My semiconductor conspiracy theories, part 1: Sept 19, 2020

Bonus post on Q-semicon

I get it, the semiconductor industry is not the most interesting one in the world. I want to present my case to you that it is fully representative of the present geopolitical and technical moment.

I think that the next world war will be with China, over Taiwan, due to the fact that the world’s most important company, the Taiwan Semiconductor Manufacturing Company (TSMC) is headquartered there.

And today I think we may have a messiah in our world who could prevent this war. With capitalism, of course.

First, a history of my interest in this. I am a trained computer architect, a fancy phrase for - I designed chips (along with huge engineering teams) that go into devices like smart phones, PCs, and most recently HPC and data center AI. And I married a guy who has a PhD in this field. He works for said messiah.

Around 5 years ago, when I was working for Samsung R&D in Austin, TX, all of a sudden several of our Chinese-origin colleagues started leaving for no-name startups. Turns out a bunch of these startups had been funded by the Chinese government to kick start homegrown chip design. The employees were offered 2-3x their current salaries, the opportunity to go back to China for extended periods of time, as well as be declared national heroes. For a lot of my colleagues, who were only children (born during the height of the 1-child policy), this was impossible to pass up. Would you blame them for wanting to be closer to their aging parents?

My husband meanwhile, was working for ARM, a British semiconductor IP licensing company (he has swapped over to working for Nvidia for a few years now). ARM’s IP is in 99.9% of every mobile and embedded device in the world, and mints royalty since they are a design-only shop that lets licensees bear manufacturing costs and risks. They succeeded (accidentally, IMHO) when Intel refused to entertain Apple’s iPhone chip business in 2005.

In 2016, Masayoshi Son and SoftBank acquired them for a 40% premium, citing grand tales of a 300 year vision and how ARM was a key piece of the “singularity”. The usual Masa fluff, except this flew under the radar because no one thinks semiconductors are sexy. Post-acquisition, the company party line heavily pushed getting into China. Masa has strong ties to China - as one of the earliest investors in many Chinese companies, most notably Alibaba. In the last few months, stories have been leaking out about ARM’s China unit CEO going rogue and preventing the UK HQ bosses from accessing the China site.

Building up semiconductor expertise and scale economics takes decades, and ungodly amounts of investment. Intel used to be the top dog in manufacturing chips - and they got their start in the 1960s. Moore’s law was created by Gordon Moore, Intel’s first CEO - an insight so powerful it made him quit the then dominant semiconductor company (Fairchild) to start Intel. However, these manufacturing capabilities were available and perfected solely for Intel’s own designs - classic vertical integration play. They paired up with Microsoft Windows and the “WinTel” partnership dominated.

TSMC, on the other hand, is a pure play shop. Their manufacturing capabilities are available to everyone. This open ecosystem allows two things -

  1. The rise of “fab-less” semiconductor companies like Qualcomm (where I used to work as well), Nvidia, ARM etc. - that can be purely focused on chip design expertise instead of worrying they need decades and dollars to also build up manufacturing

  2. TSMC to observe all these different players, learn quickly and improve their manufacturing processes (also called technology nodes) rapidly.

As a result of savvy partnerships and investments, TSMC is now unquestionably the best technology node in the industry. Ownership of the bleeding edge + scale + an open ecosystem means they can charge top dollar for their manufacturing capabilities, have lower unit economics AND robust business health because no one customer going away is going to hurt them - they have many, and can onboard new ones easily. There is a huge explosion in semi companies focused on AI right now, mine included :).

Meanwhile, Intel announced in July 2020 they are 4 years behind TSMC, which knocked 18% off their stock price the next day. In our current market environment where tech stocks seem to point solely in the up-and-to-the right direction. Gavin Baker wrote eloquently about how dire the situation is for Intel’s entire business model. They lost their CEO a couple years ago due to inappropriate workplace relationships and replaced him with the CFO. Several members of the leadership team have left. Intel stumbled on manufacturing so badly that they are relying on TSMC to bring their designs to market on time!

So here we have an American manufacturing and design powerhouse on the precipice of a downward spiral, a Taiwanese company rising to the top, and all of China’s might aimed at semiconductor independence, one of the last few things they depend on other countries for. And China adores Taiwan, doesn’t it :).

