After finishing the last GTC, more details and post-event statements are now known from the CEO of the most profitable chip company in the world. NVIDIA is reportedly still reliant on gamers for most of its revenue, specifically the 47%, but the volume is reduced as the cards are not sold and companies start to buy more and more add-on products such as A100 GPUs, Quantum systems etc …
Supply, demand, chip shortage and NVIDIA
The supply continues to exceed the demand and although we have already seen that there is a date for the end of this, Huang goes one step further and advances the date by almost a year, which is logically surprising:
“I think that during the next year the demand will far exceed the supply. We have no silver bullet to fix supply chain problems. We also have the support of our suppliers. We are fortunate to have multiple sources and that our supply chain is diverse, as our company is quite large, so we have the support of a large ecosystem around us »
In other words, next year manufacturers will have the opportunity to create more chips than assemblers and users ask for, something that will undoubtedly increase the rampant inflation that we are experiencing right now due to overinvestment, the US and China included.
NVIDIA could open up to other designers
We have to go back to 2013 to remember Huang’s own comments on the IP issue. And it is said that many of the IPs of its GPUs would be licensed to third parties, but there is little or nothing of that today.
Instead, it was the CEO himself who after this GTC 2021 has slipped that NVIDIA is open to the idea of creating processors that are not theirs as they work and integrate into the software ecosystem for the development of CUDA. Not surprisingly, Omniverse goes a bit like this as we saw in its presentation, but this has a counterpart that will not occur.
There will be no open source at NVIDIA. And it makes sense, since their software is part of the appeal of their platforms and products, as Huang himself made clear:
Underneath CUDA is NVIDIA hardware. There really is nothing of our open source. If someone wanted to create an application for CUDA or create another chip for CUDA, we are not fundamentally against it, but nobody has ever asked for it. “
Why is NVIDIA declaring this? Because Amazon, Google or Facebook They require custom chips for their tasks, to be faster and more efficient and NVIDIA will not say no if they knock on your door as they have done with AMD or Intel for example, you just have to look at the TPUs.
Ultimately, NVIDIA is covertly opting for the strategy of Intel IDM 2.0 but on a smaller scale, since it is not a wafer manufacturer, but it is open as Intel is now for companies to design under their IP, so we could see interesting products in not long with their label.
