The rapid rise of ChatGPT has underscored the explosive growth of artificial intelligence (AI), driving an unprecedented demand for Nvidia GPUs as tech giants scramble to secure the computing power necessary to fuel their AI services. With the compute intensity for AI doubling every six to ten months, the need for robust infrastructure has become more critical than ever. In response, several major public Bitcoin miners announced plans in 2023 to integrate AI and high-performance computing (HPC) services into their operations, a trend that gained momentum in 2024 as miners faced shrinking profit margins following the Bitcoin halving.
In this article, we will explore the key similarities and differences between AI/HPC compute and Bitcoin mining, as well as the unique infrastructure requirements to operate these two types of data centers efficiently.
AI / HPC Hardware and Architecture
The Hype is Real
High Energy Consumers, Different Types of Loads
Energy Density
Flexible Load
Firm Load
Data Centre Standards and Initial Capital Expenditure
Modularity and Portability
Power and Cooling
Pricing ASICs vs GPUs
Other Cost Adding Factors
Content for Premium Members
Dominant Hardware Manufactures
ASIC Miner Manufactures
GPU Manufactures
Why Bitcoin Miners Integrate AI/HPC Services
Difference in Access to Funding
Are you looking to mine in the hashrate capital of the world, Texas? Or would you rather explore the emerging mining hub, the UAE? Digital Mining Solutions has you covered! Whether you’re aiming air-cooled or hydro we provide the expertise. Let’s turn your mining ambitions into reality! Get in touch for more details!
AI/HPC Hardware and Architecture
Artificial Intelligence (AI) aims to develop systems that replicate human intelligence processes. This includes tasks such as machine learning, deep learning, and natural language processing. Data centers support these efforts through high-performance computing (HPC) clusters, which consist of multiple servers linked via high-speed networks for parallel processing. HPC commonly utilizes specialized processors, such as GPUs (Graphics Processing Units) and FPGAs (Field-Programmable Gate Arrays). Additionally, AI-specific hardware like TPUs (Tensor Processing Units) and other specialized AI chips are increasingly in use.
The Hype is Real
Over the last year many well-known publicly traded Bitcoin mining companies announced they are integrating AI/HPC into their business. Post halving there is a growing trend of Bitcoin miners diversifying into other computational services as it provides an opportunity to add a stable and lucrative revenue stream to the business.
The most significant announcement has been that of Core Scientific. Core Weave contracted 398 MW for HPC services. This represents 48% of Core Scientific’s current operational capacity, which stands at 830 MW. Core Scientific plans to retrofit a part of its existing Bitcoin mining data centers for their HPC services. Other wel-known public miners adopting an AI/HPC strategy are IREN, Hive Digital, HUT 8, Bit Digital and Terawulf.
Should all miners consider allocating a portion of their power capacity to AI/HPC services? Operating a Bitcoin mining facility and an AI/HPC data centre share some similarities but also have significant differences due to their distinct purposes, technologies, and operational requirements. Let’s have a look at some key similarities and differences between the two.
High Energy Consumers, Different Types of Loads
Energy Density
Both Bitcoin mining and AI/HPC data centres require significant amounts of electricity to power their computational hardware. They rely on powerful ASIC miners and GPUs that consume a considerable amount of energy during operation. The average size of a high-performance computing (HPC) data center typically ranges from 10 to 100 MW, depending on the scale and purpose of the facility. Bitcoin mining facilities tend to vary more in size, with the average range generally between 5 to 50 MW, although larger operations can exceed 100 MW. Due to their specialized design, ASICs provide higher energy efficiency and greater hashing power per square meter of space compared to GPUs.
Flexible Load
Bitcoin mining has the unique feature that it is a flexible and interruptible load. The industry standard for uptime is 95% but there are miners who have higher downtime than 5%. Bitcoin mining operators can participating in demand response or curtailment programmes to balance the power grid. Additionally, they can use unmarketable energy like flare gas and can be co-located behind the fence at power plants.
Firm Load
AI/HPC systems typically aim for uptime percentages well above 99%. In some critical applications, such as those in healthcare, finance, or defence, uptime requirements may be even higher, approaching or exceeding 99.9% (often referred to as “three nines” or “four nines” uptime). For this reason, AI/HPC need high quality and reliable power which means they often compete for the same electricity that residents and businesses need, driving up demand. Additionally, many HPC operators require backup power generation. This redundancy is needed to keep running when the main power source is down.
Data Centre Standards and Initial Capital Expenditure
Building and maintaining AI/HPC data centers presents infrastructure challenges that are far more demanding and costly compared to Bitcoin mining facilities. The costs for an AI/HPC data center infrastructure is estimated to be 10 to 20 times higher than for Bitcoin mining, as they require advanced engineering, redundancy and tailored designs. Hive Digital Technologies estimated the retrofitting of an existing facility to support AI/HPC workloads can cost between $5 and $7 million per megawatt (MW).
Modularity and Portability
Bitcoin mining facilities are typically more modular and portable, making them easier to scale and relocate when needed. Data centers, on the other hand, are far more rigid in design and require larger, more permanent installations to be financially viable. Data centers for AI and HPC must operate at a significant scale to justify the high initial investment in infrastructure, whereas Bitcoin mining facilities can operate profitably at much smaller scales without the need for such extensive infrastructure. Additionally, GPU rigs tend to be bulkier and require more physical space per unit of hashing power and energy.
Power and Cooling
AI/HPC data centers need reliable, high-quality power, often with backup systems like generators or battery arrays to ensure uninterrupted operation. They also require power contracts with no demand response capabilities which can be more costly to obtain. The power requirements are also much higher due to the cooling needs of GPU rigs, which generate significant heat and demand expensive cooling solutions to maintain optimal performance. Bitcoin mining operations can tolerate more fluctuations in power quality and do not require the same level of cooling infrastructure.
Pricing ASICs vs GPUs
Not only the higher quality data centres require relatively more investment but also the GPU’s are currently expensive in comparison to ASIC mining equipment. Post-halving ASICs have been bottoming. Prices are at historic low levels due to depressed demand and a big surplus of hardware on the secondary market.
The demand for AI/HPC hardware has been driven by various industries adopting AI and machine learning technologies. In the beginning of 2024 big tech companies like Meta and Google bought up a lot of GPU’s trying to secure computing power for their AI services. The most popular Nvidia H100 AI GPU has peaked beyond $40,000.
Other Cost Adding Factors
AI/HPC data centers depend on high-bandwidth, low-latency, and highly reliable network connections. This requires multiple fiber sources to ensure redundancy and speed, which is crucial for handling large-scale data processing. Bitcoin mining facilities, in contrast, do not require such high-speed internet as their operations are less dependent on real-time data transfers.
Operating an AI/HPC data center requires a team with specialized training and experience in managing complex environments. In contrast, Bitcoin mining facilities can be managed with a smaller, less specialized team, as the tasks are more focused on machine maintenance and power management.
Unlock the exclusive insights our Premium Members enjoy. Upgrade today for full access!
Dominant Hardware Manufactures
ASIC Miner Manufactures
GPU Manufactures
Why Bitcoin Miners Integrate AI/HPC Services
Difference in Access to Funding
If you’re finding this article helpful, don’t miss out on our comprehensive guide, ‘Mastering Bitcoin Mining.’ This in-depth 50-page handbook covers everything you need to know to launch and scale your mining operation. We explore all cooling systems, essential components of mining infrastructure, containerized solutions, ASIC machines, firmware, monitoring software, and much more.
