Let’s begin with some basic facts. Fact number one: The price of bitcoin, despite a recent rally is still well below the bullish peak of 2021 when many bitcoin miner bought their hardware.

Fact number two: The startup industry has seen a shift in capital and hype from Web3 towards AI.

Fact number three: High-end computer processors are required for both cryptocurrency mining and AI development, such as training large language models.

This article is part of CoinDesk 2023 Mining Week sponsored by Foundry.

Fact number four: These chips are in short supply globally.

This all raises a question. Are crypto miners shifting to AI?

While most Bitcoin-only miners do not have this option, larger shops are looking into it. Econoalchemist, an influential bitcoin miner at home, says: “I’ve never heard or thought of that.”

It’s simple. The majority of bitcoin miners are using ASICs, like the Antminer S19 Pro that is designed to perform SHA256 hashing. They’re great at mining bitcoin, but terrible for anything else. They cannot be repurposed. Every bitcoin miner that I spoke to agreed.

The math for large operations is different. The ASICs themselves cannot be converted to AI from bitcoin, but the infrastructure the companies already have in place — security, cooling systems, and access to cheap electricity — can be utilized to expand into AI.

They’re expanding. Applied Digital, a Texas crypto miner , announced a recent a deal worth $460 million to host AI cloud computing at its data center. Wall Street approved, shares immediately rose 17%. Iris Energy is another Texas-based miner that announced a revitalization of its high performance computing (HPC data center strategy), which was generally seen as a push toward AI. Wall Street again approved, shares soared 21%.

Skeptics may see it as an attempt to capitalise on a trendy trend, like when Long Island Iced Tea changed its name to “Long Blockchain Corp.” in 2017. But companies see it as a means to reduce systemic risks. Mining profits are correlated with bitcoin prices. By adding other services, such as hosting AI computing, the dependence on bitcoin is lessened.

The business wants to diversify its revenue streams. We’re less stressed if bitcoin goes up to $10k or even $20k. This allows us to act more strategically than our competitors.

Hut 8 started its pivot early. In January 2022, before ChatGPT became a buzzword, Hut8 had invested in five data centres, and two cloud regions, which could be dedicated to HPC. These data centers, unlike bitcoin-mining ASICs which are focused on a narrow range of tasks, have Nvidia GPUs. They can perform gaming, virtual reality and AI workloads. They still operate bitcoin mining rigs. Rayner says, “Our core hypothesis is that we envision a world in which mining and data centres and HPC workloads (used for AI) come together.” We’re beginning to see this more and more.

Hut 8 realized in 2022 that they had already completed some of the hard work to serve AI clients. “We already have the staff. We are compliant. “We have the expertise for operating traditional data centers,” says Rayner. Mining is closely related to that. “You have many of the same synergies.”

There’s also the shift from Ethereum mining to AI.

The ETH mining equipment was rendered useless when Ethereum changed from Proof of Work, which requires mining, to Proof of Stake. While the ASICs used for bitcoin mining can only mine bitcoin, Nvidia A40s chips that were used to mine Ethereum are much more versatile. Rayner says that the chips are capable of VFX render, gaming, and AI/ML workloads. The chips that were used to mine Ethereum are now being redeployed.

Rayner clarifies that old ETH chips have “limited scope” and “low memory and storage capacity,” but it is better to use them than let them idle.

How do the chips in Hut 8 drive the development of AI exactly? Here’s an example from the real world. XYZ AI is a startup that wants to convert text into 3D images. The startup needs to use a large data set in order to train its models. This requires extensive processing, which is only possible with chips. You can choose to purchase the chips, which are difficult to find for startups like XYZ. Or you can outsource computing to a cloud service provider like Hut 8 to handle the processing. It’s similar to the model of parking data in Amazon Web Services. Hut8 is essentially renting computing to XYZ.

This will enable XYZ users to type something like “show a sword that’s dripping in strawberry ice-cream” and magically see an image. The processing power required for generative AI is not the only AI category. Rayner says that there are “a zillion” of specific applications for which these chips are used. Rayner says that the demand for Hut 8 is exploding because of its many applications, including medical technology, games, biology and CAD drawings. Model training is used to train each of these “from a large data set.”

Mintgreen is a Canadian firm that uses heat recovered from bitcoin mining to generate power. It’s in the “conceptual stage” of exploring AI expansion. Colin Sullivan is the CEO of Mintgreen. “We have received questions from investors regarding the possibility that we could pivot,” he says. Sullivan says that the idea is only theoretical, and that he needs to “investigate the economics further.” However, Sullivan acknowledges that it would be prudent to diversify the company into other computerized electronics in the future.

Diversify is the key. The idea for mining-focused firms like Hut 8, Applied Digital and Iris Energy — as well as Mintgreen — is to create a broader portfolio, which will be more resilient in a bearish market. Rayner says that “we need to have a variety of revenue streams” so we can be the most efficient miners and stay agile and strategic in case margins compress. This could be the end of some bitcoin-only miner’s business; AI could also mean their survival.

Ben Schiller is the editor.