Original Title: Bitcoin Miners Powering the AI Revolution
Original Authors: Simrit Dhinsa, Sebastian Orejas, Gabe Parker
Original Source: https://www.galaxy.com/insights/research/bitcoin-mining-ai-revolution/
Compiled by: Tom, Mars Finance
Unlocking Exceptional Growth Opportunities in the Transition to AI
Key Takeaways
Bitcoin miners that possess large-scale land, water cooling conditions, dark fiber networks, reliable power, skilled labor, power approvals, and critical long-lead-time infrastructure components can significantly enhance their asset value by meeting the rapidly growing demand of the AI/HPC data center market.
Goldman Sachs' research department forecasts that U.S. data center demand will reach 45 GW by 2030, with an annual compound growth rate of 15% for power demand from 2023 to 2030, primarily driven by AI.
J.P. Morgan expects that by 2038, AI capital expenditures from hyperscale companies will reach $370 billion, an increase of 127% compared to AI capital expenditures in 2024.
There is now a significant demand for access capacities between 300 MW to 1000 MW and even higher, exacerbating pressure on local grids to accelerate power delivery, thus extending the interconnection and construction cycles to 2 to 4 years.
Traditional data centers themselves lack massive power capacity and cannot support high-density computing operations. Previously, the peak power consumption per rack was about 40 kW; today it needs to exceed 132 kW per rack to support state-of-the-art systems like GB200 NVL72.
The cash flow predictability, active financing markets, and significant valuation upside potential of AI/HPC business make this opportunity extremely attractive and value-enhancing for miners with suitable assets.
Miners can unlock tremendous value by transitioning to the AI/HPC market, arbitraging the current leading data center operators' valuation multiples of 20-25 times EV/EBITDA with a valuation of 6-12 times EV/EBITDA.
Introduction
The rise of artificial intelligence (AI) is generating unprecedented demand for High-Performance Computing (HPC) facilities. This surge is pushing hyperscale companies to increase their investments in new data center capacity. However, traditional data centers are struggling with limited power capacity and facility construction cycles that last 2-4 years, making it difficult to meet these demands.
Bitcoin miners have a unique advantage in this market opportunity because they have access to large-scale power infrastructure and the critical components needed for data center operations. Although not all mining facilities can transition to AI data centers due to specific requirements such as cooling, networking, and redundancy systems, those miners with the right assets and specialized skills are expected to benefit from the high cash flow margins and tremendous valuation potential of the AI/HPC business. This article will examine the current state of the traditional data center market, analyze the specific obstacles it faces in meeting AI computing demands, and explain why certain types of Bitcoin miners are likely to fill this gap. At the same time, we will explore the future trends at the intersection of Bitcoin mining and AI infrastructure.
What are the Opportunities for AI Data Centers?
In 2024, AI is expected to accelerate rapidly, primarily due to the widespread adoption of generative AI (GenAI) technology. According to Pitchbook data, over $680 billion has been invested in AI and machine learning startups since 2016, involving more than 100,000 investment cases, with 2024 investment reaching as high as $120 billion.
The growth of AI and High-Performance Computing (HPC) is creating significant demand for data center capacity. Data centers provide the necessary infrastructure and power support for AI/HPC operations, especially for GPU-intensive computing. Emerging AI applications, such as large language models (LLM), are particularly demanding in terms of power consumption. According to the International Energy Agency, a single ChatGPT query consumes 2.9 watt-hours, compared to 0.3 watt-hours for a Google search.
Emerging high-energy AI/HPC enterprises in the U.S. are driving increased demand for data centers. Goldman Sachs' research department predicts that U.S. data center demand will reach 21 GW in 2024 (a 31% year-on-year increase), with an estimated compound annual growth rate of about 15.8% from 2022 to 2033. Given the significant year-on-year growth in 2024, Goldman predicts that U.S. data center demand will reach 45 GW by 2030. At that time, U.S. data centers may consume up to 8% of the national power capacity.
This market opportunity will also be driven by the continuous ramp-up of hyperscale companies towards AI infrastructure. These hyperscale companies (like Google Cloud and AWS) are constantly expanding data center capacity to serve enterprise clients. According to estimates from JPMorgan Asset Management, $163 billion will be invested in expanding hyperscale businesses by the end of 2024, a year-on-year increase of 28%. By 2038, this figure is expected to reach $370 billion, an increase of 127% over AI capital expenditures in 2024.
