What's next for Intel?

Intel is important.

Nvidia and TSMC get most of the press these days, but Intel still matters. Why? CPU’s and FPGA’s (see Altera spinoff), which are Intel’s main businesses, aren’t the shiny object that they once were. And in the CPU business, there are competitors: AMD, of course, but also ARM and RISC-V architectures. CPU’s still run most of the non-AI computation in the world, and FPGA’s run a large fraction of what CPU’s don’t. Intel still has something like 75% market share in CPU’s.

What a sad thing to say: Intel still matters. I remember – not long ago – when people regarded getting a job at Intel as a lifetime sinecure, where you could do good, impactful, interesting work without any real career risk. A lot of people went to Intel because it was a safe place. That culture of avoiding risk resulted in a lot of bad decisions.

Intel’s CPU product line isn’t why Intel is important today. And it’s not that they’re the only place building advanced semiconductors in the United States: TSMC has commissioned some fabs building advanced chips here in the US, and Samsung is in the process of doing so.

So why should we care about Intel? Their market cap, after all, is 1/10 of TSMC’s, and about 1/40’th of Nvidia.

Intel is important because they’re the only organization that develops advanced CMOS fabrication and cutting-edge packaging processes in the United States. TSMC can put up a cutting-edge fab in the US any time they want, but it requires a transfer of technology from Taiwan. The process development capability lives in Taiwan and isn’t leaving. Without the results of this process R&D, a semiconductor fab is just a big box full of expensive machines. The process development and associated operational skill is the key, and TSMC, Intel and Samsung are the only places in the world that know how to do this kind of R&D.

Without Intel, at every generation, the United States will have to apply regulatory pressure and incentives to get manufacturing localized by TSMC, if we want a plant on-shore. The organization that develops the processes lives exclusively in Taiwan.

The challenges

I’m quite puzzled as to why the public markets regard the government taking a stake in Intel as a good thing, as witnessed by the recent improvement in the Intel stock price. If Intel needed to raise money on a dilutive basis, they could have done so with an at-the-market stock offering on better economic terms. The federal grants that Intel is getting as part of the CHIPS act come with regulatory, political and legal strings, making the money far less useful than unrestricted funds raised from the markets. And the political overhead associated with dealing with the USG as a major shareholder is likely to create enormous cognitive load and distraction for Intel’s senior executives. They’re involved in high politics now, whether they like it or not.

So… what’s the point? Do the markets expect the USG to deploy all the tools of statecraft – especially the sharp power of the state - to support Intel as a national champion, in the same way that China does with their national champions? Such an outcome seems utterly unlikely to me; we’re just not organized to do things like this, or at least we’re not organized to do things like this in a very effective way.

A lot of ink has been spilled about what should be done with Intel. In general, the consensus seems to be that Intel should:

1. Separate the fabs into a business that can serve multiple customers: In effect, the fabs would become a foundry to compete with TSMC. This is a major change in business model, and is a transition that took Global Foundries many years and a couple of acquisitions.

2. Get rid of distractions that aren’t the core CPU business through spinoffs and shut-downs

3. Take the CPU business and turn it into a fabless semiconductor business, designing and selling CPU chips, and buying wafers from foundries

This plan is appealing in many ways, but there are several enormous problems with it:

1. A fab is not a foundry. Intel’s fab is set up primarily to service the internal customers at Intel. Foundries specialize in running hundreds or thousands of different chip designs for different customers. That means enormous documentation efforts, the creation of rock-solid PDK’s, and flexible, modular processes. It also means a different mix of equipment and personnel than a single-customer fab. Running a foundry is enormously complex.

2. An advanced CMOS foundry is a natural monopoly due to the IP moat. I wrote about this a while ago:

   Fabless semiconductor companies that use TSMC’s fabrication services get to tap into a very rich ecosystem of pre-designed IP blocks, from an entire network of different vendors. As a result, TSMC users can often license 90% or more of the design of their chip, while their engineers design only the pieces that generate significant technical differentiation.

   Doing so greatly reduces risk, and improves time-to-market. This network effect has not yet been replicated at Samsung or Intel; IP providers flock to the processes where they can capture the most users, and users flock to the processes where they can get access to the most robust basket of licensable IP.

   Because TSMC is so dominant in terms of number of users, the IP vendors use scarce resources to develop for TSMC, instead of Samsung and Intel.

