Some are investing, while others are speculating.
Written by: Zhou Yixiao, Silicon Star Pro
After Trump won the U.S. presidential election, the market expected him to relax regulations on cryptocurrencies. Trump had previously stated that Bitcoin mining should be concentrated in the U.S., but such policies could have implications for the AI industry. On November 23, Bitcoin prices once reached $99,660, setting a new historical high and approaching the $100,000 mark.
Because both Bitcoin mining and AI training require a large amount of energy and computing power. The synchronized development of both is bound to create competition for electricity and hardware resources. This means that the AI training business may be affected by fluctuations in Bitcoin prices, especially when miners compete for limited hardware resources. In other words, an increase in Bitcoin prices may lead to an increase in AI training costs.
AI VS BTC
With the tremendous success of ChatGPT, AI companies are racing to train and run their own models, hoping to surpass OpenAI's flagship product. This has generated a massive demand: the inference process of AI models is much more complex than the indexing and retrieval process of search engines, and the energy consumed by a single ChatGPT query is about ten times that of a Google search.
This has led AI companies to urgently seek cheap energy and large areas of land to accommodate related equipment. In North America, some regions have implemented a queuing system to wait for large data centers to connect to the grid. However, once a company receives preliminary approval, building a data center from scratch can take several years, cost millions of dollars, and involve a lengthy regulatory and administrative approval process.
Internationally, large-scale Bitcoin mining has always been an extremely profitable business. However, it is also affected by the highly volatile cryptocurrency market. After the cryptocurrency market crash in 2022, many miners were forced to go bankrupt or completely shut down their businesses.
In 2023 and early 2024, mining companies that survived the market downturn reaped profits. However, this year's Bitcoin halving (the reduction of miners' rewards) did not trigger the same dramatic rise in Bitcoin prices as in previous cryptocurrency cycles to offset the impact of the reduced rewards. Since April of this year, the long-term stagnation of Bitcoin prices has compressed miners' profit margins, forcing some miners to seek business model diversification to hedge against the risks of cryptocurrency price fluctuations.
Four years ago, when data center and Bitcoin mining company IREN considered entering AI training, they believed that the business volume at that time was not sufficient from a commercial perspective. But now, an increasing number of large Bitcoin mining companies have begun to replace some of their mining equipment with devices used for running and training AI systems. These companies believe that providing computing power for AI companies may yield a more secure and stable source of income than mining.
Today, the cooperation between the artificial intelligence and Bitcoin mining industries is quite logical, as both sides have aligned needs. AI companies require the existing sites, already connected cheap energy, and infrastructure of Bitcoin miners. Meanwhile, Bitcoin miners pursue stable income brought by AI business, as well as the potential profits from the current AI hype.
Some Bitcoin mining companies choose to lease their facilities to AI clients. In June of this year, the Bitcoin mining company Core Scientific, which was on the brink of bankruptcy in 2022, announced that it would host over 200 megawatts of GPUs for the AI startup CoreWeave. Core Scientific stated that AI companies have started buying mining facilities at prices higher than the mining market, referring to Bitcoin mining facilities as 'power shells for the data center industry.'
Some other Bitcoin mining companies operate GPUs themselves. Bitcoin mining company Hut 8 received a $150 million investment from Coatue Management to build artificial intelligence infrastructure. Some facilities of the Australia-based mining company IREN share space for GPUs used for AI and ASIC devices used for Bitcoin mining. Bitcoin brings immediate income but is highly volatile. AI relies on clients, and once there are clients, it becomes more stable. Nasdaq-listed Bit Deer is also building its own AI data center in Singapore.
It looks like a beautiful business.
However, only a few overseas mining companies can achieve this transformation. Secondly, the equipment used for mining Bitcoin is called ASIC, which stands for Application-Specific Integrated Circuit, and 'specialized' means it cannot be used for other tasks. Mining companies cannot seamlessly transition mining equipment to AI scenarios.
