In this week’s episode of 16 Minutes, our show where we talk about tech trends in the news, what’s hype/ what’s real, and where we are on the long arc of innovation, the topic is semiconductors – specifically, the ongoing global shortage that began last summer and has intensified in recent weeks. So much so, that the U.S. president signed an executive order just last week to address concerns around the shortage, calling for reviews of supply chains for critical sectors of the economy.
Our expert is a16z Operating Partner Frank Chen, who led our research arm and has also joined past episodes about semiconductors on this show including one with Steven Sinofsky and Sonal in which they analyzed the ARM and Nvidia news.
Frank joins a16z’s Zoran Basich to cover the bigger picture of the chip shortage including geopolitics, the pandemic, and complex worldwide supply-chain dynamics — all in almost exactly 16 minutes!
How the semiconductor supply chain works [1:14], what caused the current shortage [3:10], and errors in forecasting that some companies made [4:49]
The impact of demand (types of chips needed by various industries) [5:54] and supply (cost of building factories, COVID disruptions, and the scarcity of raw materials) [8:24]
Geopolitical questions involved [11:55], including competition with China [13:58]
Thoughts about how the shortage will play out [15:14]
Zoran: Welcome to this week’s episode of “16 Minutes,” our show where we talk about tech trends in the news, what’s hype, what’s real, and where we are in the long arc of innovation. The topic today is semiconductors, a topic that has been in the news a lot in different forms. Specifically, we discuss the ongoing global shortage that began last summer and has intensified in recent weeks. So much so that the U.S. president signed an executive order to address concerns around the shortage, calling for reviews of supply chains for critical sectors of the economy.
Joining us as our expert is a16z Operating Partner Frank Chen, who led our research arm, and has also joined past episodes about semiconductors on the show, including one with Steven Sinofsky and Sonal, in which they analyzed the Arm and Nvidia news.
In this episode, we’ll cover the bigger picture of the chip shortage, including geopolitics, the pandemic, and several other factors, all in almost exactly 16 minutes. But first, Frank covers the immediate impact on the automotive industry, which was highlighted in the news of the executive order.
Frank: A bunch of auto manufacturers, this includes GM, and Ford, and what used to be called Fiat Chrysler, now called Stellantis, all of these companies have basically shut down or slowed down their production of cars because they can’t get enough chips. There are very calibrated supply chains, tens of thousands of suppliers. And right now, what’s happening is, you know, Ford’s F-150, the best selling truck, is held hostage to a $0.25 cent semiconductor that all of a sudden they can’t get enough of.
Zoran: So this news is really closely tied to the supply chain. That’s where the White House is focusing. Help us understand the landscape here, Frank. Who are all the players involved up and down the chain and how are they interrelated?
Frank: The way that chips flow is that you have semiconductor vendors like Nvidia, and NXP, and Renesis, and Panasonic, Toshiba, Samsung, TI, and so on, they make the chips. Some of these companies make their own chips like Samsung and Toshiba, so they’re called integrated device manufacturers. Other companies like Qualcomm and Nvidia design the chips, but then use other companies like TSMC, or UMC, or Samsung, to actually make the chips. These design-only companies are called fabless chip companies. But in either case, you take the chips, you sell them to companies called tier one electronic systems providers. So these companies have names like Bosch, and Delphi, and Harman, and Denso, and Siemens, Continental. And then these companies in turn sell to what’s called the OEMs, the car manufacturers. That’s what we’d recognize as a car manufacturer. So a GM, a Ford, a Tesla, an Audi, a BMW, so on and so forth. Right? So that’s sort of the supply chain.
You got a chip company selling to an electronic systems company. They might make the onboard entertainment system, they might make the anti-lock brake computer, or they might make the adaptive cruise control system. And then the OEMs assemble them into cars. And so, there is a lot of complexity.
Zoran: So how did the automakers get into this mess? What are some of the factors that caused this chip shortage?
