AI data centers have a big energy problem. SK Hynix promises to cut its cost by 30%

The proliferation of large artificial intelligence (AI) data centers is creating a very serious energy problem. So much so that the U.S. Department of Energy is being considered an opportunity for companies that have large data centers dedicated to training AI models to set up a small nuclear power plant nearby that can meet their energy needs.

But if the Energy Department believes that nuclear is the ideal option for meeting the electrical needs of large AI data centers, the total contribution of this form of energy to the overall mix would have to be much higher. And that plan, logically, requires a lot of investment. Microsoft, in fact, is one of the U.S. companies seriously considering using nuclear energy to power its data centers.

The strategy would also strengthen the U.S. commitment to energy sources that do not emit greenhouse gases. It is not yet clear what investments tech companies will make or what subsidies the government will provide. At the moment, some major tech companies have already invested in nuclear energy, although not necessarily in nuclear power. Microsoft, for example, has an agreement with Helion Energy to obtain power from its fusion reactors in the future.

SK Hynix and his memoirs have something to say

South Korean company SK Hynix is ​​currently the second-largest memory chip maker on the planet. Only South Korean company Samsung has a higher market share. However, SK Hynix has a big advantage: it is the main supplier of NVIDIA memory chips. In practice, this means that most next-generation data centers designed for artificial intelligence use their memory.

The GPUs responsible for running AI processes are largely responsible for the high power consumption of these data centers, but memory consumption This is not at all unimportant. This is where SK Hynix comes into play. And the company has just announced that it has prepared new sixth-generation DRAM memory chips manufactured using 10nm lithography.

The most important feature of the new 16GB 1c DDR5 DRAM chip, as they call it, is that it offers more than 9% higher energy efficiency than the previous generation of DRAM from this manufacturer. A 9% improvement may not seem like a big deal, but it is. This is very important because each server integrates a large number of these chips, which SK Hynix believes will allow AI data center owners to reduce their energy costs by at least 30%.

There is no doubt that this is a significant savings. It is certainly not the final solution to the AI ​​data center energy problem, but This is a significant contribution. which will increase its resilience until nuclear power plants or another strategy takes over.

More information | Reuters | DigiTimes Asia

In Hatake | Fusion milestone in Europe: JET reactor breaks crucial record on path to commercial fusion power