What to do about Intel
The Good, The Bad, The Ugly: Is Intel the next IBM?
BLUF: Intel is the US government’s best bet when it comes to ensuring secure domestic advanced semiconductor manufacturing. Because of that, there is a temptation on the part of policymakers to do anything and everything necessary to help Intel. But Intel is a large successful commercial firm that is large and successful due to its focus on commercially popular electronics, not government priorities. Intel will need to do more than create a leading edge foundry using US government money and purchase commitments if it is going to catch up with TSMC and Samsung. This challenge is all the more acute given the headwinds Intel currently faces.
The Good
It’s popular to talk about Intel’s struggles these days. But, Intel is still Intel. It basically prints money in its CPU business and maintains comically-high market shares (90-99%) in the PC and server microprocessor markets. As such, it can afford to make eye-popping investments in emerging technologies and ascendent competitors, to say nothing of burning the annual CapEx required to run copy-exactly factories in Oregon, Arizona, New Mexico, Massachusetts, China (for now), Ireland, and Israel. Intel has a lot of money, so Intel can afford to be wrong some of the time. And thats good, because Intel’s been wrong a lot lately.
The Bad
Just like the rest of the world, Intel has had a rough 2020: after a decade-plus of partnerships with Intel, Apple filed for divorce in June (on the heels of Apple’s 2019 purchase of Intel’s smartphone modem business). A month later, Intel announced it had a process-slip which meant its 10nm chips, already delayed, would arrive a full year later than forecast. Then in August it divested its NAND memory business to SK Hynix after years of struggles. Meanwhile, competitors like AMD and NVIDIA have been on an absolute tear, leveraging high stock prices to acquire Xilinx (AMD) and ARM (NVIDIA) which will cement their competitiveness in the industry, if approved.
Intel’s struggles at 10nm have meant the United States no longer has domestic state of the art manufacturing capabilities, as TSMC and Samsung have pushed the leading edge to 5nm manufacturing at some of their fabs in Taiwan and South Korea. This is troubling for consumer electronics companies in the U.S. that hoped Intel would provide an alternative to their reliance on TSMC for advanced chip manufacturing. Intel’s struggles at 10nm are also troubling for the U.S. government, which likes the idea of having commercially viable domestic advanced semiconductor manufacturing for economic and national security reasons and generally hates the idea of Intel ending up like the next IBM.
The Ugly
So, if you’re the U.S. government and you see a Fortune 100 chipmaker thats hit a rough patch, what are your options?
Do nothing: this is the approach the US government adopted when it was clear IBM’s microelectronics business was in decline. The result was IBM sold their fabs to GlobalFoundries which, though US-based, is owned by Abu Dhabi’s Mubadala Investment Company. GloFo has had plenty of struggles of its own since then, trying and then giving-up on EUV and sub -12nm manufacturing and selling off one of its fabs to ON Semi last year. Suffice to say, a foreign buyer acquiring advanced microelectronics manufacturing assets, failing to maintain the investment level needed to stay at the leading edge, and then divesting some of those assets, is not an outcome policymakers want to see again.
Money: fabs are expensive. Sub-10nm fabs are tremendously expensive. If your goal as a policymaker is secure (Made in the USA by a USA-headquartered company) commercially viable leading-edge microelectronics manufacturing, giving money to Intel to re-tool a fab is basically your only option. The government’s of China, Japan, South Korea, Singapore and Taiwan have all found creative and WTO-permissible ways to subsidize their chip industries. Perhaps its time to get creative. The CHIPS Act is a good start, but grants of up to $3 billion for new fabs don’t move the needle long-term. Before giving up on the idea, GloFo estimated just getting from 14nm to 7nm would cost that much. Meanwhile Samsung is talking about spending $116 billion in the coming years to catch TSMC, which is spending $17 billion annually to maintain their lead. These numbers are just impossible to comprehend. $116 billion is larger than the nominal GDP of 127 countries according to the World Bank.
Break up: theres been a little chatter about breaking off Intel’s non-core businesses and in fact, it sold off several in just the past year. In addition to the aforementioned sales of the smartphone modem and NAND memory units, Intel also sold off its power management chip business in November 2020. There’s also been some highly speculative talk about Intel splitting off its manufacturing assets into a standalone company - lets call them “IntelFoundry.” This decision would run in to a million commercial challenges, and there isnt much role for policymakers in this scenario, but its an interesting thought experiment and brings us to the next option:
Marriage: This is a sort of choose-your-own-adventure two-part hypothetical:
In Option #1: IntelFoundry merges with GloFo and/or SkyWater to create a sort of national champion for advanced microelectronics manufacturing in the United States, one that gets below 10nm and focuses more on servicing the military/aerospace chip market than Intel has traditionally. Theoretically policymakers have some role in this insofar as GloFo and SkyWater both proudly do business with the US government and a tie up between any/all of these companies could be facilitated by US government purchase commitments, tax breaks, expedited anti-trust reviews, etc.
In Option #2: Intel sells-off IntelFoundry to any one of the big tech companies that are sitting on mountains of cash and seem to be increasingly interested in vertical integration of their silicon-oriented business units.
Both of these scenarios sound nice on paper, which should give the reader pause. Its not really clear what IntelFoundry gains by partnering with firms like GloFo or SkyWater that are behind them technologically besides potential economies of scale many years down the road and exposure to a Mil/Aero market that is notoriously unprofitable relative to consumer electronic sales. There are also myriad commercial reasons why these ideas are bad and/or exceedingly unlikely: economies of scale being disrupted, chip design/manufacturing feedback loops broken, earning the trust of new customers etc. Also, would Apple/Amazon/Google/Microsoft really want to run fabs? It seems unlikely policymakers wade in to this hypothetical morass. Which brings us to our final option:
IntelFoundry: The final option is a standalone business unit within Intel that specifically makes chips for third-party customers. This option is different from the Break-up scenario above in that IntelFoundry would operate in addition to Intel’s existing manufacturing assets. This paradigm exists: Samsung operates factories for its in-house chip fabrication needs but Samsung Foundry also takes orders from third-parties, including some competitors of Samsung’s. The US government is clearly very keen on this option and in late April 2020 Intel’s CEO sent a letter to the Pentagon saying “we want to make a foundry.” The fact that this letter was sent to the Pentagon indicates (1) DOD interest in securing advanced microelectronics manufacturing in the United States and (2) Intel’s commitment: DOD is really the only game in town when it comes to funding projects of this magnitude. It is currently funding a multi-billion dollar Electronics Resurgence Initiative and partnering with Intel in other areas.
Conclusion
Its obvious the US doesn’t want Intel to be the next IBM. But policymakers’ options to preclude that fate are limited right now to throwing money at the problem (a public private partnership, a tax credit, or straight cash, apparently) and hoping something sticks. This is a failure of imagination, a failure to understand the history of this industry, and a failure to understand Intel’s comparative advantages. We’ll tackle all that in a later post.
Note: The views here are my own and are drawn from the publicly available information cited.