The current and future growth of AI and HPC technologies is reshaping the data center landscape. As processing demands intensify, hyperscale companies and data centers are evolving from traditional computing facilities to advanced AI infrastructure hubs. These facilities are becoming the foundational infrastructure supporting breakthrough technologies such as autonomous vehicles, advanced medical research, and next-generation AI applications. Future digital innovations will largely rely on the continuous evolution and expansion of these critical computing facilities, marking a new era for technology infrastructure.
Current Data Center Market Overview
The current data center market comprises numerous public and private participants managing large data center portfolios. Notable companies include Digital Realty, Equinix, Vantage, EdgeConnex, and QTS. The largest data center region in the U.S. is in Northern Virginia, but due to massive growth across various locations, vacancy rates have dropped to historic lows (data source: CBRE).
Data centers are the backbone of multiple industries, supporting everything from Netflix's streaming services to cloud computing, artificial intelligence, and many other applications. However, not all data centers are the same. They can be classified by functionality, such as hyperscale data centers, edge data centers, cloud data centers, and enterprise data centers. Moreover, data centers are becoming increasingly large and high-power-density. To meet the infrastructure demands of rapidly expanding industries like AI, hyperscale companies are accelerating the expansion of data centers, leading to an infrastructure arms race.
Obstacles for Traditional Data Centers to Meet AI Demand
Traditional data center providers have long served non-AI industries, typically possessing widely distributed, low-power small data center portfolios. Over the past decade, the energy requirements of these data centers have been relatively mild. While Digital Realty (with a market value of $62 billion) and Equinix (with a market value of $94 billion) are among the largest data center companies in the world, they mainly operate smaller data centers. For example, Digital Realty typically operates facilities with power ranging from 0.5 MW to 40 MW, while Equinix's xScale program has a total operational capacity of only 292 MW across 20 facilities worldwide (see Equinix's Q3 2024 Investor Report, dated November 8, 2024). In contrast, some mining sites can obtain a considerable energy capacity at a single location.
In the past, data center operators lacked the incentive for rapid expansion because streaming, telecommunications, data storage, and many cloud applications had lower computing density requirements. However, with the development of artificial intelligence and the increasing complexity of its algorithms, data centers must operate state-of-the-art facilities equipped with the latest generation GPUs and scale up significantly to optimize training execution.
Massive expansion is supported by improvements in GPU computing power and the advantages of parallel computing, enabling data centers to build larger clusters to enhance computing capacity. Parallel computing allows workloads to be distributed across more GPUs, thus scaling efficiently. When latency between GPUs is low, large centralized clusters can significantly improve parallel computing performance. A 200 MW cluster at a single site can significantly enhance the efficiency of AI training compared to a cluster distributed across four 50 MW sites. This low-latency GPU interconnection is key to achieving maximum computing efficiency. Therefore, hyperscale companies are prioritizing sites with large power capacities to meet the demands of advanced AI workloads.
This capacity is currently in short supply, and many traditional facilities struggle to meet the massive energy consumption required by modern AI/HPC workloads. Existing facilities face significant challenges in retrofitting due to the stark differences in computing load requirements (such as network, cooling, and rack density).
Today, hyperscale companies require data centers with higher power capacities to support their highly power-hungry model training, such as large language models. According to a December 2020 article from the Uptime Institute, the average rack density that year was 8.4 kW/rack (not counting high-performance extremes above 30 kW/rack). The previous limit of 40 kW per rack must now support over 132 kW/rack to accommodate state-of-the-art systems like the NVIDIA GB200 NVL72, a more than threefold increase in just a few years. Industry experts predict that increased computing density and the evolution of Moore's Law will drive server rack power demand to unprecedented levels.
Due to this demand, traditional data center operators are shifting focus towards new greenfield developments to accommodate the new generation of data centers specifically designed for AI/HPC. The energy approval and construction cycles for such projects can often take several years. A recent report from the U.S. Department of Energy shows a surge in grid connection applications requiring between 300 MW to over 1000 MW, putting immense strain on local grids and extending the interconnection and construction cycle to 2-4 years (CBRE data).
Hyperscale companies are planning to build the largest GPU clusters to train AI/HPC models, with some companies targeting gigawatt-scale data centers that can accommodate hundreds of thousands of next-generation GPUs. Although hyperscale companies are also building their own data centers, they still heavily rely on third-party providers with readily available power capacity to accelerate the GPU power-up process. However, currently, only a few existing data centers can meet such massive power demands and high rack energy density. This shortage largely stems from the market's underestimation of the exponential growth in demand for data centers.
