Bitcoin Miners' AI Arbitrage Play to Boost Revenue

This month, we update the fundamental thesis for publicly traded bitcoin miners, highlighting their increasingly important roles in AI/HPC (high-performance computing) and energy markets. Publicly traded miners now control a record percentage of Bitcoin's global hash rate, and their collective market cap reached an all-time high in July. The MarketVector Global Digital Asset Equity Index*, tracking these stocks, has risen 2.8% as of 12 August, underperforming the bitcoin price by 3,000 bps. At these levels, we believe there could be a major arbitrage between bitcoin miners and AI/HPC data centers. Keep in mind this is a risky market: since inception on February 26th, 2021, the MarketVector Global Digital Asset Equity Index returned -65.7% as of 12 August.

• Arbitrage Between Miners and AI Data Centers
• Bitcoin Miners Revenue Potential of AI/HPC
• Bitcoin Miners Value Boost at Current Valuations

Arbitrage Between Miners and AI Data Centers

We speak to many investors who are still unaware that bitcoin miners now have significant exposure to AI. The synergy is simple: AI companies need energy, and bitcoin miners have it. As the market values the growing AI/HPC data center market, access to power—especially in the near term—is commanding a premium. Due to heightened energy demands from bitcoin miners and data centers, North American grid interconnection queues now exceed four years. However, by drawing large loads and collaborating with power grids by participating in grid balancing programs, existing bitcoin miners can be uniquely equipped to support AI/HPC immediately. Suitable bitcoin mining sites could energize GPUs for AI in less than a year, compared to the 4+ years required for greenfield AI data center developments to go online. Highlighting the potential value of this time arbitrage, we found that bitcoin miners trade at an average of ~$4.5M per megawatt (MW) of installed capacity, compared to above $30M / MW for some data center stocks. If properly equipped with power, bandwidth, and cooling systems, bitcoin mining sites could be ideal for capturing this value for AI/HPC cloud services.

Early adopters like Core Scientific (CORZ) are reaping the rewards. On June 3rd, CORZ, the 4th largest bitcoin miner by hash rate, secured a 12-year contract with AI hyperscaler CoreWeave, projected to generate over $3.5 billion in revenue for providing 200 MW of infrastructure. CoreWeave exercised its first option for additional capacity three weeks later, adding $1.225 billion for an extra 70 MW over 12 years. Then, in early August, CoreWeave exercised yet another option for Core Scientific to deliver 112 more MW of HPC to host NVIDIA GPUs. Core Scientific could soon become one of the U.S.’s largest data center operators if CoreWeave exercises up to 118 MW of further expansions. Following the first announcement, CORZ has added $1.6B to its market cap, with the shares +99% vs. the MarketVector Digital Asset Equity Index (+0.29%), which is by far the best-performing constituent.

Bitcoin Miners Revenue Potential of AI/HPC

Accordingly, we believe investors might want to understand the potential size of the AI/HPC opportunity for bitcoin miners. This analysis examined the capital requirements and potential revenue if publicly traded bitcoin miners pivoted varying percentages of their energy capacity to serve AI & HPC customers. Based on an interview with Hut 8’s CEO at Bitcoin Nashville and third-party estimates, we assume revenues of $1.30 per kWh. By multiplying the $1.30 per kWh rate by 1,000 (to convert to MWh), then by 24 hours a day, and 365 days a year, we calculate the annual revenue per MW. After applying an 80% utilization rate—a rate that Hive Digital surpassed as demand for its AI/HPC servers grew in 2023—we project an annualized revenue of approximately $9.11 million per MW. The capital required to outfit these AI-oriented data centers is enormous. Borrowing from Core Scientific’s capital expense estimates of $5 million to $8 million per MW for converting Bitcoin mining infrastructure to AI/HPC, we conservatively apply a $7.5 million per MW cost of conversion in our model, arriving at a total of ~$23.1 billion for the infrastructure alone. We also assumed that the collective pivot would require 1,681,600 Nvidia H100 GPUs, each costing $32,500 or a total of ~$54.7 billion—for the first generation alone! As we further assumed that GPUs would need to be sold and replaced after four years, with depreciation derived by assuming GPU performance (FLOP/s) doubling every 2.30 years, our model includes two additional sets of replacement GPUs over the 13-year project.

