A neocloud is a cloud-computing company built almost entirely around one product: renting out the specialized chips — GPUs — that train and run AI models. Instead of selling hundreds of general-purpose services the way Amazon Web Services or Microsoft Azure do, a neocloud sells raw, high-performance GPU capacity, fast. Demand for that capacity has grown so quickly that companies with no history in cloud computing at all — including bitcoin miners — are converting their facilities into neoclouds and signing multi-billion-dollar leases with AI labs.
What makes a neocloud different from a hyperscaler
The large “hyperscale” clouds — AWS, Microsoft Azure, Google Cloud — offer everything from databases to email hosting to GPU rental, with GPUs as one line item among hundreds. A neocloud does the opposite: it strips the catalog down to bare-metal or lightly virtualized GPU access, fast networking between chips, and little else. CoreWeave, the best-known example, now runs some of the largest GPU clusters in the world for AI labs and has become a public company valued in the tens of billions of dollars.
That narrow focus lets neoclouds move faster and charge less. Industry infrastructure researchers commonly cite neocloud pricing as 70–85% below hyperscaler list prices for comparable GPUs, with clusters provisioned in minutes rather than the months-long waitlists that have become common at the big three clouds during the AI boom.
Why bitcoin miners are becoming neoclouds
The pattern isn’t new — CoreWeave itself started in 2017 as an ethereum-mining operation called Atlantic Crypto, then repurposed its GPU inventory into a cloud business after the 2018 crypto crash. The same logic is now playing out across the bitcoin-mining industry. A large mining operation already has the assets an AI data center needs most: a grid interconnection (often the hardest and slowest part of building a data center, taking utilities years to approve), industrial-scale power capacity, cooling systems, and staff experienced in running dense compute non-stop.
The economics pushed the shift too. Mining rewards were cut in half by the 2024 bitcoin halving, tightening margins for miners that depend on volatile coin prices and rising network difficulty. A long-term hosting contract with an AI lab — often 10 to 20 years — offers steadier, contracted revenue instead. TeraWulf, a bitcoin miner that has increasingly redirected its power capacity toward AI hosting, is one of several miners that have pursued this route.
How the deals get built
These arrangements are usually structured as long-term power and hosting leases rather than simple equipment rentals: the AI company commits to paying for a fixed amount of megawatts over many years, and that contracted revenue lets the miner (or a partner) borrow the money needed to actually build the data center. Sometimes a third party backs the project directly — Google, for example, has taken an equity stake in a TeraWulf data-center joint venture with the AI infrastructure firm Fluidstack in exchange for backing the lease financially.
Why it matters
For AI labs, leasing from a neocloud — miner-turned or otherwise — is often the fastest way to secure power and floor space without waiting years to build and permit a data center themselves. For the miners, it can mean trading a volatile, commodity business for predictable, contracted income. But it also concentrates a lot of a mining company’s future revenue in one or two large customers, and it means a facility once sized for mining rigs now has to draw far more power from local grids — a strain that has already drawn scrutiny from utilities and nearby communities in several states.
In the news
The pattern was on full display when Anthropic agreed to lease a 401-megawatt AI computing campus in Hawesville, Kentucky, from TeraWulf under a 20-year, roughly $19 billion deal — our report on the lease has the details of that specific agreement.
FAQ
Is a neocloud the same thing as a hyperscaler? No. Hyperscalers sell a broad menu of cloud services; neoclouds specialize almost entirely in GPU compute for AI and high-performance computing.
Why would an AI lab lease from a former bitcoin miner instead of building its own data center? Because the miner typically already holds a hard-to-get grid connection and power capacity, which can otherwise take years to secure from a utility.
Are all neoclouds former crypto miners? No, but the pattern is well established — CoreWeave, today’s largest and best-known neocloud, itself began as an ethereum-mining operation before pivoting to GPU cloud computing in 2019.
Is this only happening in the United States? The largest publicized deals have been in the US, but GPU-focused cloud providers now operate in Europe, the Middle East, and Asia as well.
Sources: Hashrate Index, “What Is a Neocloud?”; CoinDesk and Data Center Dynamics reporting on TeraWulf’s AI hosting agreements; TeraWulf investor press releases.