It’s time we had “the talk”.
It might be a little awkward. Your friends have probably been talking about it, but I want you hear the truth. I brought a few pictures to help me. Let me just find the right one.
Ok… here. Do you know what this is?
Today, we’re having “the talk” about semiconductors (aka microchips aka chips). And I started out thinking this would be a technical article, but really, it’s a geopolitical game with huge stakes. Not only is it a $500B industry growing at 12%, but winning the game means generations of technical, political, and military dominance. Each player has unique strengths, and right now they’re woven together in a complicated equilibrium. In a moment, we will meet those players. But first, I’ll do a quick intro to chips.
Chips 101
Lesson #1: Chips go in everything. They are in your thermostat, sound bar, and almost anything electronic you can think of. The average car has 1,200 of them. There are four key types:
Memory: stores information (ex. Which 42 tabs do I have open in Google Chrome right now?)
Microprocessors (CPUs): run any program (ex. Using Excel, then Zoom, then Outlook)
Commodity integrated circuits (CICs): run only predetermined programs (ex. Thermostat, Oven)
Graphic Processing Units (GPUs): render graphics (ex. Watching YouTube videos)
Lesson #2: Smaller is better. Why? Because smaller chips use less energy and can pack more punch. Firms have poured billions into being smaller, and they’ve made insane progress over the last 50 years:
1971: Intel’s leading processor had 2.3K transistors (the punch) and a node size of 10 microns.
2022: Apple’s M1 chip has 114B transistors and a node size of 3 nanometers (.003 microns)
Those improvements dramatically raise device performance. In 2021, IBM’s R&D team created the world’s first 2 nanometer chip. Mass production will take years, but here’s the impact they expect:
“…2 nm processors used in cell phones could quadruple the battery life of cell phones using 7 nm processor technology, such as the iPhone 11, Samsung Galaxy S10 and Google Pixel 5. Based on average use, this means the phone would only need to be charged once every four days.”
This improvement in chips is called “Moore’s Law”, credited to Intel co-founder Gordon Moore back in 1965. His observation was that the number of transistors in a leading chip doubles roughly every two years. While it has generally held true, we are near the end of what is physically possible. In 2022, scientists at Beijing’s Tsinghua University managed to create a chip where the transistor thickness is 0.3 nm, using a SINGLE SHEET OF CARBON ATOMS. I’m sure there will be some new innovation that drives performance, but how do we go smaller than atoms?!
This is a good time to introduce the standard, corny semiconductor joke: this isn’t rocket science, it’s much more complicated than that.
Lesson #3: Firms specialize. Because it’s so complicated, most companies have prioritized a single step in the process. This helps them stay near the leading edge, and it makes collaborating with other firms absolutely critical. Here are the high-level steps and examples of firms at each step:
Design: innovate new ways to lay-out chips and improve performance (Nvidia, Qualcomm)
Build:
Equipment: create the machines used to manufacture chips (ASML)
Materials: create the chemicals and silicon wafers needed (Shin-Etsu Chemical Co)
Manufacturing: use materials and equipment to make chips (TSMC)
Note: manufacturing plants are called “fabs”
Package & Test: put chips into encapsulated circuits and test them (ASE, Amkor)
Samsung and Intel are exceptions to this specialization rule. They are both “Integrated device manufacturers” (IDMs) that design, build, and package their own chips (though they still need to purchase equipment and materials from companies like ASML and Shin-Etsu).
Ok, enough with the overview, let’s meet the geopolitical players of the game.
The Players
China
Right now, China accounts for 5.2% of global manufacturing market share. Most of their chips are larger (not cutting edge). They want to become self-sufficient, and they also want to get their hands on the world’s best chips. This would allow them to lead in AI, in military supercomputers, etc.
United States
The US has ~12% of global manufacturing market share. In the 1990’s, it was closer to 40%, but they lost market share as other governments provided incentives for R&D and manufacturing while the US didn’t. The U.S. does still have a distinct advantage in chip design (45% market share), thanks to design-only (“fab-less”) firms like Nvidia. Fabulous.
The US also wants to lead in AI and military tech, and they want to slow China down.
Japan
Japan dominates the “materials” market. Two of their companies – Shin-Etsu Chemical Co and SUMCO, account for 58% of the global market share of materials!
South Korea
Samsung is headquartered there, and it’s known for producing some of the world’s best memory chips. South Korea has roughly 20% of global manufacturing market share for chips.
Netherlands
ASML is a Dutch company with a very unique role. They are the only firm in the world that makes the advanced equipment needed to manufacture leading chips. All firms that manufacture chips (TSMC, Intel, etc.) must buy equipment from ASML.
Taiwan
Taiwan is the juggernaut in the manufacturing space. Taiwan Semiconductor Manufacturing Co (TSMC) has 54% market share for all chips, and 90%+ market share for the world’s most advanced chips.
