Intel, now the second largest US chipmaker by market cap after Nvidia, wants to win back the top spot and stop losing market share to AMD. The pedigreed company’s plan to get back on track runs through a made-in-America future. 

Earlier this week, Intel said it would invest $3.5 billion to upgrade a New Mexico factory, on top of the $20 billion it’s spending on two new Arizona factories. The company will also try its hand in the foundry business with Intel Foundry Services, an independent business unit that will build semiconductors for other companies and report directly to CEO Pat Gelsinger. 

While the US government has yet to formally commit to subsidies for Intel’s US plants, that announcement may be coming soon. The White House has floated $50 billion in federal funding for US chip manufacturing and R&D, a proposal that has broad-based bipartisan backing. Particularly in the US Southwest, state and local governments have floated incentives for fabrication plants (fabs) to be built in their backyard. 

Plans to expand chip manufacturing capacity have taken on a new urgency. A perfect storm has thrust the word “semiconductor” into the Oval Office, earnings calls, and Brew headlines: 

1. Carmakers slashed chip orders last year. When auto demand picked back up quicker than expected, these companies were late to put in orders with contract chipmakers, which were prioritizing production for other sectors.
2. Over the past few months, chipmakers have seen a spate of force-majeure-type incidents at plants around the world, including a deep freeze in Texas, drought in Taiwan, and factory fire in Japan. 
3. Rising demand across the board, coupled with points (1) and (2), has prompted a global chip shortage, disrupting the supply chains of US-based Fortune 500s from Ford to Nvidia. 
4. Increasing a fab’s production volume is not an overnight process. 
5. The most advanced chips are made on the other side of the world, especially in Taiwan, where tensions with neighboring China are on the rise. 

If you’re a US executive, policymaker, or military planner, you want more US fabs to be built and finished, like, tomorrow. Even better, yesterday. But as you’ll read, spinning up those fabs is no small task. 

How we got here: A global perspective

In the early 1980s, the US accounted for 50+% of all semiconductor sales. Japan’s industrial policy agency MITI helped the country overtake the US later in the decade, but by the late ‘90s, American companies retook the lead, which they’ve maintained through today. As of last year, US-based firms captured 47% of the world’s $440 billion semiconductor industry. 

But the US share of semi manufacturing shrunk from 37% in 1990 to just 12% today, due to automation, globalization, state planning, subsidies, and industry consolidation. Now, advanced manufacturing of logic chips—CPUs and GPUs—is largely outsourced. Many major US semi firms, like Nvidia, Qualcomm, and AMD, don’t operate their own fabs. They design logic chips in-house, test them, contract production out to foundries, and then sell them, either integrated into a device or as a standalone product. Domestic US manufacturing of analog and memory chips is more common. 

In the foundry market, Taiwan is dominant. The country has 65% market share, according to Taipei-based TrendForce, versus South Korea’s 18% and China’s 5%. Taiwan-based TSMC, the world’s largest contract chipmaker, has 55% foundry market share alone. 

TSMC builds parts for Apple, AMD, Qualcomm, and many more. Emphasis on “many.” In 2019, TSMC says it made “10,761 different products using 272 distinct technologies for 499 different customers.”

The smaller the transistor, the more advanced the chip—and the more dominant TSMC is. The chipmaker dominates at the leading-edge (5nm). Smaller nanometers indicate new generations of manufacturing capabilities, and each subsequent “node” tends to: 

1. Shrink a zillion more transistors on a chip (i.e., power Moore’s Law)
2. Provide sizable performance and power efficiency boosts for applications like 5G and AI

TSMC and Samsung are at 5nm high-volume manufacturing, with 83% and 17% respective market shares in the last year. Five nanometers is very small. A human hair is on average 60,000 nanometers (nm) in diameter.

To put that in a geographical context, the world’s state-of-the-art fabs are mostly concentrated on an island ~110 miles off the coast of mainland China. Beijing is increasingly strident about its desire to “reunify” Taiwan and, more generally, to gain chip self-sufficiency. Even if the chances of a land invasion are remote, were it to happen, it would disrupt TSMC’s fabs and destabilize the world economy as a result. 

Last July, Intel announced it was delaying the transition to its smallest-ever size, 7nm. But pound-for-pound comparisons are apples and oranges. “Intel has higher design rules, so Intel’s 7nm is like TSMC and Samsung’s 5nm,” ASML VP of Tech Development Tony Yen recently told us. His company, along with a few others in the US and Japan, are the dominant suppliers of chipmaking equipment to TSMC and Samsung. 

