Intel CEO Patrick Gelsinger is confirmed to have landed in Taiwan and met with key TSMC executives to secure sub-5nm node capacity. Accompanied by a number of high-level execs, Gelsinger hopes to reserve enough of TSMC’s 3nm capacity to meet the demand for the next-gen Arc GPUs (and chiplet-based Meteor/Arrow Lake CPUs). As per DigiTimes, however, the 3nm contract was already finalized during Bob Swan’s reign, and won’t be affected by Gelsinger’s present visit.
The outsourcing contract between TSMC and Intel is expected to last till 2025 at the very least. Intel’s own process nodes won’t be catching up anytime before that. Furthermore, the existing capacity won’t be enough to meet the demand for the Xe GPUs, in addition to the Core and Xeon CPU lineups. This means that Intel is liable to become TSMC’s largest client by 2023, undercutting not only AMD but Apple as well.
With the production capacity for two Arc GPU lineups, paired with the I/O or GPU chiplets for the Meteor/Arrow Lake CPUs, Intel will be relying on TSMC for its flagship products more than ever. Team Blue might become one of the top two advanced foundries by 2025 with its 20A process. The chipmaker’s packaging technologies, most notably Foveros, and EMIB as demonstrated in Ponte Vecchio are some of the most advanced in the industry, giving it a unique advantage in the foundry business. Qualcomm and AWS are slated to be the first third-party clients of the node (in 2024).
As for the 3nm node itself, Apple should continue to lead the production shares, with Intel coming in 2nd, and pushing AMD to the third spot. Considering how the latter is completely reliant on TSMC for its entire portfolio, it’ll be interesting to see how the addition of Intel as the second customer will affect the dynamics.