Crypto companies that once filled data centers with GPUs for coin mining are now exploring whether to sell or repurpose their idle hardware for the rapidly growing artificial intelligence (AI) market.

Hive Blockchain, which owns 38,000 Nvidia GPUs, is transitioning its equipment to high-performance computing (HPC) and AI applications. The company began rethinking its use of these GPUs, which are still used for Bitcoin and Ethereum mining, after the crypto market took a downturn.

As the AI and cloud computing sectors are booming, crypto companies are realizing that GPUs might be more profitable in these fields. Hive ran a pilot project that dedicated its GPUs to AI and HPC tasks, and the results were promising.

“Though we still have much to do to fully utilize our GPU fleet, we’re pleased that our beta project, with only around 500 GPUs, generated $230,000 in revenue this quarter,” said Frank Holmes, the company’s executive chairman, in a recent filing.

The crypto market has been volatile, particularly since Bitcoin prices dropped in late 2021, with the value of Bitcoin averaging 49% lower compared to last year.

This crash in cryptocurrency prices came alongside software changes that reduced the need for GPUs in mining. For example, Ethereum’s switch from the proof-of-work system to a more energy-efficient proof-of-stake system meant that GPUs were no longer necessary to mine Ethereum.

Despite this, Hive has continued to benefit from the general-purpose nature of GPUs, which remain highly sought after in the AI space. Nvidia’s latest GPUs are in high demand and short supply, and even Hive’s older Nvidia GPUs are being repurposed for AI inference tasks.

The crypto market’s volatility is also affecting Sysorex, a government services company, which is now rethinking how to use its 12,000 GPUs. Sysorex’s subsidiary, TTM, had focused on cryptocurrencies but left the mining business after Ethereum’s software switch. The company is now exploring alternative uses or sales opportunities for its GPUs, but further investment in data centers would be needed for repurposing, making it a challenging prospect.

“TTM Digital is currently exploring alternative uses and sales opportunities for its GPU assets and data center located in Lockport, NY,” Sysorex stated in a filing with the U.S. Securities and Exchange Commission.

For companies like Nvidia and AMD, the rise of crypto miners had created unpredictable spikes and drops in GPU sales, which made forecasting difficult. Miners would often pay a premium and hoard GPUs during chip shortages, leading these chipmakers to adjust their strategies. In response, they took steps to make GPUs less appealing to miners while prioritizing their core audience of gamers. Additionally, Intel discontinued its Bitcoin ASIC, called BlockScale, earlier this year.

The shortage of GPUs and the growing reliance on Nvidia chips for AI tasks is pushing customers to look for alternatives from other chipmakers. While Intel and AMD have introduced GPUs for AI, they are not yet shipping these chips in large volumes.

AI is increasingly dominated by large language models, like OpenAI’s ChatGPT and Microsoft’s BingGPT, which require massive GPU clusters. However, companies like MosaicML and open-source developers are also working on models that can run locally on laptops and desktops for inference computing—using pre-trained models on new data. Intel has also integrated AI accelerators into its new CPUs, enabling inference processing on local devices.