Han Dieperink: The semiconductor supercycle

Han Dieperink: The semiconductor supercycle

Artificial Intelligence Technology
Han Dieperink

This column was originally written in Dutch. This is an English translation.

By Han Dieperink, written in a personal capacity

This week, TSMC shares rose almost 10%. Its price-to-earnings ratio is only 15 now. TSMC expects revenue to rise 25% this year, based on strong demand for artificial intelligence chips.

Based on these figures, TSMC is growing at least twice as fast as the market, which is a remarkable achievement considering the size of the company. The expectations for the entire market may therefore have been underestimated. The company does indicate that it will invest approximately the same amount in capacity this year as last year: between $ 28 billion and $ 32 billion.

TSMC is a leader in the field of 2-nanometer chips, a chip that the company already produces for Apple and Nvidia. At 2-nanometers, Samsung sees itself as the major competitor, but TSMC is still beating Samsung in terms of implementation. Furthermore, Samsung customers fear that their technology or designs will leak to other Samsung departments. TSMC doesn't have that problem. The figures show that there is still sufficient growth potential in this industry.

Semiconductor industry is flourishing

Compared to other tech companies, semiconductor companies have been relatively cheap for years. This is due to the traditionally highly cyclical image. The ingredients for such cyclicity are certainly there. Huge investments need to be made. In theory, this means that product prices can drop significantly, if only because they can make up for some of the fixed costs in addition to the variable costs. Moreover, until recently there was also a clear cycle in the PC industry, mainly depending on the latest version of the Windows operating program.

However, the PC industry has become much less cyclical, because it no longer works with software licenses, but with a subscription: software-as-a-service. Much more importantly, many new markets have emerged in addition to PCs. The most important is the mobile internet. Especially in terms of energy consumption and processing speed, mobile chips have given a strong boost to the chip industry.

Likewise, the gaming industry has created a demand for increasingly better graphics chips, something that Nvidia is now explicitly reaping the benefits of because those chips form the basis for AI applications. In addition, various developments are possible thanks to the latest chips, such as cloud services, internet-of-things, 5G, Big Data, virtual and augmented reality, etc.

Fourth industrial revolution is gaining ground

These are all developments associated with the fourth industrial revolution. Technologies reinforce each other and provide numerous new applications. If there is a cycle in the chip industry, it is multiple cycles at the same time. There are also more and more applications. A modern (electric) car contains many more chips than older models. The war in Ukraine has shown that chips (for drones and missiles) can be the deciding factor.

In addition to the many cycles, there is also structurally strong underlying growth. The lowest point in the cycle is therefore often higher than the previous low and over a longer period such fluctuations are hardly visible. That has sometimes been different, partly in view of the highly cyclical profit development of Dutch companies in the supply industry such as ASML, ASM, and Besi. However, with the advent of the mobile internet (only since the arrival of the iPhone in 2007), it has been a straight line upwards.

Given the many applications and the underlying growth, it is not surprising that the chip market will grow from more than 500 billion dollars last year to a market of 1 trillion dollars in 2030.

The importance of chips for artificial intelligence

According to McKinsey, artificial intelligence chips play a major role in the growth spurt from 532 billion last year to 1 trillion dollars in 2023. With artificial intelligence, there is no diminishing return. The machine learns based on data, and more data (and more chips) is always better. This means that it is not easy to compete with companies that have already reached a large size. It also means that demand for Nvidia's chips could increase tenfold in just a few years. Even if AMD competes with Nvidia and gains 10% market share, that 10% will soon be as large as the entire market now.

With the most complicated chips, more and more software is required to make chips work optimally. Chip manufacturers are increasingly writing that software themselves and are therefore partly becoming software companies. In addition to higher margins, this software also smoothes the cycle. This shifts some of the power to the chip manufacturers. For example, if you look at what the added value of the cloud consists of, you can look at the services of Amazon, Google or Microsoft, but the real added value in the field of artificial intelligence is now provided by the hardware and software of Nvidia. The rest of cloud is basically chips that anyone can buy anywhere.

The strong underlying growth in combination with the expansion with many new cycles ensures that the original cycle levels off and even disappears. Then only structural growth remains. Yet many chip companies are (still) seen as cyclical companies with lower valuation compared to the rest of the tech sector, while they are the beating heart of new technological applications. Add to this the expansion opportunities through software integration, both in terms of margins and valuation, and the semiconductor super cycle ensures that chip companies also outperform the rest of the IT sector in terms of price.