How crypto miners are reinventing themselves as AI inference providers
“AI is the new industrial base, and compute is the factory floor.”
Cryptocurrency mining. By kos Stiller/Bloomberg via Getty Images
- Crypto miners are turning power sites and grid connections into AI data center assets.
- CoreWeave shows the path, moving from Ethereum mining to AI cloud services.
- Inference demand is growing because AI apps need fast responses every time users interact.
- The pivot needs major upgrades, including GPUs, cooling, fiber, and backup power.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
Bitcoin miners were ahead of the curve in sourcing electricity supplies, and now they’re capitalizing on the AI revolution.
CoreWeave has become a central part of the generative AI revolution. But it began life in 2017 as Atlantic Crypto, using graphics processors to mine ethereum.
Its stock market prospectus says that, before 2022, most of its limited revenue came from cryptomining services it has since abandoned. The company now sells access to a purpose-built AI cloud for training models and running inference.
CoreWeave’s transformation has become a template for an industry squeezed by volatile cryptocurrency prices, rising mining difficulty, and the periodic halving of Bitcoin rewards. IREN, HIVE Digital Technologies, Hut 8, and Bit Digital have all moved into AI computing or data center leasing, while Core Scientific agreed to supply AMD with an initial 530 megawatts of AI-ready capacity from 2027. The deal could eventually expand to 2.5 gigawatts.
So why are cryptominers suddenly in vogue? The reason is the supply squeeze for power that AI companies have encountered, and what those who mined bitcoin and other coins can offer.\
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
Firm foundations
Crypto miners spent years securing land, substations, and grid connections for businesses that consume electricity around the clock so that they could make their crypto fortune. The same principle of always-on, high-speed access to servers is important for AI companies – and with AI developers struggling to find powered data center space, those connections may now be worth more than the coins the facilities were built to mine.
However, some changes are needed. The specialist mining machines used to hunt for crypto are not as good at AI training. Bitcoin is produced using application-specific integrated circuits, or ASICs, built to perform one narrow calculation.
AI requires expensive GPUs, alongside fast networking, dependable fiber connections, backup power, and far more sophisticated cooling.
IREN acknowledges in its annual filing that converting existing sites can require “significant capital expenditures” and cause delays or outages. The company is installing direct-to-chip liquid cooling at its Childress site in Texas and is exploring the replacement of some bitcoin equipment with GPUs.
Its cloud service, launched in 2024, gives customers remote access to Nvidia processors for training and inference workloads.
Innovation in inference
Inference – providing responses rather than training new models entirely – is a safer bet. Training a new model is an enormous but intermittent job. Inference occurs every time someone generates an image, questions a chatbot, or deploys an AI agent, and it requires the speed of response that lends itself well to pre-existing tech stacks.
CoreWeave’s prospectus estimates that inference infrastructure could become a $49 billion market by 2028, and says production workloads should account for an increasing share of AI computing.
Other former miners are chasing that demand. Bit Digital reported in early 2025 that its cloud-services revenue had risen 84% year on year, while bitcoin mining revenue fell 64%. Its AI operation has since become the separately listed WhiteFiber, whose GPU cloud supports both model training and inference.
Meanwhile, HIVE is converting a former mining facility in Sweden to higher-grade AI standards and expanding its BUZZ cloud division. It signed a 3-year GPU contract worth about $220 million in June and plans a Toronto-area campus designed for low-latency inference.
“AI is the new industrial base and compute is the factory floor,” HIVE executive chairman Frank Holmes said when announcing the project.
The pivot is not guaranteed to work. Many mines sit beside cheap electricity but far from customers, fiber routes, and skilled workers. Yet power has become AI’s scarcest raw material.
Crypto miners may not be able to turn their ASICs into inference engines – but some can certainly turn the land and electricity behind them into something far more valuable.