CoreWeave just paid $9B for a Bitcoin miner. [ANNOUNCEMENT]
They’re buying Core Scientific — yes, the 𝘉𝘪𝘵𝘤𝘰𝘪𝘯 𝘮𝘪𝘯𝘪𝘯𝘨 𝘨𝘪𝘢𝘯𝘵 — in an all-stock deal to supercharge their AI data centre empire.
Over the past few months, the traditional crypto miners are pivoting into AI data centres. And, this is one of the biggest moves to power AI.
Let’s unpack.
The Bankruptcy-to-Billions Playbook
Core Scientific filed for Chapter 11 bankruptcy in December 2022. Eighteen months later, CoreWeave—an AI cloud provider—just bought them for $9B.
What changed? The infrastructure that made Core Scientific a failed Bitcoin miner makes it a perfect AI powerhouse.
The Deal:
$9B all-stock takeover
Core Scientific’s 1.3 GW power capacity across 11 U.S. sites is now under CoreWeave’s control
Another 1+ gigawatts ready for expansion
$10B in lease savings (over 12 years), $500M in annual cost cuts by 2027
"This acquisition accelerates our strategy to deploy AI and HPC workloads at scale," said CoreWeave CEO Michael Intrator.
Translation: We just bought the power grid for the AI revolution.
The upside is massive:
AI data centre market projected to grow from $236B in 2025 to $934B by 2030
Over 40% of global data centre capacity will be AI-dedicated by 2025
Why Bitcoin miners?
Bitcoin mining and AI training share one critical requirement: massive amounts of electricity delivered to high-performance computing hardware.
The difference? The economics are night and day. And here's where it gets interesting:
HIVE Digital Technologies discovered something remarkable: 10 megawatts of NVIDIA H100 GPUs generate the same revenue as 100 megawatts of Bitcoin mining equipment.
That's 10x the revenue density.
The numbers tell the story:
AI workloads: Up to 90% EBITDA margins
Bitcoin mining: Cyclical, razor-thin returns that disappear when Bitcoin prices crash
The great mining migration
CoreWeave isn't alone. The entire Bitcoin mining industry is pivoting:
Galaxy Digital walked away from Bitcoin mining entirely, signing a $4.5B, 15-year deal with CoreWeave for AI infrastructure.
Northern Data, HIVE Digital Technologies, IREN, and Hut8 are all converting their mining rigs to AI workloads.
Global data centre investment jumped 51% year-over-year to $455B —driven largely by AI demand.
Read our deep dive below:
The AI infrastructure arms race
Big tech giants and upstart AI firms are racing to build the next generation of “AI factories” — vast data hubs packed with powerful chips that can train and run models in real time. They’re pouring billions into these facilities, not just for raw storage but for nonstop, high‑power compute.
Take xAI’s Colossus site in Memphis: a 750,000‑square‑foot former factory—about the size of 418 homes—that went from empty shell to a fully live AI centre in just 122 days. That’s half the time it usually takes to build a single house in America, and a stark sign that the AI infrastructure race is speeding up far beyond expectations.
But behind the breakneck build times sits a complex financial engine. Upfront costs come from land, power hookups, advanced chips, and next‑level cooling systems. Ongoing costs—energy bills and around‑the‑clock maintenance—are just as steep when clusters run at full tilt. Returns flow in the form of AI‑powered services, platform fees, or internal gains, yet break‑even often stretches into years.
And let’s not forget the supply chain: transformers, substations, turbines, GPUs, and high‑capacity cables can’t be conjured on demand. Securing them takes planning, contracts, and sometimes sheer luck. In this landscape, data centres are more than buildings; they’re strategic crossroads where real estate, power, logistics, and software profits meet.
AI workloads require 5-10 times more energy than traditional computing. A single AI training run can consume as much electricity as 100 homes use in a month.
Bitcoin miners solved this problem years ago. They built massive facilities in locations with cheap, abundant power. Now, AI companies are realising these "failed" crypto operations are exactly what they need.
CoreWeave's move represents a fundamental shift in cloud computing strategy. Instead of leasing capacity from data centre operators, they're buying the entire stack:
Before: Pay rent to the data centre operators
After: Own the power, own the facilities, own the margins
CoreWeave now controls 250,000 NVIDIA GPUs across 32 data centres. That's more specialised AI compute than most traditional cloud providers.
Amazon Web Services and Microsoft Azure have broader services but less specialized AI hardware. Google Cloud has advanced AI capabilities but limited GPU inventory.
Devil’s Advocate: CoreWeave came to market with huge debt and billions in lease payments. It faces significant risks:
Keep reading with a 7-day free trial
Subscribe to 51 Insights to keep reading this post and get 7 days of free access to the full post archives.