Core Scientific Doubles AI Capacity to 1.1 GW in $14B AMD Deal

Core Scientific has doubled its leased AI data center capacity to approximately 1.1 GW after signing a 15-year infrastructure agreement with AMD covering 530 MW across five campuses and representing more than $14 billion in potential base contracted revenue. Announced alongside the company's second-quarter results on July 28, the agreement begins in 2027 and grants AMD exclusive rights to reserve up to an additional 2 GW of future capacity. The initial deployment will use Core Scientific facilities to host infrastructure built around AMD Instinct GPUs, EPYC processors, and the ROCm software platform for cloud providers, AI model developers, and enterprise customers.

"Core Scientific's extensive portfolio of AI-ready data centers expands access to the infrastructure our customers need to deploy AMD AI solutions at scale," said Mathew Hein, AMD's chief strategy officer for corporate development.

Industry analysts characterize the agreement as addressing a fundamental shift in AI competition. "It's not about selling accelerators at this point, it's about selling a deployment path," Matt Kimball, vice president and principal analyst for data center technologies at Moor Insights & Strategy, told Data Center Knowledge. Kimball noted that AMD appears to be using its balance sheet to reduce deployment risk by supporting infrastructure capable of hosting high-density AI clusters, where power, liquid cooling, and specialized facilities have become as critical as the chips themselves.

The touted $14 billion represents base contracted revenue over 15 years rather than a single booking, while the additional 2 GW reflects reservation rights rather than firm customer commitments. According to Steven Dickens, CEO and principal analyst at HyperFrame Research, "The sheer scale and magnitude of the infrastructure needed for AI is driving innovative and creative collaborations to drive financing and new debt structures. We are seeing this with GPUs becoming collateralized assets and SPVs being created to structure data center build-outs."

Dave McCarthy, research vice president for cloud and edge services at IDC, observed a broader shift: "The primary bottleneck in AI deployment has fundamentally shifted from silicon supply to power access and data center footprint. It's no longer enough for vendors to simply sell high-performance processors. They must ensure the physical runway exists for customers to deploy them at scale."

The AMD agreement expands Core Scientific's contracted AI portfolio to approximately 1.1 GW of leased customer power capacity, representing more than $24 billion in potential contracted revenue. As of mid-July, the company was billing customers for 437 MW of capacity, equivalent to an annualized colocation revenue run rate of approximately $635 million. Core Scientific now has two AI customers, each committed to more than 500 MW across five campuses, providing greater customer diversification.

Chief Operating Officer Matt Brown disclosed that fully delivered AI data center capacity is expected to cost approximately $11 million to $12 million per megawatt, including construction, electrical infrastructure, cooling systems, commissioning, and utility integration. At that rate, the initial AMD deployment represents roughly $6 billion of infrastructure investment.

During the company's earnings call, CEO Adam Sullivan stated: "Value is not announced. It is delivered." Construction is underway at Core Scientific's Pecos, Texas campus, with deliveries scheduled to begin in the first half of 2027. Development is also progressing at Hunt, Texas; Muskogee, Oklahoma; Auburn, Alabama; and Dalton, Georgia, with deployments continuing through 2028.

Core Scientific's transformation has been dramatic. Colocation revenue climbed to $136.7 million in the second quarter from $10.6 million a year earlier, while digital asset self-mining revenue fell to $21.5 million as the company shifted power from bitcoin mining to AI workloads. Sullivan said Core Scientific has identified more than 2 GW of additional development opportunities that could begin coming online between late 2028 and 2030.

The agreement demonstrates that competitive advantage in AI infrastructure is no longer defined solely by accelerators. Vendors are increasingly competing on their ability to provide a complete deployment path that includes power, liquid cooling, and energized capacity years before it comes online.