Reviewing a Month of Big Deals in the Chip Industry
IntelFoundry, the M1, AMAT M&A, NSCAI recs, and Supply Chain Re-shoring
Welcome to SemiLiterate, a guide to the semiconductor industry through the lens of public policy. And greetings to new readers. Thanks to ChinaTalk for the fun podcast.
While you’re here, take a look at previous posts on What to Do About Intel, the Made in China 2025 Progress Report for semiconductors, and The pros and cons of the CHIPS Act.
BLUF: Intel’s announcement this week that it is re-entering the foundry business is a big deal. But there were several other Big Deal stories about the chip industry this month. This post summarizes five of them: (1) IntelFoundry adds to contract manufacturing’s best year ever (2) Apple’s M1 chip performs better than chips designed by chip companies (3) Applied Materials’ pending Japan acquisition is nixed by Chinese regulators (4) the NSCAI report recommends controls on DUV lithography tools to China and (5) Re-shoring semiconductor supply chains gains steam.
1. IntelFoundry and its Implications
The announcement that Intel is getting back in to the contract manufacturing business via IntelFoundry is a big deal. These will be the first 300mm wafer lines built in the U.S. in about ~10 years (or whenever it was that GlobalFoundries finished building out its Malta, NY facility) and Intel’s announcement dovetails nicely with the current administration’s focus on re-shoring semiconductor supply chains. It also (1) signals the end of Nikon’s attempts to develop Argon Fluoride immersion (ArFi) lithography as a viable competitor to ASML’s extreme ultraviolet lithography (EUV). And (2) it basically was one big mea culpa: Intel admitted it was wrong and its struggles with 10 and 7nm are ongoing while also tacitly admitting TSMC owns the 2020s for the chip industry. I’m not going to write up Intel’s announcement in great detail but will address #1 and #2 briefly below. Everyone should just read Ben Thompson’s (excellent, free, comprehensive) summary. With that said, here is how Semi-Literate characterized Intel’s options last fall:
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.
Intel is doing just that: spending $20 billion to build a commercially-focused leading edge foundry business is a great start. While this is a surprise, Intel has been testing the idea in public for a while going so far as having the CEO send a letter to the Pentagon saying “we want to make a foundry” in April 2020. $20 billion sounds like a lot, but TSMC’s CapEx for 2021 is supposed to be around $28 billion. Its going to be an uphill battle. Intel’s announcement will have two interesting follow-on effects for the chip industry:
Leading edge lithography belongs to ASML. Intel is all-in on ASML and EUV, meaning Nikon is out of the leading edge lithography game: Nikon’s ArFi lithography system (which is what Intel went with for 10nm and 7nm, thus all the delays) is done. Expect some interesting M&A activity around Nikon in the next year or two, possibly as Nikon tries to recoup some of its losses via sales of components to nascent Chinese SME firms. China remains locked out of purchases of EUV lithography so DUV is the best it can hope for (see #4 below).
The 2020s belong to TSMC: Intel wont have 7nm processors to market until 2023. We’re still in H1 of 2021 and Intel was originally supposed to have 7nm out in late 2019 or early 2020. In effect this means IntelFoundry’s competitors (TSMC and Samsung Foundry) will enjoy at least 5+ years lead on IntelFoundry as both have been producing commercial volumes of 7nm chips since Spring 2018. This lead will be all the more lucrative for TSMC because it gives them time to further lock in customers like Apple, who are currently using TSMC’s most advanced process node (5nm) for their state of the art system on chip (SoC):
2. Apple Silicon and the Vertical Integration of Big Tech
The second big deal this month is the ridiculously good reviews Apple’s M1 chip continues to get. With this chip, Apple Silicon has officially entered the top tier of fabless companies. To put that another way: A company primarily known for its consumer electronics has built an in-house silicon design shop whose performance now exceeds that of Fortune 500 companies whose only job is chip design.
Apple buys as much of TSMC’s 5nm capacity as it can, using at least some of it to fabricate its M1 SoCs (think of an SoC as “a chip of chips”). Everyone said the M1 was absolutely outstanding when it was released in 2020, but we’re only now realizing how good: Adobe says Photoshop runs 50% faster on a Macbook with an M1 as opposed to a Macbook with an Intel CPU. A fifty percent (!) improvement. The M1 is also the worst chip Apple will release in the next decade: Apple Silicon will continue to iterate and get better with each generation just like Apple’s consumer electronics have for 10+ years. But Apple is going to iterate using TSMC, not Intel.
