Balancing the Scales: Bitcoin Mining, High-Performance Computing, and Energy Consumption

On December 12, 2019, a press release from Riot Platforms (Nasdaq: RIOT) boasted about purchasing 4,000 Bitmain S17 Pro Antminers at its 12-megawatt facility in Oklahoma City.¹ Fast forward to Q3 2024, RIOT issued a press release detailing the progress of Phase 1 of their Corsicana Facility. This phase is projected to reach 400 megawatts upon completion, with the entire mining capacity at the site expected to reach 1 gigawatt by the second half of 2025.² To put that in perspective, the U.S. Energy Information Administration (EIA) states the average American home used ~29.56-kilowatt hours per day in 2022, meaning 1 gigawatt would have powered roughly 811,907 homes on average in 2022.³ In addition, the hash rate capacity of miners has increased by over 300% during this period when comparing a 56 TH/s Bitmain S17 Pro miner from 2019 to a present-day Bitmain S21 Pro 234 TH/s miner in 2024.⁴ This massive increase in RIOT’s energy capacity and miner hash rate perfectly illustrates the explosion of growth into the Bitcoin mining space since 2019.

Present-day large-scale energy consumption is synonymous with technological innovation. Whether the objective is validating transactions to secure blockchain networks or running computationally intensive mathematical algorithms in a large language model (LLM), high-powered data centers across all technology sectors share one common goal: redundant low-cost power at a large scale. Given the acceleration of AI infrastructure in all aspects of technology, established large-scale Bitcoin miners like RIOT now have the option to diversify their operations with Bitcoin mining and High-Performance Computing (HPC) for AI.

Though they are often intertwined in today’s media publications, there are significant differences in infrastructure, operations, and skill sets required for an HPC data center compared to one built to support Bitcoin mining. According to Phil Harvey, CEO of blockchain data center consulting firm Sabre56, a typical Bitcoin mining commercial operation costs between $300,000 to $350,000 per megawatt to run, while HPC data centers cost between $3 million and $5 million per megawatt.⁵ Furthermore, traditional-style data centers for HPC are classified into Tiers based on the specific infrastructure, operational sustainability, and reliability, with Tier I being the simplest and Tier IV being the most complex and resilient.⁶ The classifications involve varying degrees of paths for power, cooling systems, redundancy, and expected uptime. For instance, a Tier I data center typically features a single power and cooling path, with few, if any, redundant and backup components and an expected uptime of 99.671% (equivalent to 28.8 hours of downtime annually). In comparison, a Tier IV data center is built to be completely fault tolerant and has redundancy for every component with an expected uptime of 99.995% (26.3 minutes of downtime annually).⁶

Bitcoin miners often engage in load management demand-response programs. These arrangements with grid providers involve mandatory curtailments during sudden spikes in energy demand from consumers and residents, usually related to extreme weather conditions, to support and stabilize the grid in return for power credits. The Bitcoin Policy Institute (BPI) states that Bitcoin Miners curtailed power between 5%-31% of the time in a study conducted over 10 Bitcoin Mining companies in the US and Canada.⁷ This flexibility in uptime allows Bitcoin mining operations to shut machines off nearly instantly. It ensures grid support and benefits miners simultaneously by allowing them to give power back to the grid in exchange for power credits at peak demand prices. In contrast, HPC data centers operate more like traditional Tier III and Tier IV data centers, requiring full uptime and redundant energy supply. This critical distinction in energy flexibility will have an enormous impact on the future supply and prices of available electricity and the allocation of capacity to both Bitcoin mining and HPC operations.

Per the BPI, current HPC and Bitcoin mining electricity consumption ranges from 125 Terawatt hours (TWh) to 48 TWh, respectively.⁷ These numbers are expected to grow exponentially with the emergence of AI models and HPC data centers. A Goldman Sachs report projects AI to drive a 160% increase in data center power demand, given that an average ChatGPT query needs ten times as much electricity to process as a Google search.⁸ The research study also indicates that data centers will consume 3-4% of global electricity consumption, double the current state of data center consumption. This reality has caused Bitcoin miners to shift their focus on operations efficiency and energy harvesting, as the profitability of Bitcoin mining operations is strongly correlated with power rates. At the same time, HPC is less sensitive to the cost of energy.⁷

Customized hardware and software solutions enable Bitcoin miners to significantly improve mining efficiency, reducing operational costs and increasing profitability. Furthermore, miners are increasingly monetizing excess energy from wind, solar, and hydroelectric producers and converting landfill and flare gas to power in smaller, more modular deployments, maximizing energy efficiency and lowering overall costs. This is yet another illustration of the flexibility of Bitcoin mining from an energy perspective, benefitting both miners and power producers.

