This was in the late 1980s, and while his sexist comments need to be amended, his broader points are suddenly incredibly relevant in 2024.
A fab is a factory where semiconductors are manufactured. These gigantic facilities cost billions of dollars and take years to build, and are very difficult to operate efficiently.
For decades, Intel was the clear leader in this field, but a series of missteps began to chip away at its lead around 2018. Taiwanese startup TSMC has grown steadily and is now arguably the world’s best chipmaker.
Ian King, a veteran chip reporter at Bloomberg, wrote about this in late 2018. Since then, Intel has plummeted and TSMC has soared. Intel is now worth less than $100 billion, which puts it outside the top 150 based on that metric. TSMC is now worth almost $1 trillion, putting it in the top 10.
Why do countries need to have fabs now?
This is a stunning setback for Intel and a major strategic and geopolitical problem for the United States.
If you want the best chips, you have to go to Taiwan to have them made, or maybe South Korea, where Samsung has built up a massive so-called foundry business that makes semiconductors for other companies.
Many of the big name “chipmakers” we think of in the US today don’t actually make chips. Nvidia, Qualcomm, AMD, and all the others design their chips and then have them manufactured, typically by TSMC. Apple and other big tech companies also have TSMC manufacture the chips they design.
Again, in practice, producing these complex products at scale without small defects is extremely difficult.
If China were to invade Taiwan or tighten its grip on the island, it would be a disaster for the U.S. and Europe. iPhones wouldn’t be as fast for a while. Nvidia would have to look elsewhere for GPU manufacturers. AI progress might stagnate.
That’s why the tagline “real countries have factories” is more appropriate today. Chips power modern economies, and if we have to source these components from overseas, we become vulnerable.
Qualcomm can’t fix Intel’s alarming decline
That’s why Intel’s decline is so worrying: It’s the only U.S. company that knows how to manufacture powerful chips at scale. (GlobalFoundries also makes semiconductors, but not at the cutting edge.)
Related articles
The WSJ reported Friday that Qualcomm had recently approached Intel about a deal, following a Reuters report two weeks ago that Qualcomm was considering buying part of Intel.
Even if a deal goes through, it wouldn’t solve America’s chip manufacturing problems. One analyst called the potential partnership “strange.”
Qualcomm probably isn’t interested in Intel’s manufacturing business, though Reuters reported that the company is interested in some of Intel’s chip design business.
Intel has two main businesses: designing semiconductors for PCs, data center servers, and other applications, and manufacturing those designs.
For decades, Intel’s design and manufacturing operations were tightly integrated. This worked well for many years: the company was able to build factories exactly to the specifications of its in-house chip designers.
Then the world started to move to a different approach, pioneered by TSMC: Instead of designing and manufacturing our own chips, why not just run a fab and manufacture chips for other companies?
When TSMC first got started in the late 1980s, the idea was laughed at (which led AMD’s Sanders to make sexist comments about real men owning factories).
But helped by Intel’s missteps and advances in modern technology, TSMC’s approach gradually caught on.
The big event was when Intel missed out on manufacturing chips for the revolutionary device, the iPhone, when it first came out. Apple ultimately chose TSMC. Qualcomm, also a major designer of chips for smartphones, has most of these components manufactured by TSMC. Other chip designers, including AMD, also started turning to TSMC.
This gave TSMC the vast and diverse manufacturing volumes it needed to learn how to make chips better than anyone else. Ian King’s 2018 article provides a good explanation of this virtuous feedback loop:
There are billions of transistors on a chip, and a problem with just a few of those tiny switches can render the entire part unusable. Production can take up to six months and involve hundreds of steps that require meticulous attention to detail. Each time a mistake is made, factory workers can tweak it and try a new approach. If the changes work, that information is stored and can be tried on the next challenge. The more runs the better, and TSMC currently has the most runs.
While TSMC learned from a wide variety of large customers, Intel’s manufacturing operations were limited to just one customer: Intel itself.
As smartphone chips became mainstream, Intel didn’t have the production volume to keep up with TSMC in the manufacturing race, and AI has made this worse: Nvidia holds a commanding lead here, with TSMC making the GPUs.
The stubborn miasma of manufacturing
Undoing the manufacturing disruption surrounding Intel will be a costly, risky and complex endeavor.
Intel has started paying TSMC to manufacture some of its chips, and while that’s a start, there’s still a long way to go.
The company recently took a big step by separating its foundry business from its chip-design business, making it easier for outside customers to hire Intel to make their chips without having to compete with it.
The next challenge is the biggest challenge: getting good at making chips again.
Intel’s foundry business won’t be able to compete with TSMC until it secures a few big customers, and again, to succeed in chipmaking you need high-volume, diverse production volumes that allow you to find defects, fix processes, and feed that knowledge back into your own factories.
It’s a chicken-and-egg situation: Without volume production, outside customers will be wary of entrusting Intel with their valuable chip designs, but without customers, Intel can’t improve.
Does everyone like Raymond?
One way to break this impasse would be to lobby the US government to convince other companies to use Intel’s factories, which is exactly what’s happening right now, according to CNBC.
CNBC recently reported that Commerce Secretary Gina Raimondo is urging shareholders in companies like Nvidia and Apple to recognize the economic benefits of having U.S. foundries that can produce AI chips.
About four days later, Intel announced a partnership to make AI chips for Amazon Web Services. Intel shares soared on the news because AWS is the largest cloud provider and is designing huge amounts of chips for use in its giant data centers, which is what Intel needs.
Hope of 18A
On the technology side, Intel has a new process node called 18A, which is a set of chip design rules and accompanying manufacturing systems, and if all goes well over the next few years, Intel could once again be competitive against TSMC’s leading nodes.
The partnership with AWS is based on this 18A technology, and Microsoft announced earlier this year that it would also be manufacturing its own chips on this process node.
Intel’s foundry business needs more of these customers, and for that to happen, the 18A technology needs to work really well.
“18A should deliver stellar results, given that they are betting the company on its success,” Stacey Rasgon, a semiconductor analyst at Bernstein, wrote in a recent investor note.
Will the split work?
Meanwhile, Qualcomm does not appear to be interested in acquiring this part of Intel.
This raises the possibility of a breakup of Intel, which has been rumored for months. Reuters has reported that Qualcomm is interested in part of Intel’s chip-design business, and The Wall Street Journal reported on Friday that Qualcomm is likely to sell part of Intel to another buyer.
How will Intel’s foundry business function as a separate company, separate from its design division?
Again, the issue is about volume: Right now, Intel’s foundry business needs the manufacturing volume it can get from its own chip designs. Without it, there’s little manufacturing to learn.
“We have no idea how that is feasible right now,” Rasgon wrote in a memo in early September. “Given the significant losses and lack of scale, the manufacturing division is not currently self-sustaining.”
“A separation would only make sense if it attracted substantial third-party business, which seems years away (if it were to happen),” he added.