Every one of Apple’s fabless suppliers and competitors should be watching M1 news like a hawk. The introduction of this chip will be remembered as the start of a new age in which Big Tech began producing chips as good, or better, than fabless firms. Apple’s long term goal here is to achieve performance advantages, design-out its traditional chip suppliers, and eventually realize cost savings. Every tech industry company wants a hardware lead who can accomplish what Johny Srouji has (that link is a fantastic profile, by the way), but it takes time: he’s been with Apple since 2008. Google, Microsoft, Amazon, and Facebook are all investing heavily in chip design but really only began doing so in the past 5 years, so any notable chip products wont roll out until the mid-2020s (Google’s TPU being a notable exception).
3. Chinese Anti-Trust Concerns Nix Applied Materials Acquisition (the trade wars continue)
This headline from March 29th says it all:
Applied Materials has terminated its $2.2-billion deal to buy Japanese peer Kokusai Electric Corp from KKR & Co, the chip equipment maker said on Monday as it did not get a confirmation of timely approval from China’s regulator.
The precedent for big mergers and acquisitions in the chip industry is pretty well-established: Flashy announcement followed by a regulatory mess as companies try to seek anti-trust approval from governments around the world. Applied Materials is now 0/2 in attempted acquisitions of Japanese semiconductor manufacturing equipment firms (they attempted to buy Tokyo Electron in 2015). That 2015 deal fell apart due to (probably well-founded) U.S. anti-trust concerns but China’ decision here will reverberate much longer. This is the equivalent of Boeing trying to buy Brazil’s Embraer and France stepping in to say “nope.” Interestingly, China did approve the Marvell-Inphi tie up this past week. So maybe just mergers in the semiconductor manufacturing equipment industry are verboten now that China realizes it needs to do everything possible to protect and promote its domestic chip manufacturing equipment firms? China nixed the NXP-Qualcomm deal in 2018, but this is different.
There are three major pending M&A deals that will need Chinese regulatory approval, and the prospects aren’t looking great for #1 and #2:
AMD-Xilinx: this echoes the Intel-Altera tie up (Altera, like Xilinx, makes FPGA chips and AMD, like Intel, uses the x86 architecture). Intel-Altera was announced June 2015 and closed December 2015. AMD-Xilinx was announced October 2020 so if no progress is announced in April it could be a sign of trouble.
Nvidia-Arm: Announced September 2020... this deal seems exceedingly unlikely to secure regulatory approval, even absent Chinese intervention. The EU has already opened an investigation in to the transaction and UK regulators are getting an ear-ful from Graphcore (owned by Microsoft)) in opposition to the deal. Arm is simply too ubiquitous in the industry for Nvidia’s competitors to allow it to control that much IP. In addition, Arm has a Chinese subsidiary whose CEO has gone rogue and disregarded a 7-1 vote by Arm China’s board to remove him in Summer 2020, so all bets are off for this transaction.
Analog Devices-Maxim Integrated: This one has been flying under the radar but both companies are going to have a great 2021. If it goes down in flames due to Chinese anti-trust concerns, it will mark a new low.
Regardless of the outcome of these three deals, China’s move here should be interpreted as the start of a new chapter in the struggle for control over this industry. Just like the U.S. government’s action to stop the proposed takeover of Micron by China’s Tsinghua Unigroup in 2015 signaled a new level of scrutiny from U.S. regulators, this move to use Chinese anti-trust authority indicates China will be taking a much tougher look at any chip M&A that falls within their purview going forward.
Applied Materials now has to pay KKR a $154 million cancellation fee. Nvidia is going to owe SoftBank (the owner of Arm) substantially more.
4. NSCAI Final Report Recommends DUV Lithography Block to China
Careful readers will note one through-line from each of the aforementioned Big Deals is the Dutch lithography company ASML: IntelFoundry is committed to using ASML’s EUV tools for the foreseeable future, Apple’s M1 chip is fabricated by TSMC using ASML’s EUV tools, and Applied Materials lost the #1 spot in the semiconductor manufacturing equipment market to ASML in 2020 (its failed acquisition was part of its attempt to reclaim the top spot). Its ASML’s world, we’re just living in it.
That is what makes the microelectronics-related recommendations (Chapter 13) by the National Security Commission on Artificial Intelligence (NSCAI) all the more interesting: among other things, the NSCAI recommended keeping China two generations behind the leading edge in microelectronics fabrication capabilities. The primary means by which the NSCAI recommended the U.S. keep China two generations behind the leading edge is through export controls on Deep Ultraviolet Lithography (DUV) equipment. But this equipment is only made by two companies in the world, and one is in The Netherlands (ASML) and one is in Japan (Nikon). So this recommendation will require some diplomacy to execute.
