Table of Contents
After my last article on 7nm’s inability to kill off “dinosaur” tech over the last 8 quarters, I wanted to extrapolate on some of the Taiwan Semiconductor Manufacturing Co. (TSM) revenue data in order to get a bigger picture view. This will be most pertinent for an investor looking to analyze the potential revenue trends for the chipmakers that are TSMC’s direct customers and/or their customers’ competitors, such as Intel (INTC), Advanced Micro Devices (AMD), Nvidia (NVDA), Broadcom (AVGO), Qualcomm (QCOM), Micron (MU) and others.
It will especially help in analyzing the potential in Intel’s Internet of Things Group (IOTG), and also provide some supplemental context on how much the 7nm delay is likely to influence Intel’s long-term profitability and revenue prospects in the segments which most rely on cutting edge technology nodes for semiconductors (like high performance computing) moving forward.
My last article showed the movements in TSMC’s revenue mix for 7nm – 0.25um+ (250nm) over the last 8 quarters. This article will extend a similar analysis over 5 years to fiscal year 2016, which should prove more useful for investors looking to project long-term cash flows for valuation purposes.
While technology is obviously so unpredictable in the future, having real (data based) context on how previous technology nodes have been cannibalized over the previous 5 years should help us quantify a probable range of possibilities over the next 5 years, as well as provide clarity to how the emergence of IoT (Internet of Things), Automotive Electronics, and DCE (digital consumer electronics) is directing the hardware needs for chipmakers the corresponding silicon.
Remember that I’ve chosen to analyze TSMC in order to provide a sense of Intel’s market, competition, and industry trends because of TSMC’s position as the single dominant player in the foundry business (with 52% global market share). We should also consider that Intel has 15 fabs of their own, which have a greater influence on their own revenue but likely carry less influence on the overall (global) industry.
Estimating Consumer Demand for Semiconductors with TSMC’s Revenue Mix over the Past 5 Years
First, I’ll show how I extrapolated the 8 quarterly table into the 5-year time period. Note that these data was sourced from TSMC’s earnings call slides/presentations hosted on the Seeking Alpha website, but the company didn’t seem to release similar data for every previous quarter – which is why some columns are blank. To mitigate this, I also calculated averages for each year on as much data as was able to be sourced from TSMC’s quarterly earnings presentations.
Those who read the last article presenting demand for TSMC silicon should recall how the percentages distort revenue realities. The most obvious discrepancy is that while 7nm continued to take increased share (from a percentage basis), this didn’t necessarily correspond with decreased revenue for legacy technology nodes, as displayed with this table (calculating estimated nominal revenues from the company provided percentages):
Here’s the data condensed into the annual averages rather than the quarterly figures available online:
And then the corresponding nominal values from these annual averages:
End User Market Data – TSMC; Same Time Period
I also decided to take it a step further and additionally provide application/platform data from the earnings slides to give a decent estimation on the consumer end markets for much of this silicon. This helps understand which markets have been driving any increased demand for TSMC production, as well as which secular trends could possibly drive future growth for not only TSMC but all of the chipmakers: my self proclaimed “3 musketeers” INTC, AMD, NVDA, and other major players that could eventually breach into Intel’s territory.
Again, the percentages may be misleading, so corresponding revenue estimates are below. Note that the company shifted their presentation of consumer end market revenue data away from their traditional 4 metrics (Communication, Industrial/Standard, Computer, Consumer) into a more comprehensive dataset:
- HPC (High performance computing)
- Automotive (electronics)
- DCE (digital consumer electronics)
The problem with the change is that the old 4 metrics are no longer disclosed by the company. So, I had to make estimates for the 4 metrics based on the following slide (from 1Q19):
I combined the slide data above with the quarterly comparisons from 4Q18 and 1Q19 in order to calculate a rough formula to estimate the previous 4 metrics with each new “Revenue by Platform” presentation with the 6 new metrics (which will have to be used from 2019 onwards). The changes over quarters (QoQ) were also based on the slides, as follows:
- Communication = -27% QoQ
- Computer = -31% QoQ
- Consumer = -10% QoQ
- Industrial/Standard = -16% QoQ
According to this slide, Smartphone did see a decrease of -33% quarter over quarter, with HPC at -26% and DCE at +20%. Now that I completed the calculations to convert new metrics to old metrics, I could make rough CAGR estimates for each of the 4 legacy categories, which we can see with these tables (first the nominal value estimates).