Bitcoin miners generally have bad balance sheets, either because of too much debt, too much share issuance, too much executive compensation, or some combination of all three. This cycle is somewhat better, as many miners restructured during the bear market. What’s great about AI/HPC is that the customers are willing to fund the capex, potentially providing a lower cost of capital for the bitcoin miners to strike their next energy deal. For example, per the terms of CORZ’s first agreement with CoreWeave, CoreWeave will provide ~80% of the capex to retrofit Core Scientific’s existing infrastructure for AI/HPC and additionally pre-finance Core Scientific’s ~20% share of the capex through deferred hosting payment revenue.

Revenue Impact of Bitcoin Miners Shifting Capacity to AI

Estimated Current & Target Power Capacities of MVDAPP Index Bitcoin Miner Constituents

Source: Company filings, Bloomberg, VanEck Research of 12/08/2024. Past performance is no guarantee of future results. Not intended as a recommendation to buy or sell any securities named herein. The information and valuation scenarios in this blog are not intended as financial advice or any call to action, a recommendation to buy or sell, or as a projection of how bitcoin will perform in the future. Actual future performance of bitcoin is unknown, and may differ significantly from the hypothetical results depicted here. There may be risks or other factors not accounted for in the scenarios presented that may impede the performance. These are solely the results of a simulation based on our research, and are for illustrative purposes only. Please conduct your own research and draw your own conclusions.

Our discounted cash flow (DCF) model applies a 17% discount rate over a 13-year project duration to evaluate the profit potential for miners if they convert energy currently dedicated to bitcoin mining to AI/HPC. We estimated each miner's current and target power capacity based on publicly available investor documentation. Then, we determined the present value of each miner’s opportunity to allocate 20% of their current and target power capacity to AI/HPC. While still illustrative, please note that each of these estimates may exceed or fall short of each miner’s actual power capacity expansion, depending on what they have publicly disclosed as of our research and how they execute in the years ahead. We also assume no changes in miners’ power contracts with grid operators, typically lasting only a few years, so that miners can adapt to dynamic energy markets. While large-scale industrial customers like bitcoin miners generally have strong bargaining power to achieve stable or even reduced energy costs over time, it is important to acknowledge the risk that this cost variable could increase over our model’s lengthy 13-year duration—mainly if anti-growth regulations, such as net zero policies, are implemented. In terms of buildout, we assumed a 50% per year project completion rate, meaning that the megawatts allocated to AI/HPC revenues in our projection increase linearly from 0% to 50% at the start of year 2 and plateau at 100% at the beginning of year three. While some locations, such as in Core Scientific's first CoreWeave announcement, can achieve operational status in one year or less, we estimated two years to account for additional work required at mining sites that are less suitable or still under construction. We assumed a power usage effectiveness (PUE) ratio of 1.4, meaning that due to cooling, ventilation, and other overhead infrastructure, 1.4 MW of power would need to be drawn for each 1.0 MW of active AI/HPC.

Bitcoin Miners Potential Value Boost from AI Conversion at Current Valuations

Financial Overview, AI/HPC Strategies, & Estimated Upside Potential Among MVDAPP Index Publicly Traded Bitcoin Miners (2024)

Source: Company filings, Bloomberg, VanEck Research of 12/08/2024. Past performance is no guarantee of future results. Not intended as a recommendation to buy or sell any securities named herein. The information and valuation scenarios in this blog are not intended as financial advice or any call to action, a recommendation to buy or sell, or as a projection of how bitcoin will perform in the future. Actual future performance of bitcoin is unknown, and may differ significantly from the hypothetical results depicted here. There may be risks or other factors not accounted for in the scenarios presented that may impede the performance. These are solely the results of a simulation based on our research, and are for illustrative purposes only. Please conduct your own research and draw your own conclusions.

If AI/HPC comprises 20% of these miners’ energy capacity by 2027, and assuming they could fund the investment, we think the publicly traded bitcoin miners’ total additional yearly profits could exceed an average of $13.9 billion per year over 13 years, compared to the trailing 12 months net income for the group at -$335 million. After factoring in the 17% discount rate, in the aggregate, we estimate the net present value of this opportunity to be ~$37.6 billion, compared to the current total market cap of the 12 companies under consideration, which sits at ~$19.7 billion as of August 12th, 2024. Please note that this exercise assumes miners buy the GPUs themselves when most would likely pursue a hosting model like Core Scientific. While the hosting model will lower the margin potential, it will also slash the required investment. Either way, we think that for many miners, the NPV of converting 20% of MW capacity can easily double the value of the stock. Preliminary findings suggest that miners already exploring or working with HPC, such as CORZ and WULF, are year-to-date outperforming those with no AI/HPC plans such as MARA and RIOT. In fact, likely driven by their ambitions in constructing AI/HPC revenue streams, CORZ and WULF are the only miners who have outperformed bitcoin YTD.

