AMD Locks 529MW From Core Scientific in $14B Lease — Infrastructure Is the New AI Moat
AMD has secured more than 529 megawatts of US AI data center capacity from Core Scientific under a set of 15-year leases that could generate more than $14 billion in base contracted revenue for the operator. It is the largest infrastructure commitment AMD has ever made and one of the largest AI colocation deals announced in 2026.
What AMD Is Actually Buying
AMD is not renting cloud capacity or buying servers. It is locking up the physical layer: land that already has power, substations already connected to the grid, and fiber already in the ground. Core Scientific’s sites are operational and permitted. AMD is paying $14 billion to secure first claim on that capacity for the next decade and a half.
That is a different kind of asset than a GPU contract. GPU contracts expire. Compute capacity can be reallocated. A 15-year lease on 529MW of connected data center space cannot.
The NVIDIA Comparison
NVIDIA does not own this layer and has never needed to. NVIDIA’s real competitive moat is CUDA, a proprietary software ecosystem with an 18-year head start and roughly 85% of the AI GPU market. The infrastructure question — where the compute lives — has always been someone else’s problem.
AMD’s approach is the inverse. Its GPU software ecosystem (ROCm) has been closing the gap with CUDA slowly, but the software moat remains real. Rather than continuing to compete on software lock-in, AMD is building a physical moat: secured capacity at scale that other chip companies and cloud providers will need to negotiate for.
The calculation appears to be that owning the physical layer creates leverage. Enterprises that want AMD GPUs in production will find AMD-anchored data center capacity at the other end of the supply chain.
Core Scientific’s Position
For Core Scientific, this is a validation of its 2025 pivot from Bitcoin mining to high-performance computing colocation. The company had been building out AI-ready infrastructure at its US campuses in Texas, North Dakota, and elsewhere. A 15-year commitment at $14 billion in base revenue is not a demand signal. It is an anchor tenant.
Core Scientific’s stock has been tracking the AI infrastructure buildout since early 2026. The AMD deal removes the uncertainty about whether its capacity will find buyers. At contracted rates across 529MW over 15 years, the base revenue works out to roughly $1.75 per watt-year — competitive with other announced colocation agreements in the current market.
Infrastructure as Strategy
The broader trend running through this deal: the companies that control physical capacity are gaining leverage that chip designers historically did not need to seek. Amazon locked TeraWulf for 1GW in Kentucky for 20 years at $19 billion in March. Anthropic moved to own its data centers as Google backed leases for a 1GW US build. CoreWeave closed an $8.5 billion investment-grade GPU loan secured against its own physical assets.
AMD joining this cohort with a 529MW commitment means the infrastructure accumulation is no longer just a hyperscaler pattern. Chip companies are playing the same game. The next constraint in AI will not be the model. It will be who controls the power and the cooling.