So what?
This market is completely interdependent, like a teetering Jenga tower with expensive, complicated pieces. To make the best chips, you need US designs, Japanese materials, Dutch equipment, and Taiwanese manufacturing, which is ultimately packaged and sold as an Apple chip. With US-China competition heating up for AI dominance, and just heating up in general (cough, balloons), such interdependent chip-making is a strange equilibrium.
For China, the obvious first step towards being self-sufficient for advanced chips is roughly 100 miles away, in Taiwan. If they annex Taiwan and take control of TSMC, they are the chip behemoth overnight. If the US supply for such critical tech were to be threatened, they would probably step in to defend Taiwan (even if it wasn’t China doing the threatening). But because China is the most likely supply chain threat, AND a key geopolitical rival, there’s no way they would quietly let that Taiwan takeover happen.
For Taiwan, it’s critical to keep up the balancing act, leaning into relationships with both China and the US. Here’s a map that shows the USD volume of exports from Taiwan, with the largest Taiwanese importers in dark blue:
That’s $50B in exports to the US and ~$144B to China. As we can see from the second graph, most of those exports are chips and related electronics. This balancing act for Taiwan isn’t new – it’s part of the founding story of TSMC. It was founded by Morris Chang in 1987, who got advanced engineering degrees from MIT and Stanford and worked for 25 years at Texas Instruments (TI). His final/largest role at TI was head of the global semiconductor business… After TI, he went to work for the Republic of China (ROC), which is the de facto government of Taiwan (and an opposition force to the Chinese Communist Party (CCP) which governs the mainland). The ROC put in 50% of the capital to have Chang start TSMC. All things considered, a pretty brilliant strategic move.
The US has been slow to react, but late in 2022, the government finally took three key steps:
It turned on incentives for US-based production. The CHIPS Act, passed by Congress and signed into law by President Biden in August 2022, sets aside $52B in incentives for companies to produce more chips within the borders of the US. It won’t cover anyone’s costs completely, but it’s designed to subsidize the cost of new R&D and manufacturing facilities. Firms have to apply and agree to a lengthy set of conditions, including: using union labor, limiting share buybacks, and sharing excess revenue with the US government if they beat their projections. More here from TechCrunch.
It hit China with export restrictions. In October 2022, the US introduced export restrictions that prevent US firms from selling chips, designs, or components to China. The goal is to slow down China’s ability to produce chips, especially the most advanced kind. More here from Techspot.
It asked other countries to join forces. After introducing the export controls, the US then asked the Netherlands and Japan to make a similar investment. The Dutch had already committed (thanks to pressure from President Trump in 2019) to stop selling their most advanced ASML devices to China, and has now (as of January 2023) indicated they will further restrict sales. China will only be able to get lower grade equipment from ASML. The Japanese have agreed to restrict their sales of equipment, but there have not been any specifics about whether they’ll also restrict sales of materials (their key input to the process). More here from The Verge.
As for what happens next, I don’t have a crystal ball, but I do have opinions! Shocking, I know.
Short Term:
Japan won’t restrict sales of materials to China. The raw materials used for chips are nearly identical. This means there’s no way to restrict only cutting-edge supplies to China (the way the Dutch are for equipment). Plus, 23% of Japan’s exports go to China (mostly machinery, chemicals, and electronics). Restricting these exports would significantly hurt GDP.
Taiwan won’t produce leading chips in the United States. They did announce a Phoenix, AZ factory in 2020 and have doubled their expected investment to $40B since the CHIPS Act was announced (because who doesn’t love a subsidy). But they’ll keep the manufacturing of their most advanced chips in Taiwan to maintain the protection of US-dependence (while they can).
Because TSMC won’t produce leading chips in the US, Intel and Samsung will get preferential treatment for CHIPS Act subsidies. Intel should especially benefit as the only meaningful US company in this chip manufacturing space. Among other US chip makers, Texas Instruments does analog chips (not digital; not advanced), and Micron only does memory chips. The arms race for AI/military applications is with advanced, digital CPU and GPU chips.
Long Term:
Three US cities will benefit from all the chip attention:
I’ve already mentioned Phoenix, which is adding those new TSMC plants, but Intel is there too and is already one of the largest employers in Arizona.
Austin, TX is another developing chip hub. NXP (Dutch co, lower-tech chips) already has 2.7M square feet of factory space there, and Samsung is building a $17B chip plant that will create an estimated 2K high-tech engineering jobs.
Columbus, OH is the third city, where there aren’t currently any other chip makers, but where Intel announced a $20B factory.
Hopefully this overview helps you process (what is sure to be) an onslaught of chip-related news over the next few years. But even if it doesn’t, you now have some really impressive things to talk about at dinner – fab, nanometers, Moore’s Law?! I’m impressed just thinking of imaginary you talking about chips at your imaginary dinner.