Core Scientific lands 2.5GW AI infrastructure deal with AMD months after rejecting $9B CoreWeave bid

Core Scientific Partnership with AMD Reshapes AI Infrastructure Landscape

Nine months after shareholders rejected a roughly $9 billion takeover by CoreWeave, Core Scientific has secured what may be its most significant validation yet—a massive AI data center capacity agreement with AMD that could eventually reach 2.5 gigawatts 1

1

. The deal announced Tuesday positions the former bitcoin miner as a critical landlord for AMD's chips as the chipmaker works to challenge Nvidia's dominance in artificial intelligence.

At the core of the arrangement sits a set of long-term leases covering 529 megawatts of critical IT capacity across five data center sites in Texas, Oklahoma, Alabama, and Georgia 2

2

. AMD directly took on 377 megawatts at Core Scientific facilities in Pecos, Hunt County, Texas, and Muskogee, Oklahoma, while an undisclosed cloud provider backed by AMD signed agreements for 152 megawatts at locations in Auburn, Alabama, and Dalton, Georgia 2

2

. These 15-year leases include three five-year renewal options, potentially stretching the relationship across three decades with revenue expected to start flowing in 2027 1

1

.

Warrants and Expansion Options Sweeten the Deal

What elevates this from a sizeable colocation deal into a landmark partnership is the equity component and expansion rights bolted onto the agreement. AMD received warrants to purchase up to 30 million Core Scientific shares at $23.47 per share, with about 6.5 million vesting immediately when the initial leases were signed 2

2

. The warrants vest at 12,222 shares for every megawatt AMD brings online, tying the chipmaker's upside directly to the fortunes of its infrastructure partner .

Beyond the initial commitment, AMD secured the right to reserve up to 1,925 megawatts of additional capacity through December 2028, which is how the partnership arrives at its headline 2.5 gigawatts figure 5

5

. Core Scientific said the leases could generate more than $14 billion in base contracted revenue, pushing its total contracted revenue past $24 billion across all customers 3

3

. The capacity will run AMD's Instinct GPUs, EPYC CPUs, and ROCm software, giving AMD guaranteed real estate at a moment when it strains to be taken seriously against Nvidia in both training and inference workloads .

Pivot from Bitcoin Mining Gains Momentum

The AMD deal accelerates Core Scientific's dramatic transformation from cryptocurrency mining to AI data center hosting. In the second quarter, the AI colocation business brought in $136.7 million—representing 83% of Core Scientific's $164.2 million in total revenue—while its self-mining business saw revenue drop 66% to $21.5 million 2

2

. Colocation gross margin expanded to 59% from just 11% a year earlier, highlighting the profitability of the AI-focused infrastructure business 4

4

.

Core Scientific has been actively closing the door on its old identity. The company unwound its contract with Block for bitcoin-mining chips, taking a $41.9 million charge in the process 2

2

. Once one of the world's largest publicly traded Bitcoin mining companies, Core Scientific now maintains a comparatively modest Bitcoin treasury of fewer than 1,000 BTC while generating the bulk of its revenue from high-performance computing and AI colocation services 3

3

.

Market Response Reflects Sector Tensions

Core Scientific shares jumped as much as 10% on the news, even as AMD slipped more than 5%, suggesting investors viewed the deal differently for each company 1

1

. The split reaction came against a backdrop of broader concerns about AI infrastructure financing, with Asian chip stocks sliding on worries about Chinese competition and credit-default swaps tied to major AI infrastructure spenders hitting record highs 5

5

.

For AMD, the partnership represents another step in its multi-year effort to build credibility against Nvidia through acquisitions and strategic partnerships. The company recently announced a partnership with Anthropic to deploy up to 2 gigawatts of Instinct MI450 Series GPUs in Helios rack-scale systems and expanded a partnership with Microsoft to deploy its AI systems on Azure 2

2

. Locking in power and physical space ahead of demand removes one of the harder bottlenecks in scaling AI deployments.

Vindication for Rejected CoreWeave Offer

The deal caps what Core Scientific described as a strange and quietly triumphant year. The company spent much of 2025 as CoreWeave's target after the GPU cloud offered roughly $9 billion in an all-stock deal to fold the miner's power-hungry sites into its own operations 1

1

. Core Scientific investors were unconvinced the price captured the value of the company's land and power assets, and on October 30 they voted down the merger, forcing termination of the agreement 1

1

.

The two companies are not fully estranged, however, as Core Scientific still hosts CoreWeave's GPUs under a separate deal worth several hundred megawatts 1

1

. Instead of selling itself to one AI company, Core Scientific is now leasing power to several hyperscale computing customers, a diversification strategy that shareholders bet would prove more valuable than accepting CoreWeave's offer 5

5

. The company that could not persuade its owners to sell for $9 billion is now, on paper, worth considerably more on its own.

Credit Support Signals AMD's Ecosystem Commitment

One detail worth watching: 152 megawatts of the new capacity goes to neocloud operators rather than to AMD directly, and the chipmaker has agreed to provide credit support should one of those tenants default 1

1

. AMD is partly backstopping the smaller clouds that will run its hardware, a measure of how far the company will go to seed an ecosystem around its chips and compete with Nvidia's established developer network.

If AMD exercises its option on the additional 1,925 megawatts before the 2028 deadline, it will confirm real customer demand behind the buildout rather than just contracted capacity sitting on paper 5

5

. For now, raw chip performance is no longer the only race that matters in AI—every major chipmaker needs guaranteed power and physical space to deploy hardware at scale.