Accelerating Semiconductor Supercycle

The increasing use of artificial intelligence is accelerating the semiconductor supercycle. It is now fifteen years since there was a real downturn in the chip industry. Since then, chip companies' share prices have increased fifteenfold. Last week, the semiconductor index even rose 11 percent in one day, a feat unprecedented this century. All thanks to Nvidia, which on the same day posted its biggest ever share price gain thanks to stunningly strong results. I previously wrote about the semiconductor supercycle in April 2019 and April 2021. Even earlier this year, when there were doubts related to the approaching recession and the increasing technological battle between China and the United States, there was brief speculation about whether the end of the cycle would now be in sight. But nothing of the sort. Rather, there seems to be an acceleration.

That semiconductor companies are seen as cyclical is not surprising, given the past. Investment to produce chips is high, meaning that chip companies are theoretically willing to produce at a loss for a long time as long as they can make up at least marginal and some fixed costs. But the industry has changed significantly on both the supply and demand sides. On the supply side, there are now only a few foundries, of which TSMC is the most important. That creates more capital discipline. On the demand side, there is not only the PC cycle or the cell phone cycle but there are numerous new applications (cloud, big data, internet of things, artificial intelligence etc) that require a lot of computing power. Of course, the last 15 years was not a straight lineup for the semi-conductor index, but those who exited in the interim were soon condemned to re-enter at higher levels. For a long time, the appeal of the semiconductor sector was that the companies were valued lower relative to other tech companies because of their perceived cyclical sensitivity, but actually deserved a premium because without them there would be no IT sector at all. Moreover, not only is the speed of chips growing exponentially, but apparently so are the many applications.

Artificial intelligence requires a lot of calculations. This is faster on a Graphics Processor Unit (GPU) than on a CPU. This is because parallel calculations can be made on a GPU. In fact, a GPU consists of hundreds to thousands of smaller cores. In essence, it is a bit like a quantum computer that also does not make serial calculations, but does everything at once. For example, think of a large cabinet with all the drawers. A normal computer goes down drawer by drawer, and a GPU (or a quantum computer) opens and closes them all at once. Fast calculations are essential in gaming. A fast video card makes for a nicer and smoother display; with a faster card, you win. In the mid-1990s, Nvidia was exclusively a manufacturer of graphics cards primarily for gamers. Because of that speed, the GPUs that sit on a graphics card are also used to build supercomputers relatively cheaply. Graphics cards became faster and faster and also more expensive. Because of the fast computing power, Nvidia also attracted interest from artificial intelligence developers about a decade ago. Last year the company came out with the H100 (the Hopper), the most powerful processor ever built, with a price tag of $40,000 each, according to Nvidia the first computer chip made specifically for generative AI such as ChatGPT. Thanks to ChatGPT, everyone "suddenly" sees the potential of this technology and, as a result, many companies are investing heavily. Nvidia simply has the right product at the right time. This quarter, therefore, sales are coming in 50 percent higher than expected. In particular, companies with cloud facilities (Microsoft, Amazon, Google, Meta) are buying the H100 in abundance. TSMC is already freeing up more production capacity to produce more H100s. The waiting time is now about six months for some customers. Obviously, things are moving faster for Big Tech. By the way, Nvidia employs more software specialists than people working on the hardware. After all, making the best use of these processors also requires complicated software. Altogether, Nvidia is about two years ahead of the competition. In the IT sector, that's an eternity, but certainly, a lead that even Nvidia can quickly lose if the wrong choices are made. For now, the company is making the right choices. As a result, its stock is up 172 percent this year.

Last year, the US came up with the Chips and Science Act, effectively a subsidy of more than $50 billion for the US chip sector, a counterpart to China's "Made in China 2025. Europe, too, is eager for more technology in this field to be developed locally. As a result, this market is expected to grow at an average annual rate of 13 percent this decade. Given the opportunities and savings that can be achieved with generative artificial intelligence, this growth forecast may be a bit low, because, at such a growth rate, we are talking about a market size in 2029 of $1.4 trillion, which is then out of a world GDP of than $140 trillion. For now, the semiconductor super-cycle is not over.