The rise of AI in its different modalities has generated an enormous demand for solutions, both hardware and software, capable of facing the challenges that this represents, and according to the latest information, both NVIDIA and Meta have had to modify their initial plans to to expand and improve their data centers ready for artificial intelligence.
This has an important consequence, and that is that both NVIDIA and Meta need a greater number of components, and in this sense it seems that there are two that are not receiving as much as they really need, DDR5 memory and HBM memory. Therefore, both companies have requested a higher level of supply of both components to SK Hynix, a company that, as many of our readers will know, is one of the largest providers of both solutions.
Meta has an important ambition with its commitment to renew its servers and prepare them for artificial intelligence, implement generative AI in their social networks. If you don’t know what this quiet variant of artificial intelligence is, I’ll explain it to you in a moment. It is a type of artificial intelligence that is capable of creating text, images and other content in response to specific commands.
The company that Mark Zuckerberg runs needs DDR5 memory, while NVIDIA wants HBM memory of different generations, although it is confirmed that the company has already started using HBM3e, which is the next generation standard. This is totally understandable, since the green giant uses this type of memory in the manufacture of its graphics accelerators based on the Hopper architecture.
It is said that both companies are closing key deals with SK Hynix to ensure the supply they need of both components, and this could end up boosting the economic results of the company which, I remind you, is one of the most important manufacturers of HBM memory, so much so that today its market share in this sector around 80%a huge piece of information that allows us to easily understand its position.
HBM memory is key for graphics accelerators because allows you to stack a huge amount of graphics memory in a very small space, and because it also offers a gigantic bandwidth, thanks to the use of huge buses that can provide a bandwidth of up to 5TB/s.