Why Bitcoin Miners Can Fill Critical Gaps
Bitcoin miners meet the energy demands of hyperscale companies because they possess large-scale, immediately accessible power facilities. For years, miners have sought locations with ample and affordable power, securing substantial power capacity and long delivery timelines for infrastructure (such as substation components and medium to high voltage equipment) at a single location. Some mining sites already meet readiness conditions for power, addressing one of the biggest bottlenecks faced by hyperscale companies: obtaining reliable large-scale power.
By utilizing these already power-ready Bitcoin mining sites, hyperscale companies can bypass the lengthy energy acquisition process and focus on facility modifications to meet their specific needs. Many sites controlled by miners can reach hundreds of megawatts in scale, which is a level rarely achieved by traditional data center operators at a single location. Some large mining operations have secured energy pipelines exceeding 2 GW, giving miners a unique advantage in meeting power capacity demands. Although Bitcoin mining sites and AI data centers differ at key points, miners are experienced in large-scale construction and data center management, possessing established electrical, mechanical, facility, and security teams that help hyperscale companies get up to speed faster during expansions.
Not all miners can benefit from AI.
Not all miners can profit from the AI/HPC opportunity. Building a data center suitable for AI/HPC requires meeting several key conditions, including large land areas, water cooling capabilities, dark fiber, reliable power, and skilled labor. Even with these conditions, if a company does not have readily available power capacity, land, and planning permissions, or the critical long-lead-time infrastructure components, it will still face obstacles and delays in development.
Another reason is that the existing infrastructure of Bitcoin mining sites is not directly applicable to AI data centers, as there are differences in design and operational requirements. Although there are some similarities in power infrastructure (such as high-voltage substation components and distribution systems), AI data centers have more stringent and complex requirements, necessitating more specialized skill teams.
AI data centers have elevated operational requirements across all aspects, including mechanical, cooling, and network systems; transforming Bitcoin mining sites into AI/HPC data centers poses an engineering and design challenge. Below are some major upgrades that miners need to undertake to meet AI data center demands:
1. Network Infrastructure:
AI/HPC workloads require high-speed, low-latency connections between GPUs. The internal network architecture of AI/HPC is much more complex than that of mining environments, as frequent communication between GPUs is necessary. The key is to establish an optimal network backbone to ensure rapid execution of workloads. Additionally, dark fiber must be accessed for site connections to meet latency requirements, which mining sites do not require.
2. Cooling Systems:
Miners can adopt air cooling, water cooling, and immersion cooling methods, focusing on the machines themselves while paying less attention to auxiliary infrastructure. AI data centers, however, require more advanced cooling solutions, such as direct chip liquid cooling, to cool high-power density NVIDIA servers, along with additional air cooling systems to support networking and mechanical infrastructure.
3. Redundancy:
AI data centers have much higher redundancy requirements compared to Bitcoin mining operations. Mining businesses are flexible and do not necessarily require strong backup power. In contrast, AI data centers typically adopt at least N+1 redundancy strategies, enforcing stricter redundancy requirements for critical components such as core networks and storage to ensure continuous operation or at least to have data cached and checkpointed in the event of equipment failure. This means that for every critical infrastructure (such as cooling equipment), there must be a backup unit. When maintaining a particular cooling unit, a backup unit must remain operational continuously. Mining facilities generally do not have such continuous operational requirements.
4. Shape Redesign:
AI data centers utilize rack-mounted servers, which differ significantly from the box-like ASIC devices used in mining. To accommodate AI hardware, the internal physical structure of the facility needs a complete redesign to meet the specific cooling, networking, and power requirements of rack systems.
5. Other Differences:
In summary, transforming mining facilities to meet AI/HPC data center requirements is a daunting design and engineering task. The higher infrastructure requirements also significantly increase the capital expenditure of AI/HPC data centers compared to the construction costs of Bitcoin mining.
Potential Gains for Miners Profiting from AI Data Center Demand
Although miners may have the right infrastructure and location, transitioning to AI/HPC business is not just a matter of physical assets; it also requires specialized skills, different tech stacks, and new business models. Miners with experienced management teams capable of successfully establishing AI/HPC businesses will have the opportunity to bring significant value to their companies. Below are some potential value-added benefits of shifting power and data center resources from Bitcoin mining to AI/HPC:
1. High Cash Flow Margins and Predictability:
AI/HPC data center businesses (especially co-location/custom build models) typically sign long-term fixed and periodic cash flow contracts with creditworthy clients, often established before the construction of the data center begins. These cash flows are predictable and highly profitable, allowing data center operators to pass some costs onto tenants based on the leasing structure, including energy and operating expenses.
