How to Rescue Intel
Equity is nice, but the real solution is much simpler
There’s been a lot of Discourse recently about the Trump Administration taking a stake in Intel. The beleaguered American Icon is, without a doubt, a shadow of its former self and in dire need of a turnaround. The causes of Intel’s decline are numerous, and go back decades (TL;DR: in 2005 Intel stopped putting engineers in charge and instead was run into the ground by finance guys from Wall Street), but to put it succinctly they can’t keep up with the competition under their current business model. They are in bad need of customers to get their foundry business off the ground. Thankfully the solution is straightforward: Intel needs customers for its Foundry service. The government should be that customer.
By now most people nerdy enough to read this blog will know the vague arc of the semiconductor industry: Early transistors were invented and commercialized in Silicon Valley, with Intel being the main company to scale production of what we now know as the modern CPU. This created the Intel of yesteryear, one of the largest and most iconic technology companies in the world which was effectively a monopoly (part of the famous “Wintel” alongside Microsoft Windows). Eventually, though, Intel got lazy and stopped innovating at the rate needed to maintain that monopoly (in particular, it failed to invest in EUV technology despite playing a key role in inventing it) which allowed Taiwan Semiconductor Manufacturing Company (TSMC) to leapfrog it and become the new de-facto monopoly on high end chips. This has made TSMC arguably the most important company in the world, which is a major concern given its… um… volatile location. To understand how to rescue Intel, though, we need first to understand how a modern semiconductor manufacturing company operates. That means we need to look at TSMC.
TSMC operates under a “foundry” model, meaning it doesn’t design the chips it makes. This complements the business model driving most of the chip companies that the average person might know like Nvidia, AMD, Apple, etc: so called “fabless” semiconductor companies that design their chips in house and then outsource the actual physical construction of those chips. TSMC’s foundry model has proven highly successful; it builds nearly all of the worlds high end chips. This has provided it with the scale needed to continue to push the boundaries of semiconductor manufacturing even as the cost of building a next generation chip plant has ballooned to north of $30 Billion. Basically everyone now agrees that the old Intel model of vertical integration just doesn’t provide enough scale to maintain a leading edge on the manufacturing side. Even with margin stacking (vertical integration has its advantages!), Intel just can’t generate enough free cash flow to match TSMC’s investment without the market dominance it used to enjoy (not to mention the aggressive subsidies the Taiwanese government provides TSMC). The solution: if you can’t beat ‘em, join ‘em.
That’s exactly what Intel is doing, or at least trying: it is investing 10’s of Billions of dollars to create new manufacturing capacity in the hopes that it will entice some of TSMC’s best customers to switch, or at least dual source. Unfortunately, that hasn’t happened yet. Intel has yet to sign any significant customers, resulting in big hits to its bottom line. The response from investors has been harsh: the board of directors fired the CEO and installed a new one, who promptly pulled back on investment in fabs abroad and slowed investment in the US. The message is clear: Wall Street will not tolerate taking any more big risks that imperil their short term profits, long term consequences be damned. This has prompted calls for a bailout, or at least some kind of intervention by the US government: it is simply untenable for the global semiconductor supply chain to be so concentrated within a single company located on an island which could, at any time, be conquered by China.
The Trump Administrations response thus far has been to take a stake in Intel to the tune of about 10% of the company. This was done by converting previously allocated grants (aka money the company was already going to get from the CHIPS Act) into equity, effectively giving taxpayers free Intel shares. While not a bad deal for the public, this does not solve either of the immediate problems Intel faces: it needs money to make big investments, and it needs customers to sign up for Intel Foundry Services (IFS) to de-risk those investments. In theory the first problem could be solved by raising money through capital markets, but as the Intel Board has demonstrated investors currently view that as too risky. Fortunately, as I alluded to in the beginning of this post, the solution is simple: in addition to taking a stake in Intel, the government should be IFS’s first big customer.
“But Matt!”, you might exclaim, “The government doesn’t design chips!” Ok ok, you got me there. The government can’t be the *direct* customer of IFS. The best companies to be IFS customers are the same ones that currently buy TSMC foundry services. Those companies have, thus far, been highly skeptical of IFS. The reasons for this are many, and not all public, but boil down to a combination of cost and risk. IFS is unproven, while TSMC has a now decade+ long track record of delivering great service. Being IFS’s first customer means being their for all the initial growing pains as Intel figures out how to offer good customer service, how to respond to non-Intel design needs, and how to balance external vs internal capacity. In a sense, it means helping teach Intel how to be a foundry like TSMC. That’s an added burden nobody wants.
The US Government can solve this problem by making what’s called an Advance Market Commitment. Basically, some US department would promise that anyone who spends the time and energy to help Intel get IFS running will have a guaranteed customer paying an attractive price. This would be expensive, probably on the order of a few billion dollars, but it wouldn’t necessarily be a total loss. In exchange for making such a commitment, the US government would receive a bunch of CPUs and/or GPUs that it could then use for any of the compute intensive services the government already offers, such as building new supercomputers at DOE national labs. If it were really ambitious, it could build out a network of publicly owned datacenters and lease them to the big AI companies like OpenAI, turning the whole thing into a potentially profitable endeavor. After that initial commitment was exhausted, the new foundry capacity built would be pivoted to private sector customers, who could evaluate it’s now proven track record. In the worst case, if Intel and partners can’t deliver, then the government need not buy anything. (this is the beauty of Advance Market Commitments: you only pay for success).
This would shore up Intel’s manufacturing capacity for at least a few years, and give it a fighting chance to reclaim the lead in process technology. In the long run the US will still need to figure out a more systemic industrial policy to counter those of China, Taiwan, Japan, Germany, and … basically everywhere else, lest we be forced to repeat this process of bailing out our industries every few years. That, however, is a much bigger problem. For now, if we want to save Intel, we should keep it simple: they need customers. Let’s give them one.
Thanks for reading! If you enjoyed this post, which is my first in a long time, consider subscribing. This is a personal blog that I write when I have something to say. I used to be a Tech and Trade Staffer in the US Senate, and I lived on TSCM’s campus in Taiwan for a while, so I tend to write about semiconductors, technology, industry, and any other random topic I find interesting.