By Greysen Cacciatore, Research Associate | X | Linkedin
The repurposing of bitcoin mining infrastructure into high-performance computing (HPC) facilities for AI workloads is well underway. Prominent mining businesses, including TeraWulf and Galaxy Digital, have already signed agreements with major hyperscalers to supply over 1000 MW of power. This trend sits within a broader convergence of capital, infrastructure, and regulation that is shaping the next phase of AI development. In this article, we uncover key themes within this interplay, including the accelerating energy demands from AI, the strategic repurposing of bitcoin mining infrastructure to serve AI workloads, and how shifts in global AI policy are influencing innovation and market structure.
Bottom Line
- Global power demand from AI data centers is projected to reach 156 GW by 2030, more than 2.5x today’s levels, equivalent to powering 120 million homes or exceeding the entire electricity demand of the United Kingdom.
- Bitcoin mining firms are beginning to enter the AI infrastructure market, with companies like TeraWulf, Core Scientific, and Galaxy Digital signing deals to supply a total of over 1,100 MW of power to hyperscalers.
- As regulatory dynamics influence access to AI hardware, countries are adapting by building localized infrastructure and optimizing model development strategies, leading to a more diversified global AI ecosystem.
Energy: The Bottleneck of Intelligence
The transition from the cloud computing paradigm to a high-performance computing (HPC) paradigm, brought on by the AI era, has dramatically increased demand for energy to power AI workloads, both for model training and inference. According to NVIDIA, inference workloads alone have grown 100x in just the past few years, and show no signs of slowing down as reasoning models and agentic AI enter the market. To fill this unprecedented demand, AI research labs, infrastructure providers, and capital allocators have hyper-accelerated their investment into building out the physical sites where AI workloads are powered and executed – AI data centers.
In the cloud computing paradigm, before the rise of generative AI, data centers were primarily designed to support stable, low-intensity computing workloads such as file storage, messaging, and streaming. These tasks required relatively modest levels of compute and power, leading to a steady and predictable increase in energy demand. In contrast, the demands of generative AI, and increasingly agentic AI, has created a step-change in energy consumption, with modern AI data centers requiring exponentially more power to support massive GPU clusters and continuous inference. Global data center power demand from AI workloads alone is expected to rise from 44 GW in 2025 to over 150 GW by 2030, more than tripling from 2025 levels. This surge in demand is being matched by record levels of capital investment, with the global data center construction market projected to be over $450 billion by 2030.
Exhibit 1: Projected Global Data Center Capacity Demand (MW) vs. Projected Datacenter Construction Market (2025 – 2030)
Source: McKinsey, Grand View Research
Yet, while the demand for power from AI workloads and the investment into data centers continue to accelerate, the energy systems required to support this growth are becoming increasingly constrained. Hyperscalers now face multi-year delays in securing access to electricity, held back by long utility queues and infrastructure bottlenecks. In response, AI infrastructure providers have been pursuing alternative strategies, including building their own data centers, sourcing power from non-traditional providers (e.g., nuclear), and partnering with bitcoin miners who already control large blocks of pre-allocated energy and land configured for computing.
Bitcoin Mining: From Bitcoin Sats to Cluster Racks
As the physical buildout for AI runs into power limitations, a surprising sector has emerged as a potential solution: bitcoin miners. With access to large blocks of pre-contracted energy, land configured for computing, and experience operating digital infrastructure sites, many mining firms are now looking to diversify their operations and enter the AI infrastructure markets.
Three mining firms, TeraWulf, Core Scientific, and Galaxy Digital, have signed deals to supply AI compute providers with power and infrastructure. Galaxy Digital, which acquired a mining site from Argo Blockchain in 2022, signed a 15-year agreement with CoreWeave, allocating 600 MW of power at its Texas site. TeraWulf leased 70 MW of its New York capacity to Core42, the U.S. arm of UAE-based G42. And Core Scientific, now owned by CoreWeave, is delivering 500 MW across multiple facilities to support CoreWeave’s growing HPC footprint. These contracts represent significant deployments of power into AI infrastructure, especially compared to the average U.S. data center, which operates at just 40 MW.
Exhibit 2: AI Power Deals Signed by BTC Miners vs. U.S. Average Data Center Capacity
Source: Bloomberg, BCG, The Block
Overall, as the economics of bitcoin mining becomes less profitable due to halving cycles and rising operational costs, leasing infrastructure to AI providers can offer miners a relatively stable and diversified revenue stream. However, not all bitcoin mining firms can easily enter the AI market. Many are constrained by capital structure and infrastructure readiness. AI workloads require dense power, advanced cooling systems, and high-throughput networking, upgrades that are both expensive and technically complex. In short, while the overlap between bitcoin mining and AI infrastructure appears to be a simple arbitrage, only a small subset of Bitcoin mining businesses are well-positioned to capitalize on it.
Compute: Chips, Regulation, and Innovation
AI regulation is also a driving force shaping AI infrastructure development. U.S. export controls on high-performance chips, including NVIDIA’s H20, have significantly limited large-scale model training capabilities in China. In response, Chinese AI firms have pivoted toward building domestic compute infrastructure and developing more efficient model architectures tailored to available hardware. Models like DeepSeek exemplify this shift, designed to deliver strong performance without relying on restricted chips.
This type of regulatory environment is creating ripple effects across the global AI market. On one hand, it restricts access to cutting-edge hardware for some players. On the other, it accelerates local innovation by incentivizing jurisdictions to optimize model architectures, improve training efficiency, and invest in sovereign infrastructure. Other nations, such as the UAE, began investing in custom supercomputers built with non-U.S. chips as early as 2023, while continuing negotiations with the U.S. to secure access to advanced semiconductors.
Overall, global AI infrastructure is diverging into a more localized and diversified landscape. Rather than a single global technology stack, multiple development tracks are emerging, each shaped by local policy, capital markets, and resource availability.
