MF3d
Taiwan Semiconductor Manufacturing Company Limited (TSM) (“TSMC”) reported its Q2 2023 earnings yesterday (July 20, 2023). TSMC’s reduced revenue guidance for the year sent Nvidia Corporation (NASDAQ:NVDA) and Advanced Micro Devices, Inc. (NASDAQ:AMD) shares tumbling. However, TSMC’s commentary on the future of artificial intelligence (“AI”) chips did not receive as much attention. Once we dig into it, TSMC’s commentary on AI is positive for long-term Nvidia and AMD investors. Below I explain the reasons for this assessment.
TSMC Concurs With Nvidia And AMD’s AI Growth Forecasts
In a recent article titled “Nvidia And AMD Paint The Same Growth Picture,” I observed:
[AMD CEO Lisa] Su’s claim that the market for AI chips will cross $150 billion by 2027 aligns with Nvidia Corporation CEO Jensen Huang’s statements during Nvidia’s last earnings call about trillions of dollars of data center infrastructure being built or replaced in coming years.
It seems that TSMC broadly agrees with AMD and Nvidia’s forecasts. During the Q2 earnings call, CEO C. C. Wei stated:
Today, server AI processor demand, which we define as CPUs, GPUs and AI accelerators that are performing training and inference functions accounts for approximately 6% of TSMC’s total revenue. We forecasted this to grow at close to 50% CAGR in the next 5 years and increase to low teens percent of our revenue.
Although they are talking about different metrics, TSMC’s forecast of 50% CAGR over the next 5 years aligns well with Lisa Su’s forecast of data center AI accelerator TAM (total addressable market) growing at “over 50% compound annual growth rate –to over $150 billion in 2027.”
The commentary from TSMC further strengthens the bull case for AMD and Nvidia. The sort of growth being talked about here would go a long way toward justifying their high valuations (especially Nvidia’s)–and leave room for upside beyond that.
AI Chips Will Remain In Tight Supply
Here is what Wei had to say about the available capacity for AI chips:
For the AI, right now, we see a very strong demand, yes. For the front-end part, we don’t have any problem to support. But for the back end, the advanced packaging side, especially for the CoWoS [Chip on Wafer on Substrate], we do have some very tight capacity to – very hard to fulfill 100% of what customer needed. So we are working with customers for the short term to help them to fulfill the demand, but we are increasing our capacity as quickly as possible. And we expect these tightening will be released in next year, probably towards the end of next year. But in between, we’re still working closely with our customers to support their growth.
He further noted that the increase in CoWoS capacity would be roughly around 2x.
Moreover, there is some chance that TSMC could still be short on CoWoS capacity beyond 2024, with Wei stating:
We are planning our CoWoS capacity, although probably still not enough, but we’re working very hard to increase it. Overcapacity, today is not a concern, today’s concern is not enough capacity to support all the very strong demand.
As readers can see, TSMC forecasts that some of the advanced packaging (CoWoS) involved with AI chips is likely to be in tight supply until the end of 2024. Hence, demand is likely to continue to outstrip supply for quite some time because AMD and Nvidia will not be able to expand production past a certain point even though some N4 and N5 capacity is available.
While Nvidia and AMD may not be able to sell as many chips as they would like, they may both be able to sell as many as they can produce for several quarters. This is already true for Nvidia, and it could turn out to be true for AMD’s upcoming MI300 chips as well–simply because Nvidia will not be able to expand supply past a certain point. AMD’s chips could therefore do very well by virtue of mere availability.
Limited supply should also mean that the extremely high margins being charged for AI chips are likely to continue for multiple quarters.
Overall, while Nvidia and AMD investors may be disappointed by the capacity constraints, they should also be pleased about TSMC’s confirmation of so much demand for AI chips that it has overwhelmed its advanced packaging capacity for at least a few quarters.
The World Of Computing Is Evolving
Finally, the revelation that AI chips currently account for 6% of TSMC’s total revenue is quite useful for gaining perspective about the speed and scale of the AI revolution.
6% of TSMC’s revenues in Q2 amounted to about a billion dollars. The vast majority of this should be attributable to Nvidia. Nvidia had guided a sequential increase of a little under $4 billion in AI chip sales this quarter, and was selling some AI chips previously as well. Given the exorbitant margins on these AI chips, it makes sense that Nvidia is paying around $1 billion to produce them.
Now, although 6% of TSMC’s revenues may not seem like much, it is actually quite a lot. First, we have to note that advanced nodes (N7 and below) represent only about half of TSMC’s revenues. This means that AI chips have now already become about 12%–one-eighth–of all of TSMC’s advanced node revenues (which include both smartphones and high-performance computing). And this has happened despite capacity constraints and, of course, still very early in the AI revolution.
At least to my eyes, the fact that one-eighth of TSMC’s advanced node capacity is now deployed toward AI chips is a strong signal that the world of computing is evolving very rapidly toward AI–a strong bullish indicator for Nvidia and AMD.
Conclusion
While investors may be a bit disappointed about the capacity constraints that are likely limit growth for Nvidia and AMD in the very short term, Taiwan Semiconductor has painted a very positive picture concerning the future of AI. Long-term investors should ultimately see TSMC’s updates regarding AI as another credible data point signaling an extremely strong future outlook for Nvidia and AMD.
