Securing 1.2 GW Power Capacity
Nscale
This deal shows that in AI infrastructure, power is becoming as important as GPUs. CoreWeave already has the customer backlog and Nvidia relationship, but buying Core Scientific turns a big piece of its capacity stack from a lease into owned infrastructure. That matters because AI data centers are constrained by megawatts, interconnects, cooling, and time to energize, not just by how many chips can be ordered.
-
CoreWeave said the acquisition would give it about 1.3 GW of gross power across Core Scientific sites, plus more than 1 GW of expansion potential. It also said owning the sites would remove more than $10B of future lease overhead, which makes the move about cost control as much as raw capacity.
-
Core Scientific was already turning its former Bitcoin mining footprint into AI hosting. In 2024 it said it had 1.2 GW of contracted power and could deliver nearly 500 MW of HPC capacity, which made it a ready made conversion target for a GPU cloud that needed fast access to energized sites.
-
This is the same broad playbook seen across the category, but with different power sources. CoreWeave is buying grid connected data center capacity, Crusoe builds near stranded and renewable energy, and Nscale builds hydro powered sites. The common pattern is that winning clouds now secure electricity first, then layer GPUs and software on top.
The next phase of competition will be shaped by who can lock up power corridors and turn them into usable AI capacity fastest. As GPU supply gradually loosens, the hardest asset to replicate will be permitted, cooled, and financed megawatts, which pushes the market toward more consolidation between GPU clouds, data center operators, and legacy crypto infrastructure owners.