This is an average value per quarter, not total revenue for the year:
And then CAGR calculations based on the numbers above, noting that the YOY measurements for the 6 new categories are simply the 1-year change from the average quarterly data per year:
The overall growth in 3 out of the 4 applications has boded very well for chipmakers, with only the Consumer category showing a decline (but only of about $0.21B over the 5-year period). I estimate that the Consumer category is made up of about equal parts of IoT, Automotive, DCE, and Others, with a higher weighting on DCE than the other categories. Since those markets are strong on a YOY basis, it’s possible that there’s a little cannibalization going on which is crimping the Consumer category. Since Consumer was already close to the smallest of the 4 categories and is now firmly in the smallest, it’s not much of a cause for concern for overall chip demand, especially considering the extremely strong strength with everything else.
Now observe how this corresponds with the CAGR changes in Process Node over the same 5-year period, noting that 7nm didn’t start until 2018 and 10nm didn’t start until 2017 (according to TSMC presentations on revenue), so those CAGRs represent a 2y and 3y stretch, respectively.
The results are very much a mixed bag. From this data, it appears that there have been growth drivers for 0.11/0.13 um and to a lesser extent, 0.15/0.18um, which shows the resilience for some of the older technology nodes in their specialized end uses. Custom analog or mixed signal ICs tend to be much more prevalent in these bigger nodes, especially among communications/RF solutions, which could continue as time goes on.
However, that’s not to say that all legacy technology nodes are safe, with significant drops (-5% or more per year) present in these nodes:
- 10nm (almost completely un-used now)
The biggest segments in terms of total revenue in 2016 were the following:
- 16/20nm = $2.2b
- 28nm = $1.9b
- 40/45nm = $1.0b
- 65nm = $0.9b
- 0.15/0.18um = $0.8b
The two most noteworthy nodes to lose significant revenue per year over these last 5 years were 28nm and 65nm, with 0.15/0.18um affording itself the only “major” node to actually gain revenues as a legacy tech.
Like I mentioned in the last article on the revenues trends among technology nodes, it’s likely that the legacy tech won’t die immediate deaths and could maintain longer staying power than is widely believed by many. That breakdown has slowed considerably over the last 8 quarters. Whether the slowdown is a new normal or just a pause on a longer-term trend is something to watch in the upcoming quarters. Also, this could be a TSMC-specific trend rather than an industry wide trend, depending on what’s happening in the other major fabs.
The IoT Catalyst for Intel
We know from TSMC’s data that the YOY growth for IoT silicon is trending for an explosive 2020, with the difference in averages between quarters in 2019 and 2020 recording an astounding 27.3% YOY. Most of IoT was split relatively evenly between TSMC’s traditional segments, and so it has ridden this approximate 8% secular trend in revenue growth since 2016.
The next company to see their IoT segment grow 10%+ per year could conservatively estimate that sort of growth with a continuation of the long-term growth trend plus a few percentage points from grabbing market share, or could see growth double that if this year is any indication of the next 5 years to come.
To see which companies are poised to potentially ride this wave the best requires some more estimates, starting with a rough industry map. Obviously, the term IoT encompasses so many possible components, with new inventions popping up every day. But between the major semiconductor manufacturers and those companies known publicly to focus on IoT, we can try and summarize major parts of the market.
To surmise the Internet of Things Group, we turn to the company’s latest annual report (bolded emphasis mine):
“IOTG develops high-performance compute for targeted verticals and embedded markets. Our customers include retailers, manufacturers, healthcare providers, energy companies, automakers, and governments. We facilitate our customers creating, storing, and processing data generated by connected devices to accelerate business transformations.”
I note the embedded feature as it becomes a critical detail to understanding the overall market when reading through other competitors’ annual reports, as not every company reports relevant revenues the same.