3. Spinning off the CPU design capability will likely result in Intel’s CPU design group using TSMC, not the Intel foundry, in the absence of contractual or regulatory restrictions. TSMC is just easier to work with, and turns chips faster, and has a better IP ecosystem.

4. The ‘distractions’ were, in many cases, core to the Intel value proposition: Intel gave away a lot of IP and did a lot of engineering on all the things that went into a personal computer, so that they could give that IP away. They did a lot on things like Wifi, Bluetooth, and USB, in order to ensure that computers were as useful and valuable as possible. Intel wanted the CPU to absorb as many of the available margin dollars as possible. So they acted to ensure that there were multiple vendors competing to build everything else that went into a PC, at low margins. Those days are obviously gone, but getting rid of these ‘distractions’ is a sign that the old Intel business model is permanently gone.

5. Intel’s design team derives real value from their close relationship with the fabs. They won’t have this kind of relationship with a foundry that’s a separate business. How much value this interaction really generates is up for debate at this point. As a stand-alone fabless CPU business, they’ll have to use the same foundry processes as everyone else, and buy their wafers at price, not at cost.

A Lost Opportunity

A couple of years ago, the USG had a huge opportunity to help Intel, and failed to act. Patrick Gelsinger wanted to do a very smart thing: He tried to buy Tower Semiconductor. Tower is a lean, scrappy, very smart foundry that doesn’t do advanced CMOS. Their leaders understand the foundry model through and through. Intel needed a team of leaders who knew how to run a foundry. Tower had one.

China slow-rolled and then blocked the acquisition. Along the way they tried to extract a series of concessions. The Chinese regulators saw that bringing Tower’s expertise into Intel might put Intel into a place where they might plausibly compete with TSMC, and they didn’t want to see that happen. Having the world’s most important foundry in Taiwan, where China can strike at will, gives the CCP lots of leverage.

I hoped, at the time, that the USG would intervene and back Intel’s play: I would have loved to see a speech out of the White House saying, in essence, that if China wanted to block the deal, Intel would be shutting down all of their operations in China and would no longer sell advanced CPU’s in China. The damage to the Chinese high-tech industry would have been -enormous-, because re-designing a board to switch CPU’s is a massive effort. Alas, the will to act wasn’t forthcoming. The CCP wanted to damage Intel, and they succeeded.

What now?

So, as is so often the case in matters of geopolitics and economic statecraft, we are left to ask: What now? What is the least-worst option for Intel?

Becoming a foundry that can compete directly with TSMC isn’t in the cards. Doing that will take a decade, most likely. The TSMC moat around creating great, easy-to-use PDK’s, their design IP moat, and their advanced packaging moat are going to take enormous effort to bridge. It’s going to be a very long time before any newly funded fabless semiconductor startup in Silicon Valley has to make a thoughtful choice of whether to use the Intel foundry or go to TSMC.

But here’s what could be done: There are several very big companies here in the US who have the heft to build their chips at Intel, because they can afford to commission the creation of the design IP that they need. These are places like Nvidia, Qualcomm, Google, Meta, Amazon, Apple, Broadcom, Marvell, et cetera. Intel isn’t tooled up to handle 1,000’s of different chip designs at once, with different layer configurations and different requirements, the way TSMC does. Doing so is hard. But they could certainly tool up to handle a couple of dozen large-volume chips for these big companies. Doing so would be much easier than trying to go directly to being a full-service, open-market foundry.

If I were working in the White House and trying to save Intel, I’d put together a consortium of 5-10 of these major US-based semiconductor companies to each inject $10-20B to completely recapitalize Intel, rather than injecting more government money. I’d push to spin off the fabs to create a foundry, and take the foundry private, so that it’s not exposed to the whims of the public markets for a few years, while it retools and restructures itself.

For the first few years I would focus the foundry effort exclusively on serving the needs of a few giant, US based and western stakeholders, who would be on the board. And, if I were driving this initiative, I would use what leverage I could with regard to Cadence, Synopsys, and other major IP vendors to get them to develop cutting-edge IP into the advanced processes in the ex-Intel foundry.

Intel is worth saving. There are many actions that the White House could take to help Intel. But it’s not at all clear to me that putting the USG on the cap table is one of them.