TSMC has is said to have notified its customers that wafer and manufacturing prices will increase sharply in the near future. According to DigiTimes, the price hikes will go into effect in 2022 and will impact existing nodes differently. Sub-16nm prices, including 12nm, 7nm, and 5nm, are said to have increased roughly 10 percent, while TSMC’s older nodes have gone up by as much as 20 percent.
This differential shift likely reflects differing patterns in product demand. The ongoing silicon shortage is comprised of many smaller shortages, covering everything from GPUs to tiny display drivers built on legacy process nodes. We’ve had somewhat mixed signals from the PC market, where the guidance has been somewhere between “slow improvement but no return to normal until 2022,” and “everything could remain thoroughly screwed up until the end of 2022.” The difference between these two states, practically speaking, may be the difference between buying a GPU at 1.1x – 1.2x MSRP and getting slugged for 1.5x – 3x.
We don’t have details on every price hike, but DigiTimes reports that the price of a processed 28nm wafer will increase to nearly $3,000 in January. Currently, TSMC derives 49 percent of its revenue from 7nm or 5nm and 51 percent of its revenue from all larger nodes. 28nm and 16/12nm account for the majority of the remaining share, at 25 percent, with the last 26 percent split between six legacy nodes.
Eyeballing TSMC’s revenue share by platform, HPC contains all of the revenue from customers like AMD, Nvidia, and most likely Intel. While DCE dropped in Q2 2021, total company revenue was up 28 percent Y-on-Y and 2.9 percent Q-on-Q, with a net company profit margin of 36.1 percent.
There are several reasons why TSMC might raise prices. It’s a good way to tamp down demand, which is useful in a situation where you can’t sell as many chips as the market wants to buy. It also accounts for the fact that TSMC, GlobalFoundries, and other companies are running their fab equipment at breakneck speed, with less time to perform maintenance in-between manufacturing runs. Most of the Tier 1 and Tier 2 foundries have all indicated that they’re running at absolute maximum speed. TSMC, Intel, and Samsung have all committed to large capacity buildouts, while smaller fabs like GlobalFoundries and UMC are also investing a large amount of money in increasing production relative to their typical yearly capital outlays.
It is unclear how much of these cost increases will be passed on to customers. Companies that ship finished consumer electronics like Apple, Microsoft, and Sony can absorb the price increase more easily because the value of the CPUs and GPUs inside these systems is only part of the total unit cost. For a company like AMD or Nvidia, the cost of a wafer is more directly reflected in the cost of the chip. News that GPU prices might increase next year after years of pandemic-inflated pricing and the impact of trade tariffs before that is particularly unwelcome, but we’ll wait and see how much of this is passed on to the consumer before declaring it a guaranteed outcome.
If you are in the market for new hardware, it may be a good idea to pounce on any deals you find sooner rather than later. It’s always possible that pandemic-related demand will prove to be somewhat illusory, but the Delta variant is wrecking worldwide timelines for reopening. Lockdowns and outbreaks in Malaysia are said to be meaningfully slowing chip testing and packaging efforts in that country, and Jen-Hsun Huang does not believe GPU manufacturers will be able to fully meet demand until the end of 2022.
