Anthropic signs a 20-year, ~$19B lease to use a TeraWulf data center in Kentucky, set to have a ~400MW capacity and to start delivering power in H2 2027
WULF year to date — Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
Context & Ripple Effects
Anthropic has been assembling compute capacity through several channels: a multiyear CoreWeave arrangement, agreements with Google and Broadcom for multiple gigawatts of TPU capacity, and, according to prior reporting, more than a dozen initial direct data-center lease agreements. The TeraWulf lease extends that shift toward directly securing physical infrastructure.
The planned Kentucky site also sits alongside a broader buildout of large U.S. AI campuses, including Meta’s planned 1GW Indiana facility. The key constraint being addressed is not only access to accelerators but access to powered, deployable data-center capacity.
First-order effects
- Anthropic gains a long-duration claim on a planned ~400MW Kentucky facility, with initial power delivery targeted for H2 2027, reducing its reliance on any single external compute provider.
- TeraWulf obtains a large, long-term customer commitment tied to development of the site, making its data-center capacity more directly exposed to AI demand.
Second-order effects
- Direct leases give Anthropic another bargaining lever in its infrastructure relationships with cloud and specialized compute providers such as Google and CoreWeave, while requiring it to coordinate a more complex portfolio of capacity commitments.
- Large commitments for powered sites raise the value of data-center operators that can deliver power on schedule; competing AI developers may seek similarly long-dated capacity rather than wait for conventional cloud availability.
Third-order effects
- If direct leasing becomes a standard procurement route for frontier-model developers, the AI infrastructure market could split more clearly between providers of chips/cloud services and owners or developers of power-ready physical campuses.
- The limiting factor in AI expansion increasingly shifts from model demand alone to the ability to finance, build, and energize high-capacity sites—making power-delivery timelines a central competitive variable.
The trend: AI companies are moving from buying compute capacity to locking in long-term control over the data-center power and real estate needed to operate it at scale.