One major obstacle is that only a small percentage of existing bitcoin mining centers have the necessary proximity to major cities, bandwidth, redundancies, and other critical infrastructure needed for AI/HPC conversion. In contrast, established hyperscalers like Equinix already benefit from specialized infrastructure, economies of scale, and the resultant network effects from years of operation and customer success. Bitcoin miners entering the AI/HPC market will likely face lower margins as they work to develop similar operational expertise, recognition, and trust. Additionally, while crypto's regulatory and security demands might partially translate to AI/HPC, working in both industries introduces further challenges. Still, given the lower cost of capital from potential customers such as hyperscalers and the relatively small amount of aggregate power we assume will be converted (20% of planned capacity, or 3GW, compared to Jefferies estimates that 6GW of electricity is required to power NVIDIA GPUs already ordered for 2025), we believe these AI pivots will not cause over-supply and can be highly accretive for miners at current valuations. Moreover, if bitcoin miners can meet Tier 3 and 4 data center standards (achieving 99.982%+ uptime with multiple fault-proof redundancies) while continuing to leverage their expertise in energy grids and digital asset monetization, they could become uniquely attractive long-term investments, bridging two of this century's fastest-growing industries.

Many miners are leaning toward the complementary strategies presented by Bitcoin & AI/HPC. In its May 2024 update, Iris Energy noted that cloud services help optimize capital costs and diversify revenue streams, smoothing returns through Bitcoin’s cycle. In the same month, Terawulf (WULF) outlined models for cloud services, customer colocations, and built-to-suit data centers to optimize land, power, and capital for long-term value creation. Further underscoring the opportunity, in June, the early-stage technology private equity firm Coatue invested $150M through a convertible note into Hut 8 to build next-generation AI infrastructure. As of August 12th, WULF (+58%) and CORZ (+176%) are the two best-performing bitcoin miners YTD, while IREN (+5%) has also outperformed despite a recent power-trading loss, vs. the MarketVector Digital Asset Equity Index +2.8%.

Investors might not want to overlook another stakeholder taking part in bitcoin miners’ growth trend: electric grid operators. Bitcoin miners are attractive to grids like The Electric Reliability Council of Texas (ERCOT) because they have scalable, high-quality energy. Unlike other large power buyers, bitcoin miners can curtail consumption during peak demand, smoothing and lowering electrical costs across the grid. Exemplifying this practice, Riot (RIOT) earned $71.2M in power credits from ERCOT in 2023, generating over 25% of its revenue. While Bitcoin’s unique role as such a dispatchable energy resource or “economic battery” has been recognized as a driver of “green” energy investment for some time, we believe that miners’ more recent integration of AI/HPC further accentuates this dynamic, adding a new layer of versatility and value.

While the miner AI/HPC trend is nascent, it represents a significant merger of two high-growth tech sectors, creating a fascinating game theory dynamic. As some miners go offline to run GPUs, Bitcoin’s difficulty algorithm will automatically adjust, allowing the remaining miners to gain a slightly larger market share. Consequently, unless the US aggressively builds more electricity-generating capacity, it may lose its share in not one but multiple new digital industries. As the synergies between bitcoin mining, AI/HPC, and electrical grids continue maturing—ideally in an energy-abundant, technologically progressive regulatory environment—we believe the miners in the MarketVector Digital Asset Equity Index, collectively, should be able to easily double their market capitalization by 2028, even assuming no growth in bitcoin profits.

Links to third party websites are provided as a convenience and the inclusion of such links does not imply any endorsement, approval, investigation, verification or monitoring by us of any content or information contained within or accessible from the linked sites. By clicking on the link to a non-VanEck webpage, you acknowledge that you are entering a third-party website subject to its own terms and conditions. VanEck disclaims responsibility for content, legality of access or suitability of the third-party websites.

To receive more Digital Assets insights, sign up to our Newsletter.