Artificial Intelligence seems to be the technology or rather the phenomenon of the moment. About a year ago it was Web3, but other innovative and emerging technologies have overtaken it. While many companies are making (or on the way to) a lot of money, it’s NVIDIA that’s making a fortune, running out of graphics cards for AI.
Being a visionary is not easy. Jensen Huang is arguably one of the greatest modern visionaries (well, that and his legendary wack). Although we know their graphics cards from gaming, they are actually used for many other things like advanced maths, architecture, medicine etc.
For years they have been preparing for a more specialized field of science fiction, such as artificial intelligence. Thus Tensor Cores were born, specialized cores designed for AI that are currently the basis of DLSS. Launched in February 2019, this technology represents an advancement of its times that the company benefits from.
Years of development that pay off
Graphics cards specialized for AI, such as the NVIDIA A100, have increased in price by 37.5%. These graphics cards are not sold individually, they are integrated into specialized servers which have become more expensive.
Specifically, they are mounted on NVIDIA DGX A100 servers, which include six A100 graphics. These systems cost $200,000 per unit when announced in 2020. Currently, they may already be hovering around $275,000 per unit due to the strong demand for AI.
According to IDC China, this is not an isolated incident, it is happening with other products of the company. They also currently offer the China-exclusive A800, which is a cut-down version of the A100. As mentioned, the price of the A800 has increased by 20%. But, as if that weren’t enough, demand is so strong that there is no stock available.
Till February this year the situation was good, products were being dispatched in a month, which is normal. Currently, the situation is so complicated that the minimum delivery time is three months. For some models (not specified) it is indicated that they will not have stock until December.
To give us an idea, the NVIDIA V100, which is the easiest for AI, is already around $10,000. The H100s, on the other hand, topped out at $35,000 per unit. Furthermore, it looks like these prices are low and will continue to rise.
TSMC, which makes the chips for these graphics cards, is working to ramp up production as quickly as possible. The truth is that turning on the printer is not that easy, the process is slow and tedious.
Again, at the right place at the right time
For those of us who have been in the hardware business for so long, this should come as no surprise. We have seen a brutal shortage of stock of graphics cards for Ethereum mining in the past. Such was the demand and madness that planes were chartered to transport the graphics cards.
History repeats itself, now, in another row. For now, it doesn’t seem to affect the graphics card at the moment. As this vortex continues and it is no longer possible to meet demand, it is likely that companies will find themselves thrown out there. It cannot be ruled out that, in a few months, if AI continues to grow exponentially, we could see a shortage of gaming graphics cards.
It should not be ruled out that Nvidia may reduce the production of gaming solutions in favor of AI products. After all, it is a company and its job is to make money.