In mid-September, the Ethereum community transitioned its consensus algorithm from proof-of-work (PoW) to proof-of-stake (PoS) in a process dubbed The Merge that reduced the power consumption of the Ethereum network by 99. 95%, and sounded the death knell for many blockchain miners. Now, anyone with a few coins, an average computer that could run a basic version of Linux could, simply by consigning a few ethers (thirty-two in theory, but there are services that allow you to put half or a quarter), obtain a virtually risk-free return close to a very attractive 8%.

The consequence was immediate: the large GPU farms destined to solve the complex cryptographic equations characteristic of the proof-of-work no longer had anything to offer to an Ethereum network that didn’t needed them any longer, and had to find other things to do. Some estimates suggest that, at its peak, Ethereum mining came to support more than a million people using equipment valued at $10 billion.

The possibility of mining other cryptocurrencies was there, but with the prospect of diminishing returns in an increasingly saturated market like bitcoin, in which miners, in any case, will lose most of their appeal when the last bitcoin is mined (more than 92% of them have already been mined) and have to survive on transactions alone, and in environments with little promise in the medium and long term like shitcoins, the difficulty was obvious.

Some, after starting to lose money, simply ended up shutting down their servers.

However, in an increasingly digitalized society, owning a data center capable of providing huge computational resources has considerable potential. And indeed, we are seeing how former pools dedicated until September to Ethereum mining are now starting to offer their services to process generative algorithms, using not only their equipment, but also the expertise managing their performance and power consumption. An estimated 5% to 15% of those previously involved in Ethereum mining can repurpose their equipment for use in high performance computing (HPC) and machine learning algorithm functions of various kinds, such as text generation, image generation or computer vision.

Now, the growing supply of services based on generative algorithms is demanding energy, the new service requiring huge computing resources.

And before, there are critics, but it is important to put things in context: if bitcoin mining, at its peak, consumed about 253TWh, gold mining consumes 571TWh, video games around 214TWh and Christmas lights, 201TWh. YouTube alone consumes 244 TWh.

These are undoubtedly important figures, but we should reflect on them a little more and not just stick to sensationalist headlines such as “the annual use of YouTube could provide electricity to an average American household for approximately 2 billion years, or to all 127 million American households for approximately 8 years”, when evaluating them.

In the end, the migration of computing power and its electricity consumption is a variable that allows us to understand the dedication of resources to the development or use of a technology, no more and no less, regardless of what that technology may mean in the future, be it billions of videos of kittens, a viable neutral and global monetary system, or a machine capable of answering questions. But all in all, high performance computing is becoming a valuable commodity as there seems to be no shortage in sight of potential uses for it… which is perfectly reasonable in our increasingly digitized world.