America’s overwhelming reliance on microchip manufacturing in southeast Asia is an increasing economic and security weakness. China’s flybys into Taiwanese airspace in recent days are just the latest reminder. If the U.S. doesn’t ramp up domestic capacity, a China-Taiwan scuffle could block our most important technology companies from access to their most important components.
A provision of the latest National Defense Authorization Act, passed in December, could be one facet of a solution. It hopes to boost domestic production with direct support for leading-edge chip manufacturing. But if it is to gain broad support from across the political spectrum, including skeptics of industry subsidies (like me), we need to understand how microchips are the foundation of our economy.
For the last 60 years, the explosive advance of information technology, embodied in the microchip, has been the driving force of the American economy. Over the last three decades, however, the process of globalization has radically altered the structure of the industry. Today, the U.S. writes code and designs chips, while the bulk of manufacturing has shifted to Asia.
The big question is whether the U.S. can sustain or even accelerate tech momentum with software alone. Likewise, can we maintain supply-chain resilience and information dominance on the battlefield and in cyberspace? In other words, can we push innovation as fast as we’d like by ignoring atoms and focusing entirely on bits?
Pioneering microchip firms, such as Intel or Texas Instruments, were integrated; they designed and manufactured their own chips in semiconductor fabrication plants, or fabs, which they owned. Beginning in the mid-1980s, another model emerged. Fabs known as foundries would focus solely on manufacturing, while a far larger number of “fabless” tech firms designed chips for an exploding diversity of applications.
This division of labor generates enormous economic wealth on both sides of the Pacific. But has the U.S. overshot the mark?
The U.S. is home to just 10 percent of the world’s foundry capacity.
American apps, web services, and designs are wildly profitable. Apple builds iPhones in China with chips manufactured on Taiwan. Likewise, Taiwan manufactures Qualcomm’s 5G mobile wireless chips and Nvidia’s graphics processors, which are crucial for artificial intelligence. Amazon, Google, and dozens of other American firms similarly design chips which are manufactured overwhelmingly in southeast Asia. Even Tesla now designs one of its own high-end chips. In all, the market value of these American fabless semiconductor firms totals more than $7 trillion.
While the U.S. is home to tech firms who design 65 percent of the globe’s fabless chip volume, it boasts just 10 percent of total foundry capacity to manufacture those chips. Only five firms worldwide now manufacture chips at the leading edge of Moore’s law, down from 30 such firms in 2001.
This industry consolidation and lack of U.S. foundry capacity is thrown into stark relief by China’s recent stall, and even reversal, of 30 years of economic and political liberalization. Taiwan Semiconductor Manufacturing Corporation (TSMC) enjoys more than 50 percent of the world’s foundry share, while South Korea’s Samsung is fast approaching 20 percent. Even America’s 10 percent share is mostly not at the leading edge. Meanwhile, China is expected to add more new capacity than anyone over the next decade, bringing its world share to 24 percent.
Southeast Asian tensions thus potentially imperil the supply chains of those $7 trillion worth of U.S. fabless chip companies. The indirect effects are far larger. As the information density of the economy grows, every American firm and citizen relies on microchips, which power the Internet, which increasingly powers everything else. Our military and intelligence capabilities are ever more reliant on information and automation.
Bolstering U.S. semiconductor manufacturing, however, is more than risk mitigation. Hardware and software innovation are symbiotic. They learn from and propel each other. Manufacturing the most complex devices imaginable — 50 billion transistors on a tiny thumbnail of polished sand! — generates deep knowledge across a range of engineering disciplines and even quantum physics itself. The R&D and supply-chain synergies and spin-offs of such massive manufacturing operations provide additional ripples of economic activity.
So why not just build new American fabs? Each new fab costs between $10 and $12 billion, stretching to $25 billion to retool and operate over 10 years. And it costs around 30 percent more to build a fab in the U.S. compared to Taiwan, or 50 percent more than China. Boston Consulting Group recently found that “as much as 40 to 70 percent of that cost differential is directly attributable to government incentives.” That’s a big reason the U.S. does software while Asia does manufacturing.
Intel, the largest U.S. chip manufacturer, says it may, for the first time, enter the foundry business and build chips for the whole universe of fabless firms. Even TSMC says it wants to build a new fab in Arizona.
The U.S. government is encouraging these moves, and the 2020 NDAA authorizes support for domestic manufacturing.
I’m skeptical of subsidies for most industries, and Washington should studiously avoid micromanaging the chip industry. But compared to nearly all the government’s other activities, it’s difficult to think of a simple spending measure better focused on a strategic objective and which could generate a comparable return on investment.
Bret Swanson is president of the technology research firm Entropy Economics LLC and a visiting fellow at the American Enterprise Institute.