Are Bitcoin miners transforming into artificial intelligence a hype or an opportunity?

Original Title: 《Can Bitcoin Miners 'Do' AI? Not Really》

Written by: Colin Harper, Head of Content at Luxor Technologies

Translated by: Luffy, Foresight News

The Bitcoin industry, especially the mining sector, is no stranger to hype. And now, a new technology has emerged: artificial intelligence.

ChatGPT and other large language model programs (commonly referred to as AI) have made waves in the workplace and online world this year. As a result, references to AI naturally appear in cryptocurrency news.

For example, some Bitcoin miners have publicly stated that they are seizing opportunities in the AI space, trying to make outdated hardware work. But this does not mean that every company is executing the same functions or achieving the same scale as AI data centers. In fact, the tasks performed by this outdated hardware are quite different from those required for what people typically refer to as AI services; some miners are simply trying to profit from chasing the trend.

Can Bitcoin Miners Provide AI Services? It's Complicated

In recent months, HIVE, Hut 8, and other Bitcoin mining companies have announced new high-performance computing (HPC)/AI strategies. Most of them have computer hardware that has been rendered obsolete due to Ethereum's transition to a proof-of-stake consensus mechanism, which ended the era of Ethereum mining. Some miners claim they can rent out their computing power for high-performance computing (although Hut 8 has been doing this since 2022).

High-performance computing is a catch-all term for an arbitrary number of data center tasks. However, when miners talk about HPC, they specifically refer to things like cloud computing or graphics rendering. Hut 8 earned $4.5 million from such services in the first quarter and $4.3 million in the second quarter, while HIVE's pilot project earned $200,000 in the first quarter, but the miner did not mention any HPC revenue for the second quarter in its filings with the SEC. Hut 8 has signed a contract with the Interior Health Authority of British Columbia to provide high-performance computing services until 2028. The company purchased a 6-megawatt facility that provides these services when it acquired the Toronto TeraGo data center business in 2022.

These types of computing are different from those used to train AI models or support ChatGPT. Miners do not have the appropriate computing equipment to support such work, nor do they have the necessary power and network infrastructure (AI computing requires at least 1 TB of network bandwidth per second, which industrial-scale mining facilities fall far short of; their bandwidth is typically around 1 GB per second).

The cost of a real AI data center could be 10 to 20 times the price per megawatt of a Bitcoin mining facility (according to data from the ERCOT, one megawatt can power about 200 homes in Texas). Additionally, the power costs for such facilities would be double what miners currently pay, potentially reaching 15 cents per kilowatt-hour. Bitcoin miners in the U.S. typically pay 5-8 cents.

The cost of running HPC facilities is far higher than that of Bitcoin miners

Riot Platforms estimates that the cost of its upcoming Bitcoin mining facility near Corsicana, Texas, will be $833,000 per megawatt of power capacity, with an initial construction capacity of 400 megawatts.

In contrast, the average cost of HPC data centers in North America is $9.5 million per megawatt (according to Turner & Townsend's 2022 data center index).

The high costs stem from the complexity of these data centers. They require higher quality power infrastructure and network equipment, not to mention they need more space, better cooling, and more powerful backup generators to ensure 99.99% uptime during power outages. In short, these miners lack the funds or market capability to build AI systems, let alone a traditional data center.

This cost applies only to traditional high-performance data centers and does not necessarily apply to data centers with driving engine capabilities like OpenAI's ChatGPT and Microsoft's Bard.

Miners May Attempt High-Performance Computing, But Not AI

Many Bitcoin mining companies were born from sudden shifts in business lines, but at this stage of development, these companies should be cautious when trying to curb the hype surrounding HPC/AI.

The two largest publicly traded Bitcoin miners, Riot Platforms and Marathon Digital, announced their Bitcoin mining businesses in 2017, but these companies were listed long before they ventured into cryptocurrency. Marathon Digital was formerly Marathon Patent Co., which profited from buying and selling various technology patents, while Riot was a pharmaceutical company.

Both companies have become among the largest Bitcoin miners in the world, so this transformation has paid off.

However, for a major Bitcoin miner, transforming its mining facility into an AI data center would be extremely difficult and expensive. It is also worth noting that in the traditional data center market, miners would be competing with some of the world's most powerful tech companies like Google, Amazon, and Microsoft.

Applied Digital is a special case: it is a data center company that also helps clients host Bitcoin mining equipment. This distinction is important because traditional data center operators find it easier to deploy Bitcoin mining equipment on-site than miners do to build a data center from scratch. The AI hype wave of 2023 pushed Applied Digital's stock price up to $10.24, but as of this writing, the stock has significantly retreated to $5.68. As shown in the chart below, Applied Digital and mining stocks are currently on fire, so we advise investors considering these stocks to proceed with caution.

This year, Bitcoin mining stocks have soared alongside related assets, but many hope that AI can reclaim bullish momentum.

Conclusion

Once the hype dies down, the market will tell us which companies are in a favorable position to ride the wave of AI.

For miners, they can provide graphics rendering services, but they will not use their mining hardware for ChatGPT. Therefore, when you see AI in press releases, do not be misled; it may not be what you imagine.