2. Cash Flow Diversification:
This type of revenue is not only more predictable than Bitcoin mining but also uncorrelated with the cryptocurrency market, thereby smoothing the income curve for companies with high exposure to the volatile crypto market. In a Bitcoin bear market, this can enhance financial stability and make it easier for miners to raise capital through equity or debt without excessive dilution or bearing high-interest burdens.
3. Strong Capital Markets Supporting Scalable Expansion:
Although AI/HPC infrastructure is costly, its predictable cash flow makes investment evaluation easier, attracting more debt and equity capital. Investors such as private equity, infrastructure funds, pension funds, and life insurance companies are eager to capture the returns from data centers. Data center operators with creditworthy tenant agreements can leverage those leases to raise significant project financing to build data centers.
According to the Newmark 2023 Data Center Market Overview report, regular debt financing reached a historic high in 2023, and the growth trend continues. In the first quarter of 2024 alone, $18 billion in development financing was approved. Interest rates are also relatively reasonable, with spreads added to SOFR rates ranging from approximately 2.25% to 4.50%, depending on the lender.
4. Huge Valuation Upside Potential:
Once assets are built and operating stably, the valuation difference between Bitcoin mining and AI/HPC businesses makes AI/HPC an extremely attractive opportunity. Bitcoin mining companies traditionally fluctuate in the 6-12 times EV/EBITDA range, while the largest global data center operators have valuations of 20-25 times EV/EBITDA. This difference makes sense as the AI/HPC industry has high-profit margins, a defined growth path, predictable cash flow, and lower volatility compared to the crypto market. For example, the enterprise value of hybrid mining/AI companies is only 23% of Digital Realty, but their potential capacity is 3.5 times that of the latter.
Therefore, for miners with the right assets, the cash flow predictability, active financing markets, and significant valuation upside potential that AI/HPC offers make this opportunity extremely attractive and value-enhancing. These miners are expected to make significant progress in the traditional data center market and become one of the largest operators in the industry.
Outlook for Bitcoin Mining
Although AI/HPC has garnered attention in recent months, we still expect Bitcoin network computing power and mining operations to continue to grow. The mining industry has not stagnated amid the rapid development of AI/HPC. The rise in Bitcoin prices has improved miners' profitability; if prices continue to rise and exceed the increase in network difficulty, mining may become even more profitable.
So, against the backdrop of the simultaneous rise of Bitcoin and AI/HPC, what will the future mining landscape look like? Here are the main trends at the intersection of Bitcoin mining and AI/HPC:
1. Maximizing Energy Value for Miners:
The vast majority of Bitcoin miners have prioritized maximizing energy value. Currently, AI data centers represent the most profitable path for those companies capable of adjusting their sites. Considering the value enhancement of AI/HPC data centers, if a particular mining site can be converted into an AI/HPC data center, miners may choose this direction to maximize shareholder value. This does not represent a retreat from Bitcoin mining, as we still expect network computing power to grow, albeit potentially at a slower rate than if all large U.S. miners maintain the mining model. For those miners still in the network, this transition effectively reduces competitive computing power, benefiting their profitability.
2. Bitcoin mining as a driver of developing remote capacity:
As AI/HPC acquire large-scale sites at higher prices in more developed markets, Bitcoin miners will be more inclined to deploy capacity in more remote areas to leverage the local surplus power generation capacity. The permissionless, location-insensitive, and flexible nature of Bitcoin mining makes it one of the best methods for utilizing surplus capacity in remote areas.
We expect more Bitcoin mining to shift to edge areas to monetize the use of residual power in remote areas, especially in the remote regions of the U.S. and emerging markets like Ethiopia and Paraguay, which have abundant and inexpensive excess energy.
3. Bitcoin Mining as a Strategic Bridge for Infrastructure Investment and AI/HPC Flexibility:
As regions across the U.S. ramp up the construction of transmission infrastructure and fiber connections, Bitcoin mining can serve as a transitional means for pre-financing larger energy infrastructure projects (such as substations and power generation facilities), even when there is no clear current use for AI/HPC. By leveraging Bitcoin mining for opportunistic land and power investment, investors can gain returns while awaiting other long-term energy use scenarios to emerge, making it an attractive infrastructure growth and investment strategy.
Miners unable to transition to AI/HPC data centers can still remain profitable in Bitcoin mining for the long term. Some miners have purchased large-load facilities and are investing in sites at different stages, even without AI/HPC tenants. As previously mentioned, these sites may not have the optimal conditions for AI/HPC, but they can still be used for mining. Other miners lack the teams or internal expertise to contract with major buyers and cannot afford the challenges of engineering and large construction projects. These miners hope to attract AI customers, but if AI/HPC opportunities do not materialize, they can still choose to build a profitable Bitcoin mining business.