Frank: Let’s start with the demand side. So, three things. First, software is eating cars, which is to say that the percentage of a car that is electronics has been steadily increasing. So it used to be a car was rubber, plus glass, plus steel, and that was pretty much it. But now, in 2020, electronics are about 40% of the cost of a car. For a point of reference, in 2000, it was probably 18%. And, you know, these days, as you think about all of the cool safety features, right? Somebody’s in your blind zone, or adaptive cruise control, or automated stopping if there’s a pedestrian in front of you, right? All of those are obviously electronics.
The second thing is, this was a classic case of under-forecasting. So what happened was, COVID hit, and everybody battened down the hatches. Everybody was like, oh, car demand is going to drop off the cliff here. And so we better be conservative and cut back all our orders. And what happened roughly at the same time was, there was a set of things that people wanted that demand went the other way, right? So think of Chromebooks, and laptops, and webcams, and everything that makes Zoom possible. All of this work at home stuff created a lot of demand for TVs, and computers, and so on. And so, the demand for semiconductors to feed those things sort of leapt into that vacuum that the car guys left. And then once we figured out that car sales weren’t going to be as dramatically impacted by COVID, they went back to the suppliers, and then they discovered the supply is not available. In other words, they got spoken for by the computers, and the cell phones, and the webcams, and IoT devices.
Zoran: So there have been lots of headlines about this chip shortage, lots of angles to unpack, but let’s home in on the pandemic for a minute. How and why did this under-forecasting happen? And why specifically is that a problem when it comes to chips?
Frank: I think it was human nature. Which is, COVID happened and you were like, gee, when will demand really ever come back to normal? And I don’t want to be the guy that over-ordered everything, right? Like, once the TPS reports come out, I’m a complete outlier because what kind of idiot would I be, if COVID happened and I was the guy who over-ordered everything by two orders of magnitude. The other thing is, it’s very long lead time to spin up a new semiconductor, in other words, to change the line so that it’s making your chip versus somebody else’s chip. This is measured in tens to dozens of weeks, right? So, 30 weeks. And so, you can’t turn it off and turn it on, on a dime, right? And so, when the orders got cancelled, the lines got retooled to build other chips. And so, to turn the line back on to build your chip, we could be talking the better part of a year, which is why most people think that the shortage is going to be another couple quarters before we’re done with it.
Zoran: Okay, so on the demand side, you’ve mentioned the rise of electronics and cars. We talked about the forecasting mistakes. Let’s get a bit more into what’s hype and what’s real here. What else do we need to know to make sense of this?
Frank: On the scale of things, car manufacturers aren’t the biggest customers for chips. So, by far, the biggest customer for chips is cell phone makers. Like, when Apple places an order, that’s an order, right? Like, that’s tens of millions, hundreds of millions, billions of components. And so, like, they’re more reliable as a customer. You can sort of see why if you are a chip manufacturer, like a Panasonic, or Toshiba, or Infineon, or STMicro, or TI — when Apple, or Samsung, or Dell, or HP comes to you and says, “I need more chips.” You say, “Of course. When do you need them?” Right? And then Ford, and Jaguar, and, you know, all of the other guys have to sort of kind of wait in line.
And then the other thing that is true about car chips is, they can be harder to make, which is, their temperature range, their operating lifetime, the failure rates of these chips, right? So it’s like one thing for your Chromebook’s light sensor to go bad, it’s another thing altogether if, like, the radar that powers your adaptive cruise control goes bad. Like, that thing can’t go bad, right? Because you’re going to crash into somebody.
Zoran: Yeah. And some of the chips that are used in cars are older generation chips. So, because they’re more expensive on the consumer electronics side, that’s another reason that they’re seen as more higher priority customers.
