Crypto miners repurpose GPUs for AI workloads amid market shift

Crypto companies that once bought GPUs to mine cryptocurrency are now redirecting those same cards toward artificial‑intelligence (AI) workloads. Hut 8, Hive Blockchain and IREN have each announced plans to lease or sell GPU capacity for AI training, seeking to offset the loss of mining revenue after the crypto market slumped and Ethereum moved to proof‑of‑stake [1].

Key takeaways

• Hut 8 earned $16.9 million from GPU‑time services in 2023, about 11 % of its revenue [1].
• Hive Blockchain spent $66 million on Nvidia GPUs in 2021 and aims to generate $10 million from AI services in 2024, rising to $20 million in 2025 [1].
• IREN added 4,200 Nvidia Blackwell B200 GPUs, bringing its total to 8,500 GPUs at a Canadian site, and secured $102 million in financing for the purchase [2].
• IREN’s AI cloud expansion targets a 50 MW power capacity capable of supporting roughly 20,000 Blackwell GPUs and an estimated 50 EH/s mining capacity [2].
• All three firms will compete with major cloud providers such as Google Cloud, Microsoft Azure and Amazon AWS for AI‑focused customers [1].

From crypto mining to AI cloud services

Hut 8 and Hive Blockchain, both owners of large GPU racks, have begun offering high‑performance computing (HPC) services to external clients. Hut 8’s GPU‑time revenue of $16.9 million in 2023 represented a modest but growing share of its earnings, while Hive reported a $1 million AI‑related income in 2022 and projects $10 million in 2024, doubling to $20 million the following year [1]. The shift reflects the broader industry trend of repurposing mining hardware for AI, as Nvidia’s AI‑optimized cards such as the H100 “Hopper” remain scarce and expensive, making older gaming‑class GPUs a more viable interim solution [1].

IREN (also known as Iris Energy) has taken a more aggressive approach, purchasing 4,200 Nvidia Blackwell B200 GPUs and expanding its fleet to 8,500 units at its Prince George data center in British Columbia [2]. The company says the new GPUs, together with existing H100, H200, B200 and B300 models, will support both continued Bitcoin mining (estimated at ~50 EH/s) and a growing AI cloud business. Financing for the GPU acquisition is being arranged through a 36‑month lease at a “high single‑digit interest rate,” funded largely by existing cash and a $102 million loan for earlier GPU purchases [2].

Why it matters

The conversion of mining GPUs to AI workloads illustrates how crypto‑related firms are adapting to a market where traditional mining profits have eroded. By leveraging the parallel‑processing strength of GPUs, these companies can tap into the rapidly expanding demand for AI training infrastructure, though they will face stiff competition from established cloud providers that already offer specialized AI services [1]. The success of this pivot will depend on the firms’ ability to manage operational costs, maintain hardware suitable for AI models, and secure financing without diluting shareholder value. As the AI cloud market continues to grow, the repurposing of mining hardware could become a critical bridge for crypto firms seeking sustainable revenue streams.