Looking at top competitor (in the Internet of Things Group), Texas Instruments (TXN), we see them describe a potential TAM (Total Addressable Market) as about $18 billion in 2019, with this to say about their Embedded Processing Segment (their other segment being Analog; bolded emphasis mine):
“Our Embedded Processing segment generated $2.94 billion of revenue in 2019. Embedded Processing products are the “brains” of many types of electronic equipment… Our Embedded Processing products are used in many markets, particularly industrial and automotive. Sales of Embedded Processing products generated about 20% of our revenue in 2019. According to external sources, the market for embedded processors was about $18 billion in 2019. Our Embedded Processing segment’s revenue in 2019 was about 16% of this fragmented market, which is among the leaders. We believe we are well positioned to increase our market share over time.”
Texas Instruments breaks down their Embedded Processing Segment into two primary solutions: Connected Microcontrollers and DSPs.
Other companies also report similar end use markets for their IoT related segments as well. One great example is Silicon Labs (SLAB), which defined the following applications for their “Internet of Things Products” business segment, under their Microcontrollers and Wireless Products line:
- Home automation /security
- Industrial automation
- Smart metering
- Smart lighting
- Commercial building automation
- Consumer electronics
- Asset tracking
- Medical instrumentation
Within many of these competitors we should be able to confirm the trends in recent IoT growth as described by TSMC’s revenue earlier in the article, and one more competitor shows an example of exactly that in their latest earnings presentation:
Note the estimated 8-11% CAGR for Industrial & IoT, suggesting that the latest 5-year trend of 8% CAGR for the type of silicon that tends to serve this market might be on the low end of what investors can expect.
IoT Industry Map
Taking all of this research into mind, and making some estimated calculations on how to divide segment revenue and operating income (for the companies which don’t provide a defined IoT group), I’ve come up with the following rough industry map (with this S&P 500 industry map as a guide):
Other more direct competitors (reported by Maxim Integrated Products (MXIM)) include Cirrus Logic (CRUS) and Semtech Corporation (SMTC), but both companies reported total revenues below $1.5B in 2019.
These companies are all worth a watch, especially Monolithic Power Systems (MPWR), the small competitor carrying lots of market optimism for high potential growth. If any sort of additional excitement brews within the industry, it will likely ferment itself into this expensive stock with a P/E over 90 and high P/FCF.
How the major players Analog Devices (ADI) and Maxim Integrated Products manage their merger could also change the tide in the upcoming years and threaten Intel’s position, particularly if other business segments bleed the company and/or reduce its investment plans in IoT, allowing someone like Analog Devices or Texas Instruments to pick up the slack.
Intel Reverse DCF
I think most investors understand by now that Intel is a value stock by many traditional metrics. I’d like to quantify how cheap the company’s current valuation presents based on a reverse DCF.
Intel’s WACC (weighted average cost of capital) the last 4 quarters, to show why I’ll be using a 6% discount rate for this calculation:
Using last year’s figures for free cash flow and EPS, which are conservative because the company has actually earned more in both categories over the last 12 months, the model returns the following implied growth rates:
- FCF per share = $3.79; Implied growth rate = 1.28%
- EPS per share = $4.71; Implied growth rate = -1.81%
I know that the company has extreme pessimism surrounding the competitiveness of its CPUs against the likes of AMD’s and Nvidia’s GPUs, which is something I’ve touched on before.
Obviously, the IoT industry is still young, fragmented, and a small proportion of Intel’s complete operating business.
But, if the growth in IoT, 5G, and cloud computing can spill into areas like IOTG or Mobileye or Intel’s other operating segments, then even a smaller business unit like IOTG could possibly provide enough small growth to justify the low valuations of today – without even considering the possible (valuation) multiple expansion if sentiment does turn around.
The revenue split based on TSMC’s technology nodes shows that legacy tech does tend to decrease (in demand as expressed by TSMC revenue) over a 4-5 year period, but that some tech (like 0.15/0.18um) has staying power even as the rest of the semiconductor world moves to 7nm and beyond. It’s important for Intel to be proactive with maximizing any staying power of old tech, while at the same time limiting the fallout of decreased demand if they can’t produce highly efficiently at the most advanced nodes.
***If you’d like more deep dives into company filings like this one, please do consider Following my profile with email alerts. To see how my picks for the Real Money Portfolio have done, check out: Post-Pandemic Performance***
Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.