The surge in demand for advanced chips used in artificial intelligence (AI) has presented a new opportunity for cryptocurrency miners who survived the previous tech boom and subsequent decline. Companies like Satoshi Spain, which previously sold and leased mining rigs for cryptocurrencies, have now repurposed their machines for AI tasks. These refurbished rigs, equipped with graphics processing units (GPUs), are being utilized by startups, universities, and individual developers across Europe to handle AI workloads.
The rising popularity of AI applications, such as ChatGPT, has led to a shortage of GPUs, prompting a competition to secure computing power and find alternatives. In the past, miners rushed to acquire GPUs to mine cryptocurrencies during the price surge. However, with the decline in crypto prices, many machines were left idle, creating an unused resource known as “dark GPUs.” Now, startups and miners are reprogramming or reselling these GPUs to cater to the computational demands of AI.
While converting mining rigs for AI purposes may not always be straightforward or inexpensive, they offer a more affordable and accessible option compared to the infrastructure provided by major cloud companies. Startups and universities, facing challenges in obtaining AI computing power from larger providers, can benefit from these repurposed rigs. However, meeting the requirements of training AI models still demands a significant amount of identical chips and memory.
The change in how one cryptocurrency, Ethereum, is minted has resulted in a surplus of used chips. The elimination of mining for new coins and the associated intensive computation led to tens of millions of GPUs becoming available. About 20% of these chips can be repurposed for training AI models, leading to an increasing number of crypto miners exploring participation in the AI field.
Companies like Together are using former crypto miners and their hardware to build virtual server farms, offering cloud services for AI developers. Demi Guo, founder of Mellis AI, has turned to repurposed hardware provided by Together’s cloud due to pricing and capacity limitations at leading cloud providers. Similarly, Monster, a startup founded by Saurabh Vij, leases chips previously used for mining to customers in need of AI computing power.
However, it remains uncertain how much of the surging demand for AI computing power can be met by crypto miners. Repurposed mining rigs, typically constructed by individuals, may not possess the necessary graphics cards suitable for intensive AI training. Companies like CoreWeave have refurbished mining rigs for AI tasks by replacing most of the components and investing in high-end Nvidia data center chips. Such radical and expensive upgrades are essential for miners seeking to pivot towards AI.
Many miners underestimate the significant investment required to upgrade their machines to handle AI workloads, including additional components that can cost up to $5,000 for devices with lower-level graphics cards, according to Mark D’Aria, an entrepreneur specializing in purchasing and reselling mining rig components.