Core Scientific and AMD announce a deal under which AMD will secure 500+ MW of US data center capacity starting in 2027, with the ability to scale up to 2.5 GW
Quick Take — AMD will secure more than 500 megawatts of Core Scientific data-center capacity beginning in 2027 …
Context & Ripple Effects
AMD’s capacity reservation follows its recently reported up-to-2 GW MI450 supply commitment with Anthropic beginning in 2027, linking expected accelerator deployments to a defined U.S. data-center footprint. It also extends AMD’s earlier 500 MW AI-compute partnership with Humain, which included U.S. facilities.
For Core Scientific, the agreement advances the company’s stated shift from crypto mining toward AI data-center leasing, a transition it previously sought to finance through a planned $3.3 billion bond sale. The deal gives that repositioning a named semiconductor customer rather than only a prospective infrastructure narrative.
First-order effects
- AMD secures more than 500 MW of U.S. capacity from 2027, with contractual room to expand to 2.5 GW; this gives its data-center roadmap a dedicated deployment option alongside chip sales.
- Core Scientific gains a major prospective tenant for its AI-oriented capacity, tying part of its future buildout to AMD’s demand.
Second-order effects
- The reservation makes available powered capacity at Core Scientific scarcer for other customers and raises the importance of long-duration capacity commitments in competing for large AI deployments.
- AMD can pair infrastructure access with its accelerator and server offerings, strengthening its ability to support customers whose rollout plans depend on both hardware and sites.
Third-order effects
- If similar agreements proliferate, chip suppliers will increasingly compete through secured power and data-center access—not only processor performance—pushing the market toward integrated compute-and-capacity offerings.
- The model shifts more execution risk to long-lead infrastructure delivery: the value of large compute commitments will depend on whether power, facilities, and hardware arrive on coordinated schedules.
The trend: AI infrastructure is moving toward long-term compute offtake arrangements that lock in powered capacity years before the associated hardware is deployed.