High Performance Computing (HPC) infrastructure for artificial intelligence (AI) offers a growing opportunity for Bitcoin miners. However, the sustainability of this business model and the capacity available to miners remain in question, according to a report by Needham & Company.
Converting existing mining sites to HPC infrastructure involves large capital expenditure. While some components can be reused, most of the existing infrastructure would need to be rebuilt to support HPC operations.
Currently, HPC data centers require an investment of $8-10 million per megawatt (MW) in capex, excluding GPUs, whereas a Bitcoin mining site typically costs between $300,000 and $800,000 per MW, excluding ASICs.
Most Bitcoin miners do not possess GPUs, which were commonly used for Ethereum mining, and instead employ ASICs for Bitcoin mining. Consequently, they lack the excess GPU capacity necessary for HPC tasks.
Bitcoin miners venturing into AI are predominantly acquiring GPUs, typically H100s, and co-locating them in third-party sites as a quicker and less capex-intensive way to generate AI revenue compared to building dedicated data centers.
However, Needham believes it is more advantageous for miners to own their data centers, especially if they have access to cheap power, rather than relying on third-party sites for GPU co-location.
Large publicly traded Bitcoin miners plan to more than double their power capacity over the next 12-24 months, inclusive of their mining and AI/HPC expansion plans. Location is a critical factor, with Needham favoring northern, less dusty climates over regions like West Texas due to temperature and dust/wind conditions.
According to the report, Applied Digital Corp (NASDAQ:APLD), Core Scientific Inc (NASDAQ:CORZ), and Terawulf Inc (NASDAQ:WULF) are leading in their ambitions for HPC sites. Companies like Hut 8 Corp (NASDAQ:HUT), Bit Digital Inc (NASDAQ:BTBT), HIVE Blockchain Technologies Ltd (NASDAQ:HIVE), Iris Energy Ltd (NASDAQ:IREN), and Bit Brother Ltd (OTC:BETSF) are currently focusing on co-locating GPUs in third-party sites.
The report further details that Applied Digital has the most ambitious AI/HPC site plans among covered companies. Originally a Bitcoin miner, the company no longer classifies itself as such after selling a ~200 MW Bitcoin mining hosting site to Marathon Digital (NASDAQ:MARA).
Meanwhile, Bit Digital has become the first cloud service platform in Asia to offer NVIDIA DGX SuperPod H100s. "We do not include any revenue from AI/HPC in our model but will be monitoring for updates as 2024 progresses," Needham noted.
HUT 8 Mining also launched a third-party HPC service, despite currently having no revenue from it. The company executed a purchase order for 1,000 Nvidia (NASDAQ:NVDA) H100s and secured a $20 million annual run-rate contract with a VC-backed AI large language model customer, with revenues expected to begin in the second half of 2024.
Elsewhere, Bit Brother has a three-year contract to provide GPU resources to a customer developing its own proprietary large-language model (LLM) and expects $50 million in annual revenue following a 2023 funding round. Moreover, Core Scientific’s multi-year contract with CoreWeave is expected to generate more than $100 million in total potential revenue.
Finally, HIVE Blockchain had majority of H100s delivered in March 2024, and the company reports growing positive HPC operating income with $25 million in run-rate annual HPC revenue guidance as of Q1 2024.