This recommendation’s timing was also a bit ironic because the same week the NSCAI final report was released, ASML extended its $1.2 billion dollar sales relationship with SMIC (China’s leading foundry, and a U.S. Entity Listed firm as of December 2020) to sell Deep Ultraviolet Lithography (DUV) equipment through the end of 2021. But DUV has been in China for a while. Notably, SK Hynix announced it was brining DUV to its China factory in 2019. Nikon also has sold its DUV tools to China in the past as its chipmaking equipment business continues to struggle.
So it would appear that there is already substantial foreign availability of DUV equipment in China, making any new controls likely too late. The Bureau of Industry and Security’s guidance on foreign availability is pretty clear:
Foreign availability exists for a national-security controlled item when a non-U.S. origin item of comparable quality is foreign available, and in sufficient quantities, such that the U.S. export controls on that item would be rendered ineffective.
In addition to this foreign availability, Chinese firms are trying to buy and build their way in to the DUV lithography market. Shanghai Micro Electronics Equipment (SMEE) will be a name to watch in 2021 and beyond as they represent China’s best shot at developing an indigenous DUV tool. No doubt they are trying to buy lithography components from Nikon to advance their efforts.
With that said, this was a great report by NSCAI and it accomplished a rare feat by government report standards: in addition to diagnosing problems it actually managed to prescribe tangible solutions. The NSCAI chapter on microelectronics is genuinely quite good. And for those who think otherwise, at least it offered something to argue with rather than platitudes.
5. Re-shoring Semiconductor Supply Chains
Re-shoring American Chip Supply Chains:
The recent Biden Administration supply chain executive order specifically called out chip supply chains as an area of concern, and The Commerce Department’s Federal Register notice seeking comment on semiconductor supply chain risks is pretty interesting (as far as FR RFIs go). Commerce is hosting an open virtual forum on semiconductor supply chains in early April. On top of the IntelFoundry announcement and The CHIPS Act passage in January 2020, this political momentum comes after several favorable developments for the domestic US chip industry:
TSMC has broken ground in Arizona on a 5nm fab coming on line in 2024 and is in the midst of a hiring blitz. Rumors of a mega fab valued at $35 billion abound.
Samsung is building a 3nm factory in Texas. $17 billion potentially.
Micron is building out their fab in Manassas (which mainly does long life cycle/automotive chips) to expand capacity and considering a bid for Japan’s Kioxia (formerly Toshiba Memory).
GlobalFoundries has been oddly silent to this point. Expect an announcement sooner rather than later, probably when the CHIPS Act is funded by Congress.
Re-shoring European Chip Supply Chains:
Meanwhile Europe announced a $145 billion fund to shore-up European competitiveness in chips and Europe is courting Samsung and/or TSMC to build a new chip factory in the region. TSMC went on the record in the Financial Times last week to say “thanks but no thanks:”
In the US, we committed to building a fab after the authorities made clear that they would subsidize the cost gap. In Japan, our investment is focused on an area of technology that is key to our future,” says a senior TSMC executive. “But in Europe, the case is not that strong, and [the Europeans] really should figure out what exactly it is they want, and whether they can maybe achieve it with their own chipmakers.”
This is smart by TSMC. The value proposition of an advanced chip fab in Europe seems odd given how far behind the rest of the world the European chip industry currently sits. The crown jewels of the EU chip manufacturing industry (Infineon, NXP, and STMicro) don’t do (and don’t need to do) leading edge chip manufacturing and most major consumers of chips in Europe (Nokia, Ericsson, VW, BMW etc) don’t need the most advanced chips for their products. This $145 billion fund seems like an expensive vanity project with uncertain return on investment.
It will however be a boon for whichever company/companies benefit from these funds. Europe should probably give all the money to IMEC in Belgium, encourage them to establish some consortium with TSMC to accelerate “beyond” EUV research, and re-tool the GlobalFoundries fab in Dresden and ST Micro’s in France, and call it a day. Interestingly, Intel did make clear its factory in Ireland will get some of the IntelFoundry investment. And Apple recently announced a $1.2 billion chip design center in Germany. So maybe Europe will be able to claim victory after all.
Conclusion (other big deals)
Just as important as these five subjects are some things that didnt make the cut:
TSMC and Samsung are 43% of whole industry’s CapEx: Between 2017 and 2020 Samsung spent $93.2 billion on chip CapEx, more than double the $44.7 billion spent by all indigenous China semiconductor suppliers combined. Meanwhile TSMC plans to spend $25-28 billion this year. Governments in the EU, U.S., and China would need to spend $30 billion/year for five years to catch up with Samsung and TSMC’s lead.
Automakers continue to struggle with the tight supply of chips, especially in Japan and the US. Suzuki had to shut down 2 of its 3 auto plants and GM is just shipping cars without some electronics now.
Thanks for reading.
Note: the views here are my own and drawn solely from the documents I cite here.