Update March 16, 2023: Intel has informed us that an interpretation of a comment given by Nvidia CEO, Jensen Huang, to CNBC’s Katie Tarasov—which seemed to suggest that it was open to using Intel’s Arizona fab for new GPU production—is incorrect.
When Huang was quoted as saying: “oh absolutely, we’ll use Arizona,” he was referring to TSMC’s advanced node facilities being constructed in the state, and not Intel’s own fab in Arizona.
Original story: Ever seen a man eat his own head? Probably not, but you’re very likely going to see something almost as strange, an Nvidia GPU made by none other than arch rival Intel.
Nvidia’s CEO and leather jacket aficionado Jensen Huang has seemingly confirmed that the company plans to use Intel’s new cutting-edge chip production facility in Arizona.
At this point Nvidia isn’t detailing what chips exactly. So, it could be something other than GPUs. But Nvidia is primarily a GPU company, so Nvidia GPUs made by Intel just became a very realistic possibility.
Of course, we’ve seen oddities like this before, as arch rivals come together in some pragmatic shared cause. It wasn’t that long ago that Intel teamed up with AMD to insert Radeon Vega M graphics into an Intel mobile CPU package.
That never really seemed to catch on and Intel has since embarked on its own graphics chips with the Arc series of GPUs. But it’s an example of how a competitive relationship can also be symbiotic.
Making chips for customers is central to Intel’s plan to reinvent itself as not only a full integrated chip design and manufacturing powerhouse, but also as a foundry to compete with the mighty TSMC.
As we recently reported, Intel is sticking publicly to its aggressive plans to roll out no fewer than three new chip production nodes in the space of a year, starting with its new Intel 4 node, formerly known as 7nm.
Intel 20A supposedly follows in early 2024 with Intel 18A in late 2024, at which point Intel claims it will have the most advanced chip production technology around.
For Nvidia, a cutting edge alternative to TSMC would be very handy indeed. Partly, that will help to keep TSMC honest when it comes to pricing. But in strategic terms, being a fabless company relying on just one option for manufacturing must be a bit terrifying, especially when that option is located on an island in the South China Sea that is currently the subject of extreme tense geopolitics.
Admittedly, Nvidia did go to Samsung for its Ampere generation of GPUs including the RTX 3090. So TSMC isn’t literally the only option. But Ampere, though fairly effective in performance terms, was a pretty inefficient architecture, in no small part thanks to the characteristics of Samsung’s 8nm node, which most industry observers consider far inferior to TSCM’s 7nm technology of the time.
Anyway, it’s not a nailed on certainty that Nvidia GPUs will be made by Intel. Heck, it’s far from guaranteed that Intel will get its manufacturing technology sufficiently up to speed that Nvidia will even want to use it. But Nvidia graphics made by Intel very much seems to be the plan. And that will take some getting used to.
