NVIDIA CEO Allegedly Booked Secret Meeting With TSMC To Secure 3nm Wafers

leather jacket man jensen huang
According to Chinese-language site MyDrivers, the famed leather jacket aficionado (who also happens to be the CEO of NVIDIA) Jensen Huang recently returned from a trip to Taiwan. While there, he reportedly visited TSMC to ensure that his company would be among the first to have access to the foundry's 3nm process when it enters mass production next year, but arguably more interesting were his rumored visits with AIB partners.

It's no surprise to anyone that follows this stuff that NVIDIA wants dibs on 3nm chips. GPUs are the biggest processors around, and NVIDIA's in particular, especially since there's no indications that the company is moving to chiplets anytime soon. The kinds of workloads you run on GPUs tend to scale very well with additional parallelism, and improved density means that Mean Green can stuff MOAR COARS into its chips.

hopper die nvidia
NVIDIA's Hopper H100 GPU is one of the largest microprocessors ever manufactured.

The visits with the company's mostly-Taiwan-based AIB partners are arguably more interesting because rumors have been going around for a while now of NVIDIA's poor treatment of its partners. Those rumors predated the EVGA bombshell announcement back in September, but that whole debacle, where EVGA claims that its complete exit of the GPU market—some 80% of the company's revenues—was in response to "disrespectful" behavior from NVIDIA, certainly fanned the flames.

When NVIDIA canceled (or "unlaunched") the GeForce RTX 4080 12GB, one of the first thoughts in some enthusiasts' heads was about how it would affect AIB partners. Those cards are likely manufactured already, to say nothing of printed packaging and marketing materials. The company publicly committed to bearing the cost of pivoting those products to a new branding, so if we had to guess, we'd say that Jensen's visit was probably to smooth ruffled feathers in person, and assure the company's partners that it will continue to work with them.

geforce rtx family
NVIDIA's not talking about lower-end Ada yet, likely to move Ampere product in the channel.

Jensen also probably discussed the matter of the purported Ampere overstock situation. It's been rumored by several sources—including MyDrivers—that NVIDIA and its AIBs are struggling with a major overstock of GeForce RTX 3000 parts, particularly those based on GA102 silicon, which includes everything in the 3000 series from the original GeForce RTX 3080 10GB model on up.

That would certainly explain the bargain-basement pricing we've seen for GeForce RTX 3090 boards, but the lower tiers of Ampere product haven't come down in price the same way. The situation may change in the coming weeks if the product doesn't start to move more quickly, as AIBs are likely eager to launch the GeForce RTX 4080 (16GB), a card that's sure to see wider adoption than the rather limited supply of GeForce RTX 4090 cards.