The next best semiconductor manufacturing fab is owned by Samsung in South Korea, not too terribly far from North Korea’s supposed nukes.

Lots of funny stuff started happening in our industry. In 2018, Bloomberg reported of a backdoor that China shoehorned into chips manufactured for our data centers (edit: apparently this has been debunked; h/t @esiedell!). All of the shenanigans with Huawei. The Trump administration’s ban on China’s state-backed semiconductor manufacturing company. Said company poaching top engineers from TSMC. China blocking US-based Qualcomm’s takeover of NXP due to regulatory concerns. Qualcomm, by the way, tried for years to crack the Chinese market, only to be slapped with the largest fine in China’s corporate history. U.K. being up in arms (no pun intended) about losing their last big semi company to the U.S.

There are a lot of people worried about a war involving TSMC’s assets. The US recently funded TSMC developing a fab in Arizona, but it’s too little, too late - it will be an old node and only for military use.

In comes Jensen Huang, founder and CEO of Nvidia. He is the Elon Musk of the chip industry, with a penchant for black Fendi leather jackets (retail price $4000), and showboating. He’s even trying to name a new law of technology after himself, ala Gordon Moore and Moore’s law.

Jensen is a formidable CEO. He founded the company on his 30th birthday when he noticed the graphics card market (used in video games) was under-served. Nvidia rode the wave of gaming, tried its hand at smartphones, dominates scientific super computing, and was very early to understand that the computing required for AI and blockchain mapped very well to its inherent strengths. He’s also been taking from the “commoditize your complements” playbook. The software stack to program their graphics card, CUDA, is free for developers and is heavily pushed in universities to get young developers interested and contributing to the ecosystem.

This week Jensen announced their intent to acquire ARM from Softbank, for a bargain. The story that got pushed out was heavily centered on AI dominance.

Nvidia’s stock has been on a tear, 20x in the last 5 years. In typical SoftBank fashion, they acquired a $4B stake in Nvidia in 2017, turned around and sold their 5% stake in Dec 2018 for a $400M loss, during a blockchain-implosion-induced blip. Now SoftBank is getting a 7% stake in Nvidia for $21.5B (which as of today has a ~$300B market cap).

Nvidia also recently concluded their acquisition of Israel’s Mellanox, which builds the networking gear required to connect compute components together in the data center.

They will be expanding ARM’s Cambridge UK HQ into an AI center of excellence.

There are no details on how the ARM China snafu was resolved, if it even was. I would not be surprised if they just let this slide as the cost of doing business, in service of the vision.

A U.S. semiconductor company flexing itself in the current political environment.

Need I mention that Huang is of Taiwanese origin and Taiwan, China and the rest of APAC are their three largest geographic markets (note 16 in the 10k). The US is only their 4th largest.

So it appears that Huang is savvily setting Nvidia up for the near term future of geopolitics where every big player seeks to nationalize the internet and localizes data centers. Intel’s fate is too closely tied to that of the U.S. And Nvidia, by virtue of playing in the scientific super computing market, is used to playing politics.

My pessimistic theory is that we go to war with China over TSMC in the next 5 years. My optimistic theory is that Jensen Huang will diffuse these tensions and be awarded a Nobel Peace Prize.

Of course, the ARM deal is going to take 18+ months to bear out, and people are already complaining. Regulatory approval is uncertain. Further, ARM has been a company that has perpetually failed to live up to lofty visions - enough so Masa gave up on it in less than 4 years.

Jensen is a fantastic executor. It will be interesting to see if this leads to a loss of focus at Nvidia. I certainly hope not, for my husband’s sake. Nvidia is the best managed company he has ever worked for.

Colette Kress, Nvidia’s CFO is the MVP in all of their developments the last few years BTW. I’ve built up so much admiration for her over the years. Her responses/demeanor during analyst calls is just brilliant. Her training ground was Cisco’s legendary M&A operations team.

(Edited on 9/20 at 11pm PT to correct a couple errors, add links and pictures. Obligatory: I am not a fiduciary and these are just my opinions. I titled this post Q-semicon after all.)

Update: Part 2 is here!