Miners limit cryptocurrency mining, GPUs are also used for HPC and AI tasks

Two years ago we had a huge crisis here with the unavailability of graphics cards and their absurdly high prices. The prices of cryptocurrencies at the time forced a lot of people to buy modern (and later also older) graphics cards and start mining cryptocurrencies (and there wasn’t much left for players). Mining companies caught on to it in a big way, and scalpers who resold insufficient graphics cards at insane prices also joined in. Now the situation has reversed and the price of cryptocurrencies is still much lower than it was then, despite the slowly increasing value again. It is not surprising that mining stops paying, for which there are several reasons besides the lower price. The complexity of mining continues to increase, we also had Ethereum transition from proof-of-work to proof-of-stake. Cryptocurrencies that still make sense to mine on GPUs are no longer among the biggest. On the contrary, the high interest in artificial intelligence, led by the chatbot ChatGPT in particular, is moving the use of GPUs to other areas.

E.g. crypto mining company Hut 8 experienced a huge drop in income and while last year it mined 942 Bitcoins in the first quarter of the year, this year due to the increasing complexity of mining, despite a slightly higher hashrate of 2.6 EH/s against last year’s 2.5 EH/s, it was only 475 Bitcoins (428 of them has been sold). Add to that the drop in the average price of Bitcoin sold from last year’s $52300 (a figure according to Hut 8, I’m not aware that the price of BTC will be at that level in early 2022) to now $30600, it’s no wonder that her income has plummeted for mining from 49.3 to only 14.5 million USD. This means that mining revenue is now only 30% of last year. In addition, ether cannot be mined on the GPU for the above reasons, so this part was stopped and the GPU more or less idled, which was another reason for such a significant drop.

The company is trying to improve this by focusing more on HPC and using GPUs for AI tasks instead of mining cryptocurrencies (especially the no-longer-mined Ethereum). GPUs are not suitable for Bitcoin mining, for that you need special ASICs. It already increased it last year with the purchase of TeraGo. Unfortunately for the company, it is only a very weak band-aid. While last year HPC generated revenues of USD 3.3 million (ie about 6% of all revenues), this year it has grown to USD 4.5 million. Although HPC now accounts for almost 24% of the company, it still means a huge drop in the company’s revenue from 53.3 to 19 million USD (-64%).

Other mining companies are also planning to increase the share of HPC and AI, but it is not expected that more than some 5-15% of purchased GPUs, which were mainly used to mine Ethereum, will be reserved for this. For the time being, they are rather trying to find a use for already purchased graphics cards, rather than buying new ones because of AI. So far, it doesn’t look like we should see another GPU crisis anytime soon, with a shortage of graphics cards and skyrocketing prices, even though AI projects are mushrooming. However, if the number of projects were to increase dramatically and there would be an acute shortage of computing power, something like this cannot unfortunately be ruled out. However, AI projects are not expected to run on smaller machines very often (similar to how you can mine in a pool on a single GPU at home). AI usually requires a faster response, however, who knows, maybe in the future we won’t be mining cryptocurrencies on GPUs, but processing algorithms using AI for money. Do you think something like this is realistic and will be common?