The AMD Core Scientific Deal: A $14 Billion Bet on Independent AI Infrastructure

Advanced Micro Devices has placed a $14 billion wager on an independent AI infrastructure future by leasing more than 500 megawatts of data center capacity from former bitcoin miner Core Scientific, with options to scale that footprint to 2.5 gigawatts. The 15-year agreement, announced alongside Core Scientific's second-quarter results on July 28, is both AMD's most aggressive bid yet to break the Nvidia-hyperscaler lock on AI compute and the definitive arrival of bitcoin miners as a new asset class in the AI data center market.

The AMD Core Scientific deal covers 529 megawatts of critical IT capacity across five campuses in Texas, Oklahoma, Alabama, and Georgia. AMD directly leases approximately 377 megawatts through triple-net leases, while an undisclosed neocloud provider backed by AMD's credit accounts for the remaining 152 megawatts. Lease terms run 15 years with extension options that could stretch to 30 years. Customer deployments are expected to begin in 2027.

Core Scientific values the initial contracts at more than $14 billion in base contracted revenue, pushing the company's total contracted revenue past $24 billion. Build-out costs for the capacity are estimated at $11 million to $12 million per megawatt, a figure that reflects the high power density required for AI workloads.

AMD's Infrastructure Playbook

For AMD, this partnership solves a structural problem. While the company's Instinct GPU line has won design wins at major cloud providers, AMD has lacked the dedicated, AI-optimized data center footprint that Nvidia's largest customers (the hyperscalers themselves) can offer through their own infrastructure. By pre-leasing capacity at Core Scientific's sites, AMD can offer its enterprise and neocloud customers a turnkey deployment path built around AMD Instinct GPUs, EPYC processors, and the ROCm software platform.

The arrangement also includes market-priced warrants for 30 million Core Scientific shares, with 6.5 million shares vesting immediately upon lease execution. That equity stake gives AMD a financial interest in Core Scientific's success beyond the lease terms, aligning the two companies as the infrastructure is built out. AMD has the right to reserve an additional 1,925 megawatts through December 28, 2028, bringing the total expansion potential to the full 2.5 GW figure cited in the announcement.

This rights structure is unusual in scale. Most chipmaker-data center partnerships involve a few hundred megawatts at most. The 2.5 GW ceiling approaches the capacity of a mid-tier hyperscaler fleet, suggesting AMD intends this relationship to function as a long-term infrastructure arm rather than a one-time capacity fix. The 15-year base term and the 30-year extension option reinforce that reading, as does the equity warrant structure that ties AMD's fortunes directly to Core Scientific's operational success.

The Bitcoin Miner Transformation

Core Scientific's pivot from bitcoin mining to AI hosting has been underway for quarters, but the AMD deal makes it definitive. The company's colocation business already generated 77 percent of revenue in the second quarter of 2026, and management is repurposing remaining mining facilities toward high-density AI colocation. The company terminated its ASIC agreement with Block earlier this year, signaling the strategic shift away from cryptocurrency.

The AMD Core Scientific deal also illustrates how bitcoin miners have matured into an AI data center asset class. Companies like Core Scientific built massive power and cooling infrastructure for crypto mining that is directly adaptable to AI workloads. The same high power density, the same need for waste heat management, and the same demand for cheap land and reliable grid access in the American South all transfer cleanly from mining to AI compute. What was once a speculative cryptocurrency bet has become a tangible AI infrastructure asset.

The five AMD sites sit in Pecos, Texas, and other locations across Oklahoma, Alabama, and Georgia. Core Scientific had already secured power capacity and land for mining operations at these locations. Retrofitting those assets for AI hosting costs less than greenfield construction, giving Core Scientific a cost advantage that pure-play data center operators cannot easily match. Build costs of $11 million to $12 million per megawatt are competitive with new data center construction in secondary markets.

Structural Implications for AI Compute

The AMD Core Scientific deal has consequences that extend beyond the two companies. For AMD, it creates an independent infrastructure channel that routes around the hyperscaler cloud providers who are also Nvidia's largest customers. Enterprises deploying AMD Instinct-based workloads now have a direct infrastructure path managed by a dedicated data center operator, without the margin stack of a cloud intermediary.

For Core Scientific, the partnership validates the thesis that mining infrastructure is a first-class asset for AI computing. Other bitcoin miners with similar power assets and land positions are likely to follow the same playbook, accelerating a trend that is already reshaping the data center construction pipeline. The U.S. data center industry faces a power availability bottleneck, and the mining sector controls a meaningful share of the interconnected capacity that can be repurposed faster than new builds can come online.

For the broader market, the deal shows that the AI compute supply chain is diversifying. Rather than routing exclusively through AWS, Azure, or Google Cloud, enterprises and model builders can access dedicated AI facilities running AMD silicon. That diversification introduces price competition, alternative hardware ecosystems, and deployment models that do not depend on hyperscaler procurement cycles.

Construction on the initial 529 megawatts will proceed through 2026 and 2027, with the first customer deployments expected to go live in 2027. Core Scientific's stock climbed more than 5 percent following the announcement, reflecting investor confidence that the pivot to AI colocation is delivering results. AMD's stock moved more modestly, as the deal's impact on the company's data center revenue will take quarters to materialize fully.

Why This Matters

This deal breaks the implicit assumption that Nvidia-powered hyperscaler clouds are the only viable path to large-scale AI inference and training. AMD has secured its own infrastructure channel with enough scale to serve serious enterprise deployments, creating genuine competition in AI compute supply. For decision-makers evaluating AI strategy, the existence of a second, independent infrastructure ecosystem built on AMD silicon and independent data center capacity changes the risk calculus around vendor lock-in, pricing leverage, and deployment timelines. Bitcoin miners turned AI hosts now sit on a strategic resource that the largest chipmaker in the world was willing to back with $14 billion and an equity stake, and that shift will ripple through data center real estate, GPU procurement, and cloud pricing for years.

Related Articles

• AMD France AI Partnership for Exascale Infrastructure
• AMD Helios rack AI system lands Microsoft as first customer
• AMD Debuts Ryzen AI Max+ Pro 495 and Pledges $10 Billion for AI Infrastructure Expansion

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team.