Where the U.S.’s Chip Strategy Is Still Falling Short

The United States is investing billions of dollars to bring the most advanced semiconductor manufacturing onshore. New factories are rising across the United States, and the new Taiwan Semiconductor Manufacturing Company (TSMC) chip fabrication facility (fab) in Arizona will make some of the most advanced chips in the world. Despite this progress, the United States still faces a major dependency on other countries for a key part of the value chain: the back-end factories that perform all the steps needed to turn processed silicon wafers into packaged chips. Wafer testing, cutting the wafers into individual chips (known as “singulation”), packaging, and testing finished chips moved offshore in the 1970s because of its high labor intensity. Even companies like Intel that have made chips in the United States for a long time send most of them to Malaysia, Vietnam, or China for packaging. The CHIPS & Science Act of 2022 offered a broad range of subsidies for companies—including Intel, TSMC, Samsung Semiconductor, Micron Technologies, and Texas Instruments—to build new fabs in the United States. While it also provided subsidies for some companies to build packaging plants—such as one that SK Hynix will build in Indiana and another that Amkor Technologies is constructing in Arizona—they have been the exception. There is nothing close to the level of activity needed to create the full semiconductor ecosystem in the United States. In recent interviews, managers at eight major companies across the semiconductor supply chain—including two of the world’s largest chip-packaging companies, chip designers, a foundry, and electronics manufacturers—told us that the problem is primarily economics. (All asked that they not to be identified.) Their comments offer some valuable insights into a lower margin and more labor-intensive part of the ecosystem that is a necessary complement to the more glamorous wafer fab business. From those interviews, the work of one of us (Willy) on the U.S. Commerce Department’s Industrial Advisory Committee during the Biden administration, and ongoing research by one of us (Willy), we have identified three actions—shifts in how buyers price risk, changes in regional trade rules, and the creation of demand-side incentives—that could help address this deficiency. Why the Back End Remains Largely in Asia The front end of semiconductor manufacturing is where circuit patterns are etched onto silicon wafers. This is the well-known part, and it is where most U.S. policy attention and the big dollars have been focused. But a processed wafer is not a finished product. It still needs to be tested, sliced into individual chips, and packaged into the modules that are then assembled into phones, servers, and a wide range of electronic products. This work takes place in the back end. Almost all that work is still done in China, Taiwan, Malaysia, and Vietnam. Even chips made at TSMC’s new Fab 21 in Phoenix go back to Asia for packaging, although the company is building two advanced packaging facilities in Arizona (whose capacity Nvidia has mostly booked). Why aren’t more back-end plants being built in the United States? Our interviews found that for labor-intensive packaging work, U.S. manufacturing costs are roughly double those in Southeast Asia. Construction costs are even steeper. An executive at one back-end company that had constructed a U.S. plant told us that its initial projected cost of building it was two to three times the Taiwan baseline. The actual figure was five to six times. He said building a factory takes one year in Taiwan and three years in the United States. That is consistent with reports that TSMC’s Fab 21 construction costs were four to five times greater than in Taiwan. For this back-end company, construction labor represents approximately 60% of U.S. building cost versus 20% in Vietnam. Because labor is so expensive in the United States, it makes more sense to design buildings that require fewer workers to construct—for example, by using prefabricated components instead of on-site welding. But most arriving firms simply import their labor-intensive construction methods from Asia, driving costs even higher. That raises a basic question for them: Is each product valuable enough—with a large enough gross margin—to justify making it in an expensive American factory? Another issue is cluster density: how close together the players in the industry are. The Asian semiconductor cluster benefits from a high level of density. In Taiwan, a wafer moves from fabrication to test to assembly to shipment within a single compact geography. Replicating that in the United States would be hard. It would require substrate suppliers, chemical providers, chip testers, and packaging houses co-located with the fabs, none of which exists at scale locally today. When chip-packaging companies diversify their locations, they are heading to Southeast Asia, not to the United States. According to the people we interviewed, Malaysian labor rates are roughly one seventh of those in the United States; Vietnam’s are roughly 30% lower still. In one case, a major packaging company turned down a U.S. expansion even when the government offered free land and the end customer offered to fund all capital expenditures. Shared languages, familiar business practices, and overlapping time zones were cited repeatedly as decisive factors that U.S. locations simply cannot match. Even companies that have invested in U.S. back-end capacity told us that domestic production will represent only 10% to 20% of their total output. Another obstacle to luring the back end of the ecosystem to the United States is the currently inadequate “demand signal”—indicators