Given the rising demand from Bitcoin mining and high-performance computing (HPC), the future prediction of energy prices indicates a potential increase due to the substantial energy consumption these activities require. As both industries expand, the strain on power grids may drive up electricity costs, especially in regions where supply struggles to keep pace with demand. However, this scenario could also accelerate investments in renewable energy and energy efficiency technologies to mitigate rising prices and stabilize the grid. According to a report by BloombergNEF, clean energy investments rose 17% to $1.8 trillion in 2023.⁹

Tech giants have also ramped up considerable interest in nuclear energy. Google has invested in Kairos Power leveraging nuclear technology, and Amazon recently purchased a nuclear-powered data center from Talen Energy.¹⁰ Additionally, in September 2024, the infamous “Three Mile Island” nuclear power plant near Middletown, Pennsylvania, announced its reopening to power Microsoft data centers.¹¹ Molten salt as a coolant and nuclear fuel offers multiple safety and efficiency benefits by naturally changing its power level to match heat removal for electricity production, providing flexibility and eliminating the need for refueling outages.¹² These technological advancements in energy production and storage, along with market dynamics and regulatory policies, will play a crucial role in shaping the future landscape of energy prices and energy supply for data centers.

Sources

1. “Riot Blockchain Announces Purchase of Additional 1000 Next Generation Bitmain S17-Pro Antminers, Completing Upgrade of Its Oklahoma City Mining Facility.” *Riot Platforms*, www.riotplatforms.com/riot-blockchain-announces-purchase-of-additional-1000-next-generation-bitmain-s17-pro-antminers-completing-upgrade-of-its-oklahoma-city-mining-facility/. Accessed 20 Aug. 2024.
2. Riot Platforms. “Riot Announces July 2024 Production and Operations Updates.” Riot Platforms, https://www.riotplatforms.com/riot-announces-july-2024-production-and-operations-updates/. Accessed 20 Aug. 2024.
3. U.S. Energy Information Administration. “How Much Electricity Is Used for Lighting in the United States?” *EIA*, www.eia.gov/tools/faqs/faq.php?id=97&t=3. Accessed 20 Aug. 2024.
4. “Antminer S17 Pro Specifications.” *Bitmain Support*, www.support.bitmain.com/hc/en-us/articles/360021167174-S17-Pro-Specifications. Accessed 20 Aug. 2024.
5. Nelson, Tom. “Converting Mining Sites to AI Data Centers Is Not Seamless: Sabre56 CEO.” *Cointelegraph*, 4 Oct. 2023, www.cointelegraph.com/news/converting-mining-sites-ai-data-centers-not-seamless-sabre56-ceo. Accessed 20 Aug. 2024.
6. “HPE Data Center Tiers.” *Hewlett Packard Enterprise*, www.hpe.com/us/en/what-is/data-center-tiers.html. Accessed 5 Oct. 2023.
7. BPI. BPI 2024 Margot Policy Report. https://cdn.prod.website-files.com/627aa615676bdd1d47ec97d4/66a02960ca5d7628f2080909_BPI%202024%20Margot%20Policy%20Report.pdf. Accessed 20 Aug. 2024.
8. Goldman Sachs. “AI Poised to Drive 160% Increase in Power Demand.” Goldman Sachs, https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand. Accessed 20 Aug. 2024.
9. BloombergNEF. “Energy Transition Investment.” BloombergNEF, https://about.bnef.com/energy-transition-investment/. Accessed 20 Aug. 2024.
10. “Powering the Future of AI: Addressing the Looming Energy Challenge.” Forbes, 16 July 2024, www.forbes.com/councils/forbesbusinesscouncil/2024/07/16/powering-the-future-of-ai-addressing-the-looming-energy-challenge/#:~:text=The%20power%20requirements%20for%20AI,17%20GW%20consumed%20in%202022. Accessed 20 Aug. 2024.
11. “Three Mile Island Nuclear Power Plant Microsoft AI.” *NPR*, 20 Sept. 2024, www.npr.org/2024/09/20/nx-s1-5120581/three-mile-island-nuclear-power-plant-microsoft-ai. Accessed 20 Aug. 2024.
12. “How Molten Salt Could Be the Lifeblood of Tomorrow’s Nuclear Energy.” *Idaho National Laboratory*, www.inl.gov/molten-salt-reactors/how-molten-salt-could-be-the-lifeblood-of-tomorrows-nuclear-energy/. Accessed 20 Aug. 2024.