An AI Infra industry practitioner told Silicon Valley Star, 'For example, the H100 is generally used for training models, while mining uses the 4090.'
In other words, if Bitcoin miners want to serve the AI industry, they must purchase brand new equipment, and the requirements for data centers for artificial intelligence and Bitcoin mining are different. Entering a brand new and highly complex industry is already fraught with difficulties, not to mention competing with well-funded tech giants like Google, Amazon, and Microsoft.
Therefore, not all mining companies can replicate the high-level cooperation between Core Scientific and CoreWeave. Especially smaller miners, who actually have very few resources to offer to the AI industry.
Domestically, virtual currency mining has been banned, and there are no miners transitioning to AI. However, businesses from other industries want to share a piece of the AI wave. They either enter the field directly or establish computing power subsidiaries to engage in 'computing power leasing' business. According to statistics, more than 100 listed companies in the A-share market are involved in computing power leasing concepts, including 'lottery printing king' Hongbo Shares and 'monosodium glutamate king' Lianhua Holdings, among others. On video platforms, there are even stories like 'I拆了老家一套房，买了八百张显卡，和初中老同学一起合伙做算力租赁' (I tore down a house in my hometown, bought 800 graphics cards, and partnered with my middle school classmates to do computing power leasing).
Ideally, the business model of computing power leasing only requires an initial investment in GPU server equipment, hosting the hardware in specialized intelligent computing centers, and leasing the computing power to end users, with hardware operation and software services managed by the intelligent computing center.
However, in reality, this may not be a good business. The demand for computing power leasing comes from the development of the AI large model industry, while the rental costs for high-end hardware to train AI are plummeting. Featherless.Ai CEO Eugene Cheah pointed out that the rental price for overseas H100 GPUs once reached as high as $8 per hour, but has now fallen below $2 per hour. This is mainly because some companies signed early power leasing contracts and, to avoid wasting idle capacity, began reselling reserved computing resources, while most of the market opted for open-source models, leading to a decreased demand for new models.
The domestic computing power leasing market is also experiencing a similar 'computing power surplus' phenomenon, but 'the leasing market is unlikely to drop prices because all were bought at high prices in the early stages,' a practitioner in the intelligent computing industry told Silicon Valley Star.
'Still this fast'
There is a famous saying in the cryptocurrency circle: 'computing is power,' and computing power is power; this saying has now circulated into the AI circle.
Behind computing power is energy, and there is a close relationship between developed countries and high energy consumption. If we compare the per capita electricity generation (kilowatt-hours, kWh), we can see this. In other words, obtaining excess energy is a necessary condition for the manifestation of civilization's progress. After all, the basic survival level of agriculture is layered above manufacturing, transportation, public services, urbanization, and computation, all of which require energy support.
In this dimension, the infrastructure originally established to serve cryptocurrency is now providing solutions for the computing power demands of the AI era. This is undoubtedly an opportunity for the overseas digital currency mining industry, which has always hoped to rid itself of speculation, to prove its value. As long as this trend continues, leading manufacturers will benefit from the enthusiasm and liquidity brought by AI.
Every wave of technological revolution is accompanied by a 'gold rush.' For speculators among them, what they pursue is always profit itself, regardless of whether the asset is cryptocurrency, artificial intelligence, or tulips from three centuries ago, it seems unimportant.
After the Bitcoin halving, some miners faced a dilemma: either continue mining and hoarding coins, hoping for a rise in Bitcoin prices, or transition to AI data centers, hoping to ride the wave of artificial intelligence and make quick money. Now that the coin price has reached a historical high, some people begin to lament, 'Still this fast.' But there is also a saying in the cryptocurrency circle, 'Holding coins is harder than holding a mistress.'
As people oscillate back and forth between the cryptocurrency circle and the AI circle, this repetitive process reminds one of what Keynes said: speculators do not care who is the most beautiful; they only care who will be chosen in the beauty contest.
And this 'beauty contest' will continue endlessly.