Frank: So, yeah, two classes of chips, the older stuff, right, for the anti-lock brakes, and then the newer stuff for what the car industry roughly calls ADAS, right? So, the advanced driver assistance systems. One way to measure this is, what’s the nanometer process technology used to create it? And we’re headed towards seven, six, five nanometers. Think of that as sort of how fine is the etching on a semiconductor that defines the circuit path, right? And car manufacturers — some of their chips, you don’t need [a] 7, or 6, or 5 nanometer, you’re fine at 180. But the second class of chips that the car companies are increasingly buying are basically the exact same chips that go into a smartphone core, right? So if you think about machine learning, as we head towards autonomous, you know, they will be the most advanced chips, with a ton of transistors on them to do linear algebra, because that’s what machine learning demands. And so, this sort of emerging class of chips are exactly the same set of chips.
Zoran: Okay, so that’s the demand side. When the pandemic hit, automakers pulled back on their orders, while at the same time, demand was rising for consumer electronics. And the chip manufacturers turned to that segment of their customer base, and started producing for them. And it’s very hard to stop on a dime, and then take the automakers’ calls, who now suddenly are calling you and saying, “Hey, we want to restart our production.” That all takes a long time. So, let’s talk about supply. What happened on that side of the equation?
Frank: So it turns out that the world has a fixed amount of manufacturing capability for semiconductors, which kind of seems bizarre, because we all know that software is eating the world. And, you know, the world has nearly infinite demand for chips. Most semiconductors are made outside of the United States, despite the fact that the United States dominates revenues for semiconductor design, through companies like Intel. But most of the fabs, which is the factories that make semiconductors, are overseas. But it turns out these fabs are incredibly expensive and relatively low margin to build. Think of a factory that might cost $10 billion to build, and it’s obsolete in five years. The rate at which semiconductor fabrication changes is so fast that, like, all of the equipment that you just bought from Applied Materials and JLA-Tencor, like, that thing is going to be obsolete in five years.
And so you have these incredibly expensive factories that depreciate very, very fast. The equipment in it sort of needs to get replenished very, very quickly. And so, the industry’s kind of rewarded companies that are called fabless designers. In other words, they’re companies that design chips, but don’t actually make them. So the fabless designers that we all know are companies like Qualcomm, and Broadcom, AMD, Nvidia, Apple itself, right — all of the, like, super awesome chips that they design, they don’t actually make. They go to semiconductor fabs, the largest of which is the Taiwan Semiconductor Company, that actually operates the factories. The top three countries that actually make chips, South Korea, Taiwan, and Japan, they have all of the factories because fabless is rewarded by the investing community. You’d much rather have the higher profit margins of a Qualcomm than the lower profit margins of a TSMC, the Taiwan Semiconductor Company.
Zoran: Okay, so you have this issue of how incredibly expensive it is and how frequently you have to accommodate the new technology they have to build. What are some of the other supply factors?
Frank: Well, like every other manufacturing facility, the semiconductor industry got hit by COVID itself. People need to be in the factories feeding the wafers, and doing quality control. And so there was some slowdown as a result of that. Now, the good news is, South Korea and Taiwan are the biggest manufacturers, and they were much less exposed to COVID because they were very aggressive with their lockdowns. But there are other things about semiconductor manufacturing. So, Taiwan is going through a drought, just like California went through a drought. And it turns out, you need a lot of water to make chips. TSMC’s daily water consumption is 156,000 tons a day. In the northern part of Taiwan, where these factories sit, that amount of water is 10% of the region’s daily supply of water.
And so, you have all of these weird supply things. You have, like, raw material shortages. We’ve got the COVID hit. We’ve got the very, very small number of fabs, just because they’re expensive, and sort of the financial community willingness to fund these super expensive factories is sort of low.
Zoran: So we’ve talked about the factors like drought, the economics of chip production, the pandemic, obviously. Now, where do the geopolitics come in? How big a deal is that, really?
Frank: Yeah. So, geopolitics is a big deal. So, you’ll remember that the Trump administration basically forbid American companies from buying from Huawei. And then they later extended it to, you know, requiring a license to sell to Huawei. And then there are licenses that the semiconductor tool chain now has to apply for in order to sell to Chinese companies. And so, if there were no politics, you would do what manufacturing has done for the last four decades, which is, you’d fire up China. You’d make it possible for Chinese fabs, not Taiwanese ones, not South Korean ones, not Japanese ones, to bloom. But the problem there is that the United States, rightfully so, wants to be a little careful about what kind of semiconductor manufacturing equipment they will sell. Because the worry is that, they’ll buy it, they’ll reverse engineer it, they’ll infringe on the intellectual property. And lo and behold, they can make their own semiconductors. And by the way, that is China’s explicit goal, which is that they want to have the number one semiconductor design and manufacturing industry in the world by 2030.
