As demand for client PC chips is slows, some of TSMC’s largest customers are reportedly cutting down their orders to the world’s number one contract maker of chips. So far, Apple, AMD, and Nvidia have either reduced their orders, or plan to do so, reports DigiTimes. In particular, AMD is winding down orders for products made using N7/N6 nodes, whereas Nvidia is mulling cutting down orders that could impact the upcoming Ada GeForce RTX 40-series GPUs.
AMD reduced orders for products made on TSMC’s N6 and N7 process technologies by 20,000 wafers for Q4 2022 and Q1 2023. Since AMD is about to release its next-generation desktop CPUs (Ryzen 7000-series ‘Raphael’) and GPUs (Radeon RX 7000-series based on the RDNA 3 architecture) using TSMC’s N5 node, it’s logical for the company to reduce orders for chips made on previous-generation fabrication technology. However, AMD‘s current-generation Rembrandt processors are made using the foundry’s N6 technology. Furthermore, even next-gen Raphael uses an I/O die fabricated on the N6 node, so a potential reduction in N6 orders might affect supply and availability of AMD‘s future Ryzen 7000-series CPUs as well.
Nvidia is also looking forward to reducing orders for TSMC’s N5 (or perhaps 4N) production from TSMC. The company’s next-generation ‘Ada Lovelace’ GPUs are set to be made at TSMC using one of the company’s N5 nodes (presumably 4N, which is customized “4nm for Nvidia“). Nvidia‘s partners have plenty of GeForce RTX 30-series GPUs and boards, and miners are dumping their graphics cards and feeding a second-hand cards market at dirt cheap prices. Adding an all-new GeForce RTX 40-series into the mix could further oversupply the GPU market.
Since Nvidia (and AMD) made pre-payments to TSMC for capacity and agreed to certain terms, TSMC is apparently unwilling to reduce production of chips for Nvidia, but it agreed to delay the first deliveries, possibly up to Q1 2023, which essentially means that Nvidia could delay the GeForce RTX 40-series launch. This would allow Nvidia and its partners to sell off existing cards, but it could also provide AMD with an opportunity to release its Radeon RX 7000-series GPUs before Nvidia launches its next-gen offerings.
More likely is that Nvidia will launch the new GeForce RTX 40-series ‘Ada Lovelace’ GPUs, but it won’t ship as many models at first in order to let its partners sell off their existing GeForce RTX 30-series inventory. Regardless, it’s unclear whether demand for previous-generation products will be strong enough once RDNA 3 and Ada Lovelace graphics cards make their appearance, especially not at the wildly inflated prices of the past year, which is why so many companies are heavily pushing “GPU sales” right now — sell them while you can.
Even Apple, one of the world’s largest makers of consumer electronics, is reportedly cutting down orders for iPhone 14 components by 10% from 90 million initial units, DigiTimes notes. No clear reason was given, though softening demand and no real need for users to upgrade are likely factors.
While there might be a temporary oversupply of client CPUs, GPUs, and SoCs, the industry seems to be optimistic about the long-term demand. At present AMD, Apple, Broadcom, Intel, MediaTek, Nvidia and Qualcomm are all queuing up for TSMC’s next generation N3 (3nm class) process capacity, reports DigiTimes. It remains to be seen who is going to use which process — there are currently five variants slated to arrive in the next year or two: N3, N3E, N3P, N3S, N3X. N3 will be a long lasting node for TSMC, so we are likely talking about multiple fabrication technologies here.