GPUs and SoCs will become more expensive.
According to reports, TSMC will raise the price of wafers processed using its cutting-edge N3 (3nm-class) process technology by 25% when compared to the N5 (5nm-class) production node. This immediately raises the cost of complex processors such as GPUs and smartphone SoCs, raising the cost of devices such as graphics cards and handsets. Meanwhile, the prohibitively high costs of multi-chiplet designs will make them more appealing.
According to DigiTimes (via @RetiredEngineer), one wafer processed on TSMC’s cutting-edge N3 manufacturing technology will cost more than $20,000. According to the report, a N5 wafer costs around $16,000.
Making chips on the N5 and N3 production nodes is expensive for a variety of reasons. To begin, both technologies make extensive use of extreme ultraviolet (EUV) lithography for up to 14 layers in N5 and even more in N3. Each EUV tool costs $150 million, and multiple EUV scanners must be installed in a fab, adding to TSMC’s costs. Furthermore, producing chips on N5 and N3 takes a long time, which means higher costs for TSMC.
Alleged TSMC Wafer Pricing
Except for actual customers, TSMC does not generally reveal wafer prices. It’s also worth noting that contract pricing — which larger orders from companies like Apple, AMD, Nvidia, and even competitor Intel are likely to use — may be lower than base prices. Nonetheless, here’s what the report has to say about current prices.
| Price per Wafer | $20,000 | $16,000 | $10,000 | $6,000 | $3,000 | $2,600 | $2,000 |
| Node | N3 | N5 | N7 | N10 | N28 | 40nm | 90nm |
| Year | 2022 | 2020 | 2018 | 2016 | 2014 | 2008 | 2004 |
Chip developers who use TSMC’s services are expected to pass on the costs of new chips to downstream customers, raising the prices of smartphones and graphics cards. Even now, the iPhone 14 Pro starts at $999, whereas Nvidia’s flagship GeForce RTX 4090 costs $1,599. We can expect Apple and Nvidia’s products to become even more expensive once they adopt TSMC’s N3 node.
Of course, the actual chip cost can still be relatively low when compared to the rest of the components in a modern smartphone or graphics card. Take, for example, Nvidia’s AD102, which measures 608mm2. According to dies per wafer calculators, Nvidia can get about 90 chips out of an N5 wafer, for a base cost of $178 per chip. Packaging, PCB costs, components, cooling, and so on all contribute at least twice as much, but the true cost of modern chip design is in the R&D aspects.
| Header Cell – Column 0 | N3E vs N5 | N3 vs N5 |
|---|---|---|
| Speed Improvement @ Same Power | +18% | +10% ~ 15% |
| Power Reduction @ Same Speed | -34% | -25% ~ -30% |
| Logic Density | 1.7x | 1.6x |
| HVM Start | Q2/Q3 2023 | H2 2022 |
While there are rational reasons for TSMC’s price increases, it should be noted that the company can get away with it because it currently lacks competitors who can produce chips using cutting-edge fabrication technologies with decent yields and in high volumes. While Samsung Foundry is formally ahead of TSMC with its 3GAE process technology (3nm-class, gate all around transistors), it is believed that due to insufficient yields, it is only used for tiny cryptocurrency mining chips. Meanwhile, Samsung Foundry’s 4nm-class process technology has fallen short of expectations in terms of performance.
When (and if) Samsung and Intel Foundry Services offer process technologies that outperform those of TSMC, the world’s largest foundry will be forced to limit its prices, though we do not expect chip prices to fall as fabs become more expensive, chip development costs rise, and fabrication technologies become more complex.
When Intel, GlobalFoundries, Samsung, TSMC, and UMC adopted FinFET transistors in the mid-2010s, the costs of making chips on leading edge nodes began to rise rapidly. Costs rose for everyone at the time, despite intense competition among contract semiconductor manufacturers.
Apple is expected to be TSMC’s first client to use N3, as the company can afford to develop an appropriate SoC, mass-produce it, and still make money on its hardware. Apple has not stated what kind of processors it plans to use on the N3, but a successor to the current M2 and A16 Bionic seems logical. Other chip developers may defer using TSMC’s N3 for the time being due to prohibitively high costs, and instead use chiplet-based designs due to lower development costs, risks, and production costs.
Please keep in mind that TSMC does not comment on its quotes and has not responded to reports that it will charge $20,000 per N3 wafer. These figures are provided by industry insiders and may or may not represent actual high volume prices.