And so, now the whole thing is a geopolitical dance. Like, you could imagine any administration, this is not a red or a blue issue, saying, like, we’re not sure that we want to sell China the equipment to make the most advanced semiconductors.
Zoran: It’s interesting because, you know, there’s this perfect storm that happened with the pandemic, and also some other factors, which we haven’t mentioned. You know, there were a couple of fires in Japanese factories that had a negative impact on supply. Even in Texas, there were a couple of factories that because of the recent cold spell, had to shut down for some time. So all these things kind of just built upon each other to create this somewhat perfect storm. And so that created this instant problem, but it also highlighted this larger problem [that] needs to be addressed. So what’s the bigger picture about what this means for innovation, given these geopolitical pressures?
Frank: Yeah. I mean, one of the big things that the industry is asking itself is, you know, China’s ambition to be number one, can they get there? And they don’t want to stop at just the chips, right? They want to be the complete, fully vertically integrated stack. So if you think of the vertically integrated stack, it’s sort of iOS running on top of a bunch of chips, with iOS. Or if you think about Windows, running on Intel. Like, in the next 10 years, will there be a Chinese operating system running on a Chinese chip, right, with all of the motherboard, etc., etc., design done by China. And if that were the case, will we really have two competing world ecosystems? It kind of reminds me of, like, the early days of communication. There was Docomo in Japan, and then there was Minitel in France. And like, they didn’t talk to each other. They were just little islands. And we knew that that world wasn’t good. What the world wanted was the internet, the connection of networks.
But if we go back to this sort of geopolitically motivated desire to have your country own the factors of production in a completely integrated vertical stack, hardware and software, then we might go back to the bad old days, where compatibility was hard. And, you know, we sort of were kind of wasting “R&D” dollars building the exact same thing, just in slightly different ways.
Zoran: So we have the White House, President Biden, calling for this review. Short term, what’s going to happen here? And how long will this last?
Frank: So most people are forecasting a couple more months, maybe quarters of pain. I don’t think, unless something surprising happens, I don’t think this is going to last that long.
Zoran: I thought the solution was more manufacturing facilities need to be built in order for this truly to become solved. So how is this going to solve in the short term?
Frank: You know, I don’t think it’s going to take, like, a brand new fab to unlock the current snarl that we’re in. So in the short term, you know, look, we’re not going to be at post-COVID highs on webcam orders forever, right? Like, they will go back to normal. And so the heat on alternative demand will sort of cool some. And then, you know, we’ve got the automaker chips in the queue now, right? And so, like, eventually, things will sort out. We’ve always had component shocks in the tech ecosystem. It’s just every now and then, we’ll hit a bad one. This one’s a pretty bad one because we have so many car factories making very, very expensive products, stymied by their $1 semiconductor being missing.
Zoran: And that brings us right back to the news, full circle. So, let’s go to our bottom line, Frank. What are your takeaways and final thoughts on the topic?
Frank: We have some soul searching to do about what the shape of our supply chains ought to look like and how much they should reflect the geopolitics of the time, or if technology wants to and should be a country-independent thing. Where, you know, the best ideas, meritocracy, take the day, as opposed to, we’re going to have the U.S.-led tech stack, and then a Chinese-led tech stack, and then they don’t really speak with each other, and view each other with mutual suspicion. The supply chain for technology has always had shocks. Because software is eating everything, these shocks are now rippling beyond technology.
Zoran: Frank, thanks so much for being with us today.
Frank: All right. Thanks, Zoran.
16 minutes is a short news podcast covering the top headlines of the week, separating what’s real from what’s hype.