In recent weeks, we've heard that TSMC's N2 process will start raising prices next year. We've been thinking about the implications ever since, and we think they're even more important given Intel's growth. If TSMC raises prices like we've heard it does, many companies will have no choice but to abandon the Moore's Law curve. Maybe having an alternative like Intel isn't such a bad idea.
Simply put, in the absence of effective competition, TSMC will transform from an "effective" leading monopoly to a true monopoly. This allows them to raise prices as much as they want. Doing the math, we will soon find that many companies designing cutting-edge chips today will have to exit the Moore's Law curve because it is no longer economically feasible.
Of course, TSMC will not raise prices to infinity, nor will they cut off all demand, but they will use pricing to maximize their own value. This could lead to a significant reduction in the number of customers capable of designing leading-edge chips.
Let's give an example. Imagine a big customer for TSMC - not top three, but maybe top ten. Today, they might pay TSMC $20,000 per wafer, with low-volume customers paying closer to $25,000. Assume that this company's chip area is 170 square millimeters. Using the convenient Semi-AnalysisDieYieldCalculator, 325 chips can be produced per wafer, which is $61 per chip. If the company prices the chips at $140, the gross margin would be 55%, which is good but not great.
Now assume that TSMC raises the price of the next process to $40,000. While estimates of N2 process density improvements are still in progress, we assume a 15% increase in die per wafer (375KGD). However, the cost per chip will jump to $107. This illustrates the core of the slowdown in Moore's Law - that increases in density now lag significantly behind increases in price. If the design company cannot pass on the cost increase to customers and still stays at the price of US$140, the gross profit margin will drop to 22%, which is unsustainable.
We can argue about the extent to which chip design companies can pass these costs on to customers, but the conclusion is the same: As TSMC raises prices, more and more customers are unable to produce chips at a leading level.
The examples above are loosely based on Qualcomm, so they fall into this category as well, but so does AMD. The hardest hit will be customers with smaller volumes, from startups to hyperscalers. For many, Moore's Law has become extremely challenging. Of course, NVIDIA and a few other companies have more flexibility in absorbing these costs, but many, if not most, companies don't have that flexibility.
We expect TSMC to be unlikely to push customers that hard, but the reality is that they could.
Some may say that TSMC has had a virtual monopoly for several years and could have used this method to raise prices long ago. The fact that they did not do so means that they will not do so in the future. However, things are changing.
Until recently, a cautious and paranoid TSMC had to worry about Intel or Samsung becoming competitive again. Now it seems that this possibility is getting smaller and smaller. This is why Intel foundries matter. Today, some might argue that Intel foundries are not commercially necessary in the industry -- customers don't need a second supplier besides TSMC.
However, looking to the world a few years from now, TSMC can raise prices at will. In this case, everyone will desperately look for alternatives.
Author Jonathan Goldberg is the founder of multi-functional consulting firm D2DAdvisory, which develops development strategies and alliances for companies in the mobile, web, gaming and software industries.