Breakfast Bytes Blogs

I'm going to summarize the case for "yes" from McKinsey & Co, and the case for "it's a stretch" from an interesting presentation that took place in the "Chiphead Theatre" in the exhibit hall at DesignCon in which Michael Yang of Omnia presented The Next Tipping Point for Semiconductors.

McKinsey

Recently, McKinsey produced a report, The Semiconductor Decade: A Trillion Dollar Industry.

Here's one of the money paragraphs:

As the impact of digital on lives and businesses has accelerated, semiconductor markets have boomed, with sales growing by more than 20 percent to about $600 billion in 2021. McKinsey's analysis, based on a range of macroeconomic assumptions, suggests the industry’s aggregate annual growth could average from 6 to 8 percent a year up to 2030.

The result? A $1 trillion dollar industry by the end of the decade, assuming average price increases of about 2 percent a year and a return to balanced supply and demand after current volatility.

They dug into market segments and came up with something that readers of Breakfast Bytes could have written in their sleep:

The overall growth in the semiconductor market is driven by the automotive, computing and data storage, and wireless industries.

McKinsey's numbers assume that there will be continuous growth between now and 2030, and that the semiconductor industry has completely stopped being cyclical. One person who hasn't forgotten this is...

Michael Yang

Michael started by pointing out that the combination of stronger prices (especially in memory) and strong demand meant that last year the semiconductor industry had a revenue of $587B. This wasn't just a record; it was $100B more than ever previously recorded. Of the top 10 semiconductor companies, 9 had double-digit growth (only Intel at #1 did not). Moreover, 7 of those 9 had over 30% year-on-year growth. Since Michael's background is in DRAM and flash, he was pleased to see that memory companies were 3 of the top 5.

The semiconductor industry has been subject to a perfect storm, but it is worth noting that not much on this list is in the control of the semiconductor industry. We don't control natural disasters, power outages, politicians, or demand shifts such as the move to electric vehicles.

Post-Covid is a new norm. Covid-driven demand will go down, newer advanced fabs will come online. Legacy technology (that is to say, not EUV) on 12" will reduce demand on 8" fabs. There are a total of over 80 fabs that have been announced for the next few years. So the equation is changing, the equation being supply versus demand. Michael polled the audience as to who thought we were on the left of the crossover, who thought we were at the crossover, (my vote) and who thinks we are already on the right. He thinks we are already on the right but that it takes time to ripple through the whole supply chain. But from a pure manufacturing point of view, it is all on the right: microcontrollers, memory, processors, etc.

Application drivers tomorrow. Smartphones will continue to be the biggest sector (but not grow). PCs will continue to decline. Severs and automotive will grow a lot. Automotive will grow from 11.5% to 20% of the industry. Hyperscaler from 14.5% to 22% of the industry.

Some analysts think that the industry will grow to $1.2T by the end of the decade, which requires 8% per year through 2030. Michael doesn't believe that but thinks $1T is reachable.

So that was demand. It will continue to grow. What about supply? In particular, what about Capex?

What we don't control is how geopolitics wants to have a hand in this. Governments want to "help." That means more capacity around the world versus today, where we have supply roughly matched to demand. Capex has been $100B/year for the last 4 years. It will grow to $120B per year for the next 5 years, so that is $1T in 9 years.

This is a fantastic time to be a fab engineer. Go and get your degree and move to Arizona.

The big risk in all this is that we end up with a lot of overcapacity since every region wants its "own" fab, which will probably add up to more than demand.

One key question is whether the government can protect the industry during oversupply and rising inventory. There is already a lot of DRAM and flash all revved up with no socket to go to. This is the geopolitical-induced self-sufficiency capacity war among various geographic regions like US, Europe, and more. I wrote about this topic recently in my post China, US, Europe: Everybody's Got a CHIPS Act. But most semiconductor companies are public companies and cannot just do the governments' bidding and, say, build a new fab because the government wants one if there is no demand. One lesson that I learned working for VLSI for years is that you have to "fill the fab," or the financials will look horrible.

Another challenge is that a whole ecosystem (not just fabs) needs to be built to support the manufacturing. Materials, chemicals, gases, and equipment. They could all be bottlenecks.

Just because you can build a fab doesn’t mean you can buy the equipment and the gases, etc. you need.

So, the big question is when is the tipping point of too much investment and not enough demand? Michael thinks that there will be a dip in the next 12-36 months. If there is, it will probably not get back to $1T by 2030. Like when an investment portfolio drops by 50%, it has to double to get back even.

The semiconductor industry is cyclical, always has been, and always will be. Or, as Michael put it:

One thing is certain. Boom and bust will be alive and well in the semiconductor industry.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.

.