of customer demand—for finished semiconductors that are made entirely in the United States. Chip buyers tell their suppliers they want sourcing plans that are “NCNT,” which stands for “No China No Taiwan.” But when they see what some alternatives cost and how long they take, most quickly back down. For the most advanced AI chips, the gap is even wider. The advanced packaging process these chips use, called chip-on-wafer-on-substrate (CoWoS), is today almost entirely done in Taiwan. TSMC’s Arizona facilities will not produce at volume for another two years. And even when they do, they will represent only a fraction of what is needed. Until then, a wafer fabricated in the United States must fly across the Pacific for packaging before it becomes a finished product. As one executive at a major fabless chip designer put it, if the most critical component of your product still originates in Taiwan, what is the real value of spending heavily to move one part of the process elsewhere? What is more, most of the factories that assemble printed circuit boards (the boards onto which chips are ultimately installed) are still in Asia. This is true even for the electronics and servers that go into AI data centers. Even if companies no longer want to do final assembly in China, Taiwan and Southeast Asia remain the preferred locations. If the boards are assembled in Asia, the packaging has reason to stay there, too. The talent pipeline compounds the challenge. Every company we spoke with identified the availability of a qualified workforce as a bottleneck. One company told us that hiring packaging engineers in Arizona is so difficult that the company flies in expatriates from South Korea. Another noted that even when you can hire local operators, the skilled technicians from Taiwan who actually know how to run the equipment still need to be temporarily assigned to the United States, adding cost, complexity, and months of delay to getting factories operating at full operating capacity. Multiple firms, including foundries and equipment suppliers, compete for the same small pool of workers in the Phoenix area. Local community colleges in Arizona are struggling to produce enough graduates to meet demand. This gap between talent supply and demand will likely take years to close. Three Ways to Finish the Job We have three ideas that we think will accelerate the creation of the full semiconductor-production ecosystem in the United States. Change chip buyers’ assumptions about risk. The past five years of tariff shocks, pandemic shortages, and geopolitical crises have made one lesson clear: disruption risk is not free. Yet most companies still do not price it into their sourcing decisions. CFOs and procurement leaders need to start including the cost of potential disruptions when they compare prices between domestic and offshore suppliers. Paying a premium for domestic or near-shore production is not a cost increase. It is an investment in continuity. Use trade tools to encourage North American production. The U.S. government should consider offering incentives that reward companies for keeping more of the chipmaking value chain in North America. There is already a model for this. Under the United States-Mexico-Canada Agreement (USMCA), cars must have at least 75% of their value produced in the United States, Mexico, or Canada to qualify for tariff-free treatment. A similar standard for semiconductors could measure how much of the front end and back end work happens within those three countries. Mexico already has significant electronics-assembly capacity whose costs are well below U.S. levels, so that can provide a strong demand signal for localized production. And data center components are already increasingly flowing from there. This gives companies a near-shore alternative to Asia. It also gives back-end suppliers a reason to set up in North America: test and packaging companies go where their customers are, and a regional content standard would help grow that customer base. Offer government incentives to buyers. The logic of the CHIPS Act was straightforward: Subsidize the large, visible anchors of the value chain—the fabs—and the surrounding ecosystem and customers will follow. But that assumption has not held up. So, instead of just subsidizing chip factories, the U.S. government should also give buyers incentives to purchase domestically. Tax incentives for companies that buy American-made electronic assemblies for critical infrastructure like data centers, communications networks, and medical equipment would create strong demand for domestic sources of supply. Then the whole semiconductor-production ecosystem—and not just the fab companies—will build the capacity to serve them. The U.S. government has used this approach before. Demand-side incentives in the Inflation Reduction Act generated significant investments in clean energy technologies and substantial demand for electric vehicles until they were largely discontinued by the Trump administration and the Republicans in Congress. And looking back to the 1910s and 1920s, U.S. tax incentives for domestic production of oil and gas set the United States on course to be a global leader in the sector for much of the 20th century. . . . The onshoring of semiconductor manufacturing is happening in the United States, but the job is not finished. The country still lacks sufficient facilities that turn wafers produced domestically into finished chips. If America doesn’t take steps to nurture the back end of semiconductor production, it will end up with an impressive collection of factories that can fabricate semiconductors but cannot turn them into a final product.

Where the U.S.’s Chip Strategy Is Still Falling Short

Summary.

Partner Center