Bitcoin Miners Are Selling Coins to Buy GPUs as AI Hosting Overtakes Mining Revenue

Bitcoin mining has spent a decade being described as a business about one thing: converting electricity into hash rate. That description is quietly going out of date. CoinShares projects that AI hosting and compute could account for roughly 70% of listed miners' revenue by the end of 2026, up from about 30% at the start of the year — a shift large enough that calling these companies "bitcoin miners" is becoming a misnomer for several of the biggest names on the Nasdaq.
The clearest sign of how seriously the pivot is being taken: MARA Holdings sold approximately $1 billion worth of its bitcoin treasury to fund its move into AI infrastructure — a company that spent years building a balance sheet specifically to hold bitcoin, liquidating a meaningful chunk of it to bankroll a different business entirely. Across the sector, public miners reportedly sold around 32,000 BTC in the first quarter of 2026 alone to fund the transition, according to industry trackers.
Why Miners Have a Head Start Nobody Else Has
The pivot isn't really about bitcoin mining companies discovering AI — it's about AI companies discovering that bitcoin miners already own the one thing the AI buildout is short on: power. Mining operations sit on sites with existing grid interconnections, substantial contracted power capacity, and land entitlements that would take a hyperscaler years to replicate from scratch. For a data center developer facing multi-year interconnection queues in most US grid regions, a miner's existing 200-megawatt site with power already flowing is worth far more than its hash rate.
That's the asset hyperscalers are buying when they sign colocation and lease deals with miners. TeraWulf has built its strategy explicitly around splitting capacity between bitcoin mining and AI colocation, hedging between the two revenue streams rather than picking one. It's a structurally sound position: mining demand for the site's power is flexible (miners can throttle down when the grid is stressed or power prices spike), which is exactly the kind of flexible load that makes a site attractive to utilities approving new AI data center interconnections in the first place.
The Catch: Power Access Isn't the Same as AI-Ready Infrastructure
The transition is harder than "flip a switch from mining rigs to GPUs" implies. Hyperscalers evaluating a converted mining site care about power density, redundancy, and uptime guarantees that most mining facilities were never engineered for. A typical bitcoin mining data hall is built around air-cooled ASICs tolerant of power fluctuations and brief outages — acceptable because a few minutes of downtime just means a few minutes of un-mined blocks. An AI training cluster running a multi-week job across thousands of GPUs has none of that tolerance; a power interruption can corrupt a checkpoint and cost days of compute.
That mismatch is why the pivot is playing out unevenly. Core Scientific's September 2026 operating data shows it mining 9% less bitcoin even as it expanded its hash rate 2.4% — a company visibly shifting capacity and attention away from pure mining optimization while it retrofits sites for higher-value, non-mining tenants. The retrofit itself (liquid cooling, redundant power paths, higher-density racking) is a real capital outlay, not a paper transition, and it's part of why the miners best positioned to benefit are the ones that already had access to larger, newer power contracts rather than older fleets sized purely for ASIC density.
What It Means for Bitcoin's Network, and for the AI Compute Market
For bitcoin, the concern is network security economics: as the largest public miners redirect capital and site expansion toward AI rather than additional hash rate, the long-run growth curve of global hash rate — and the capital available to defend the network against a hash rate attack — becomes more dependent on private and smaller operators who face the same AI-pivot incentive without the balance sheet to act on it.
For the AI compute market, the pivot is a genuine new supply source at a moment when GPU capacity and grid interconnection are the two tightest bottlenecks in the industry. Converted mining sites won't replace purpose-built hyperscaler campuses, but they offer something scarce: megawatts that are already contracted and substations that are already energized. In a market where the wait for new grid interconnection routinely runs into years, that head start is worth more than the hash rate it used to produce — which is exactly why miners are selling bitcoin to buy GPUs instead of ASICs.