4. Emerging Synergies between AI/HPC Data Centers and Mining:
ASIC manufacturers (like Bitmain) have begun developing ASICs that resemble GPU racks for data center racks. If ASICs take on the form of next-generation GPUs, data centers could install server-grade mining machines in available rack space, thereby aligning designs with AI/HPC. In the future, miners may be more inclined to purchase these types of machines to maintain flexibility in data center design, facilitating easy transitions when higher-value AI/HPC opportunities arise.
As AI/HPC data center capacity grows, its impact on the power grid will also increase. While AI/HPC data centers need to remain online long-term, this does not mean energy consumption is constant. In fact, the load curve for AI/HPC training can be very unstable; peak computing loads consume more power, while checkpointing phases consume less. As models grow larger and more data is stored, the time required for saving data will also increase.
Similarly, for AI/HPC inference workloads, the load curve will be highly correlated with client demand. The more model queries, the more energy the data center consumes; initially, this demand can be very volatile. However, as specific models become more established, the load may peak during the day and drop overnight. Daytime peak and valley fluctuations provide an ideal opportunity for Bitcoin mining, allowing the mining business to flexibly adjust computing loads, increasing mining load when AI inference business loads decline.
Therefore, in the future, Bitcoin mining can serve as a load-balancing mechanism, increasing mining energy consumption when AI loads are low and reducing mining when AI loads rebound. This can provide additional value to data center operators and bring load stability to tenants, benefiting the overall stability of the power grid. As data center clusters grow in scale, the impact on the power grid will receive more attention, making load stability crucial.
5. Shifting loads to AI/HPC can slow the growth rate of computing power:
Miners participating in AI/HPC business are diverting capacity that could be used for Bitcoin mining, slowing the growth rate of network computing power. This is particularly important in a potential Bitcoin bull market, as the increase in Bitcoin prices will not be completely offset by the rise in network computing power, thereby driving up hash prices and increasing profitability for all miners. Nonetheless, we still expect network computing power to continue to rise as more efficient mining machines are deployed, whether replacing old machines or being deployed at new sites unsuitable for AI/HPC.
Conclusion
Demand for U.S. data centers is poised to surge at an unprecedented rate, with a projected year-on-year growth rate of 31% in 2024. Over the next five years, U.S. data center capacity is expected to increase from the current 21 GW to more than 45 GW. This explosive growth, coupled with commitments of hundreds of billions of dollars in investments from hyperscale companies over the next 5-10 years, creates a highly attractive opportunity for companies that can provide ample cheap energy and have a solid infrastructure to support AI/HPC operations.
The current prosperity of AI and HPC reveals the fundamental weaknesses of traditional data centers, which cannot retrofit existing facilities to meet the robust power demands of modern AI workloads. This market gap creates significant opportunities for Bitcoin miners. Miners already possess the critical resources needed by AI/HPC enterprises: large-scale sites that can be powered up quickly. Hyperscale companies have limited options, and if they want to expand in time to keep up with the explosive demand growth from AI/HPC, Bitcoin miners become their reasonable and viable partners. However, this historic opportunity is not suitable for all miners; only a few possess the infrastructure and capabilities to meet the stringent requirements of modern AI/HPC workloads. Those with scarce assets willing to maximize their value will transition into AI/HPC data centers.
Despite criticisms that Bitcoin miners transitioning to AI/HPC services could weaken network security (due to a reduction in computing power used for block mining), this transformation may actually benefit the overall development of the mining ecosystem. Miners unable to meet AI/HPC site requirements will benefit from increased hash prices. When some miners go offline and Bitcoin prices rise, the increase in hash prices will significantly improve the profitability of all Bitcoin miners. Since the beginning of 2024, Bitcoin prices have soared by 143%, and with a newly elected pro-Bitcoin president in the White House, U.S. Bitcoin mining is expected to enter its most prosperous era.
In 2024, the intersection of crypto and AI is set to be one of the hottest areas in the crypto industry. By December 2024, the total market capitalization of crypto projects with tradable tokens engaged in AI project development is estimated to be around $33 billion. Furthermore, Galaxy Research estimates that early-stage crypto AI startups have raised over $382 million in venture capital by early 2024. Although most crypto AI projects lack product-market fit, the intersection of Bitcoin mining and AI/HPC business growth is clearly visible. Bitcoin mining entering the AI space is more meaningful than other crypto and AI combinations since it can provide the most critical element for AI/HPC enterprises—energy—on a large scale. Therefore, Bitcoin miners who can convert to AI/HPC assets might be among the few pure and scalable dual investment opportunities available in today's market.
