The Great Semiconductor Shift

First off, we will do this week’s chat at our usual time — (Wednesday, September 13) at 3pm ET in the TradingWithCody.com Chat Room or you can just email us at support@tradingwithcody.com.

The world is in the midst of a multi-trillion-dollar shift in the semiconductor industry. The Semiconductor Revolution (which Cody has discussed for years) has enabled most of the technological progress we have witnessed over the last 50 years. More powerful semiconductors give us more computing power which allows us to complete ever-challenging and more complex tasks. The advances we are seeing in artificial intelligence right now, from large language models to self-driving cars, are only possible because companies like Nvidia and Tesla have developed chips with enough computing power to process and analyze the massive amounts of data needed for AI to work.

The largest tech companies in the world are also some of the largest, and fastest-growing, purchasers of semiconductors. Companies like Apple (AAPL), Google (GOOG), Microsoft (MSFT), and Meta (META) purchase billions of dollars of chips each year to run phones, PCs, servers, etc. Big tech companies purchase chips from a huge swath of semiconductor designers that make chips for different applications. You probably know many of these companies already and there is no need to include an exhaustive list here (as it is VERY long), but we are talking about companies like Intel (INTC), Nvidia (NVDA), Broadcom (AVGO), Qualcomm (QCOM), and Advanced Micro Devices (AMD). Nearly all of these chip designers are “fabless,” other than Intel, Samsung, and a handful of smaller names, meaning they outsource the manufacturing of the chips to a third-party foundry, like Taiwan Semiconductor (TSM) or Global Foundries (GFS).

So here is the multi-trillion dollar shift: big tech and the other major semiconductor buyers are currently designing more and more of their own semiconductors in-house, rather than pad high margins at the fabless semis. Why? The big tech companies have the scale and capability to design application-specific chips, at lower costs, that also significantly improve the quality of their products when compared to the same product built around the generic off-the-shelf chips designed by third parties.

This trend has already started and will only continue to accelerate. For example, nearly all of Apple’s products come with chips designed by Apple in California (they never let you forget that). Apple cut Intel and its x86 architecture from the Macs starting in 2021. Apple has also tried to cut Qualcomm’s Snapdragon modem from the iPhone, but just this week signed a new contract with Qualcomm to supply the iPhone with 5G modems through 2026 (at least in part). Tesla also designed its own CPU to control its cars as well as its own AI chip, known as the D1, which powers the Dojo Supercomputer (more info here).

A lot of the other major tech companies have or currently are in the midst of designing more of their own chips:

Google Has Developed Its Own Data Center Server Chips (published February 14, 2023)

Meta bets big on AI with custom chips — and a supercomputer (published May 18, 2023)

Amazon is making its own chips to offer generative AI on AWS (published August 15, 2023)

If Microsoft is looking into its own ARM chips that could set off alarms at Intel and AMD (published May 02, 2023)

Microsoft developing its own AI chip (published April 18, 2023)

As this trend continues to accelerate, we are concerned that many, if not most of the fabless chip designers may find themselves losing their biggest customers in the years to come.

Continuing with this line of thinking, we wondered to ourselves: What value do the fabless semis offer to their customers? We see four key attributes of the fabless semis that encompass their value proposition:

1. Intellectual Property
2. Software/Integration
3. Talent/Know-How
4. Distribution

The reason that the big tech companies are starting to replace third-party chips with in-house silicon is that the relative value provided by the fabless semis is diminishing as big tech continues to grow in scale and capability. Apple clearly has enough talented engineers and capital to design chips itself. Critically, the big tech companies are also one step closer to the end consumer. It can also design the software needed to integrate its own chips with its existing hardware and operating system(s). Further, the big tech companies do not rely on the semiconductor companies to get their products to market. The same cannot be said of the semis.

But as the Apple/Qualcomm deal makes abundantly clear, switching to in-house chip design is in no way a walk in the park. Despite a clear indication from Apple that it wants to stop purchasing Qualcomm modems (and QCOM’s apparent belief in the same), Apple this week nevertheless re-upped its contract with Qualcomm for three more years. Thinking about what is preventing Apple from leaving Qualcomm behind led us to identify the factors that go into the decision to design chips in-house versus using a third party. In our view, there are at least five considerations that any chip buyer analyzes when making this decision:

1. Scale
2. Capability
3. Specificity
4. Efficiency/Power Consumption
5. Cost

Let’s discuss each in turn. First scale, a chip buyer needs to consume enough chips to justify the cost of hiring engineers and dedicating resources to chip design. Product improvements alone may not be enough to justify the added cost of hiring your own chip designers if you do not have adequate scale.

