, the majority is being positioned for AI and colocation. Tydal, Norway, is the site management has flagged as furthest along in tenant discussions. Rockdale, Texas (563 MW online today, plus 179 MW in pipeline) is under active evaluation for an AI transition. Niles, Ohio (300 MW grid-interconnected, targeting Q4 2028 energization) was developed with colocation in mind from the start. Clarington, Ohio (570 MW under contract with a local utility) adds further AI-suitable capacity, though timing depends on the resolution of ongoing legal proceedings. Two of the U.S. sites (Knoxville and Wenatchee) are also being converted from mining to AI cloud, with phase-one data center design work underway and core equipment beginning to arrive.

But what about the Bitcoin mining business? The mining business is the operating-cash-flow engine that anchors the AI buildout. Bitdeer reported first-quarter 2026 self-mining revenue of $146.9 million, up 295% year-over-year. Adjusted EBITDA flipped from a loss of $45.6 million a year ago to a positive $14.4 million. Self-mining hashrate is up roughly 5x from last year. That’s mainly because Bitdeer designs and manufactures its own chips. 

The annualized AI cloud revenue doesn’t represent a large portion of its $620 million total revenue, and the colocation deal is yet to be closed. But the trajectory is clear, and the data is moving fast.

The AI infrastructure stack is increasingly going to be assembled by operators who can vertically integrate four things at once: chip design, power procurement, data center construction, and GPU operations. The hyperscalers can do all four because they have unlimited balance sheets, but only a small handful of public miners are quietly demonstrating they can do most of it on a fraction of the balance sheet. The market is paying huge premiums for contracted AI infrastructure plays, and some efficient miners are demonstrating they can do most of it without hyperscaler capital

Bitdeer has gigawatts of energized power, a fast-growing GPU cloud with named hardware and external contracts, a colocation pipeline aimed at AI tenants, and a profitable underlying business funding the buildout. It is a combination the market has not seen often, and one worth watching as the AI power build moves into its next phase.