“Overall, the 7nm development is progressing well and we are on track to restore process performance parity and resume leadership as quickly as possible,” an Intel spokesperson told the Brew. 

And the unrelenting quest for shrinkage continues: In early May, IBM revealed a 2nm production process for chips that it says would be 45% faster than today’s top 5nm chips, and consume 75% less energy. 

So, you want to build a fab? 

If Taiwan is frontrunner for the title of chipmaking “linchpin,” then Veldhoven, Netherlands, is the dark horse candidate. The Dutch town houses ASML, which has a monopoly in the extreme ultraviolet lithography (EUV) market. EUV machines support the most advanced chipmaking processes.

“If you are ordering from a country that does not require export control,” ASML’s Yen said, “an EUV order has about two years of lead time.” These machines have 100,000+ parts, cost $150+ million, and ship in 40 freight containers that fill three Boeing 747s. ASML’s big customers—TSMC, Samsung, and Intel—have to think technological generations ahead when placing orders.

“When Intel initially designed 7nm, EUV was nascent technology so we developed our process to limit our use of it, but this also increased process complexity,” the Intel spokesperson said. “We have now fully embraced EUV. We’ve rearchitected and simplified our 7nm process flow, increasing our use of EUV by more than 100%.

What about the fabs? How long does building them take?  

“You need years,” Will Hunt, a research analyst at Georgetown’s Center for Security and Emerging Technology, told the Brew. First, you must secure land and permits where you plan to build. 

• “Then, you have to construct the building where manufacturing will take place—the ‘shell’—which can take a year or more.” 
• Next, “you have to install equipment and get it qualified, which can take another year.” 
• Finally, “it takes another three months or more to ramp up to volume production.”

You’re not just building a large shack for computers. Your shell needs a “substantial amount of structure to maintain cleanliness and minimize vibration in the fab itself to meet the precise performance specifications,” ASML EVP of Applications Jim Koonmen told the Brew. 

You also need a highly skilled workforce: “These new fabs rely on connection to the global [integrated circuit] ecosystem,” Koonmen said, “a highly effective collaborative knowledge network that has taken 50 years to build.”

What’s underway?

Intel expects the Arizona fabs to be operational sometime in 2023. TSMC, which will spend $100 billion to expand capacity over the next three years, has committed $12 billion for a Phoenix, AZ, fab, with construction slated to start next year. And this week, Reuters reported that TSMC may build up to five fabs in Arizona. 

Samsung already operates a high-end memory fab in Austin, TX, which also looks like a leading candidate for a $17 billion additional US fab from the South Korean conglomerate. 

TSMC’s Arizona fab will be a fraction of its overall 5nm capacity, and by 2023, the company will have moved on to a more advanced generation. In Taiwan, TSMC has already broken ground on a $19.5 billion 3nm fab, with the goal of beginning production in 2023. 

Due to fabs’ lead time, these new plants aren’t being built in response to the current shortage. They’re meant to expand long-term capacity. To pull the lever on a chip shortage, “you increase utilization” at existing fabs, Hunt said. At those fabs, utilization rates are already high, and increasing them more takes time (like everything else we’ve discussed here). 

New US fabs will expand capacity across the board and help diversify some production from Asia. “As the current chip shortage demonstrates,” Intel told the Brew, “there has never been a greater need for more semiconductor manufacturing capacity and a more globally balanced supply chain.”

Looking forward

Remember that the US, where the semiconductor was invented, is still the biggest player in this broadly defined, complexly constructed, and highly distributed market. 

The US has many chip hubs, from (duh) Silicon Valley to Oregon’s Silicon Forest to Austin’s Silicon Hills and more. US companies maintain global leadership in designing chips and creating other associated intellectual property. And while China is spending tens of billions to bolster its own chip industry, it still relies on machinery, technology, and parts from elsewhere (including the US and allies). 

Two years can go by quickly. But the next two may feel glacially slow for foundries, fabless chipmakers, and the cascading web of industries that depend on chips but can’t get them (basically everyone). 

Even when the Great Shortage™ is alleviated in a year or two, the US won’t have cutting-edge logic chip production capabilities. Losing that edge was a multi-decade process—It can’t be unwound in a matter of months.