Capability is next. The big tech companies are in a prime position to cut out the chip designers because they already have the talent, IP and capital required to design a chip. On the other hand, less sophisticated chip buyers like the traditional automakers (Ford (F), GM, etc.) probably do not have the capability to design their own chips. However, we think the ratio of sophisticated/capable chip buyers to incapable buyers is higher than ever before, thus leaving more and more customers assessing the possibility of bringing chip design in house.

Next is specificity. The specificity analysis takes place at the application level, rather than at the company level. Standardized chips do not offer equal utility across applications. A chip selected to run a car’s operating system is not typically going to be the same chip selected for a phone, even if they have equivalent computing power, because the application and operating environment in the two scenarios are so radically different. Tesla demonstrated this when it opted to design its own self-driving chip, rather than rely on existing technology from a third-party chipmaker.

Efficiency and power consumption are the next factors. Going back to Tesla, it did not need a super-tiny, powerful, and power-efficient chip, because it is not limited by the same constraints faced by Apple or Dell when selecting chips for phones or PCs. Apple and Dell, on the other hand, are both looking to optimize battery life and space in the much smaller hardware that houses the chips, so they focus much more attention on size and efficiency than say Tesla does in the car chip it designed.

The last is cost. Apple reportedly saved billions of dollars by switching away from Intel CPUs in its computers. Tesla can make its D1 chip for 1/6th of the price of Nvidia’s H100. It’s the same old story of cutting out the middleman. Why would Tesla, Apple, Google, or anyone else pay chipmakers 50-70% gross margins if they can justify doing the design themselves?

So why does this multi-trillion-dollar shift in the semiconductor industry matter? We think that the market is largely overlooking the potential risk to the scores of multi-billion dollar fabless semis (with trillions of dollars in collective market cap) that are at risk of losing 40%-60% of their revenue as their customers decide to start designing their own chips. Just think what would happen to Intel and AMD if Microsoft designed its own CPUs that shipped in all of future PCs? We are not saying that this scenario will play out overnight, but these companies (MSFT, AAPL, GOOG, etc.) are already taking major steps in that direction and yet most of the semiconductor companies have not even acknowledged this shift in demand, much less altered their company strategy. This trend should only accelerate as we believe many, if not most of the major chip buyers today satisfy the four criteria above and have the incentive to move silicon design in-house.

We think this multi-trillion-dollar shift in the chip industry further highlights the strategic importance of the foundry businesses. Regardless of who designs the chips, somebody has to actually fabricate them. However, building fabs is extremely expensive, and there are no major companies investing in fabs other than TSMC and Intel. In our view, the lack of fab investment from the remainder of the semiconductor industry is representative of the misdirected focus and the lack of recognition of the trend discussed herein.

Intel’s old CPU business has been on the front end of this trend, with its chips (the demand for which was red hot for decades) being designed out of many applications (e.g. the Mac). We think Intel learned this lesson the hard way, but perhaps fortunately for us, Intel learned it early enough to make a strategic shift before the rest of the market wakes up to the trend we are discussing. By the time AMD, AVGO, QCOM, etc. recognize that their customers are doing everything they can to get away from their chips, it may be too late for them to shift strategies to building fabs. This process takes time and A LOT of capex. TSMC and Intel will each spend about $30 billion this fiscal year on capital expenditures. For reference, this combined spend ($60bb) is roughly equivalent to the entire market cap of KLA (KLAC), the 15th most valuable semiconductor company in the world.

There is obviously a lot to unpack here and we intend to spend a lot amount of time researching and analyzing this trend in the coming months. But we think there is a multi-trillion dollar sea change happening in the semiconductor industry right before our eyes that the markets have largely ignored, and there will be a lot of money to be made and lost in the semiconductor stocks in the coming years. More to come.

Don’t forget, we will do this week’s chat at our usual time — (Wednesday, September 13) at 3pm ET in the TradingWithCody.com Chat Room or you can just email us at support@tradingwithcody.com.