TSMC recently released their Q4 2020 quarterly
financial report. The results were quite good. TSMC has become one of the world's most profitable
companies. In 2020, the company generated net income of something in the neighborhood of $17.5
billion (depending on the USD-TWD exchange rate). TSMC generated more 2020 profit than Wal-mart
or Tencent and roughly as much as Facebook. Its products make 53% gross margin, which matches
Intel's reported 53% gross margin in Q3 2020. Though Intel has a lot of moving parts and
that probably brings things down a bit. But the big surprise was not the profit but TSMC's
company guidance for Q1 2021 capital expenditure. Capital expenditure represents expenses
spent on expanding overall capacity and upgrading manufacturing
technologies. This number is a long term leading indicator of where
TSMC believes the market is to be. In 2020, TSMC spent $17.2
billion in capital expenditures. This is a billion over the $15-16 billion
range that they had projected back in 2019. They had raised it midway through the year.
The $17 billion was a record for the company. Analysts had expected the company to announce
$20 billion of capital expenditure for 2021. It would roll over the $17 billion from 2020
and add the capital for the Arizona plant. TSMC had earlier announced in November
2020 that they were going to build a small N5 fab in Arizona. The board
allocated some $3.5 billion for this. Well, this $20 billion in retrospect
was far too little. For 2021, the company announced that that they will spend
something in the range of $25 to 28 billion. A lot of words have been spilled on the internet,
trying to figure out why the jump in capital expenditure. Right now I am just trying to
grapple with the sheer size of the number. $25 billion, the lower end of that number,
would represent as much revenue as Mondelez, the Oreos maker and number 117
on the Fortune 500. It would be more than Tesla's 2019 fiscal year
revenues ($24.5 billion). $28 billion would be the size of Starbucks ($26.5
billion) or ViacomCBS ($27.8 billion). In terms of sheer numbers, TSMC’s budget
rivals capital expenditures from tech giants and telecoms. That includes
Verizon ($17.9 billion) and Google ($25 billion). Anytime you’re building big
data centers and installing new networks, you are spending billions of dollars on building
up things and assets. Oil and energy companies like Exxon ($25 billion) and Royal Dutch
Shell ($20-29 billion) are also peers. Amazon has a demonstrably
higher budget with $32 billion. But Amazon is also scaling up immensely as it
tries to dominate the e-commerce and cloud spaces. For me, I think you don't spend $25-28 billion just for one reason alone. There
must be a multitude of reasons. The first has to do with demand from
the unprecedented economic environment. The first wave led to everyone shutting
down and the economy crashing hard - but then it came roaring back faster
than people thought was possible. At the same time, the work from home
economy came back harder than ever after the summer ended and
the pandemic surged. You need chips to fuel all those cloud software
services connecting today's companies. Another source of demand comes from
your car. On the earnings call, TSMC mentioned the automotive industry, where
revenue surged 27% from the previous year. Your car uses a surprisingly high
number of chips to help it operate. The Wall Street Journal recently ran an
article noticing that a chip shortage in the automotive industry has caused some
factories to wait for supply to replenish. Automotive, financial stimulus, a new
generation of video game consoles, lockdowns from the pandemic's second surge.
In addition, we have chip demand stemming from the new crypto surge and the ongoing 5G
transition. It's all insanity and craziness. The second big reason to really push
hard has to do with competition. TSMC is dealing with a challenger unlike
few others that it has gone up against. Samsung Foundry, part of Samsung Electronics is making a real push into the market -
capturing reliable business as a second supplier. The Korean giant is estimated to spend some
$26 billion in capital expenditures for its own semiconductor business in 2020. There
is the caveat that Samsung is also the world's dominant memory maker and it is assumed that
a vast majority of that capital is going into reinforcing that business. Not all of that
can be leveraged for making logic chips. But that is a lot of money
regardless. It needs a response. But Samsung is really swinging for the
fences. For its cutting edge 5 nm process, the company has collected customers like Ambarella
(the image sensor maker and supplier to Hikvision) and Qualcomm. They have a big Austin semiconductor
foundry and are considering a $10 billion Texas investment, similar to Taiwan's Arizona
facility, so to win more American clients. And then there is the anticipation that
Samsung is seeking to implement a brand new semiconductor structure for its next
generation 3 nanometer node; something called "Gate All Around". TSMC's N3 will not be
using this structure, and for this reason Samsung has been cited as seizing the technology crown
upon the release of that 3 nanometer process. Well, since we have already touched on the
topic, let us talk about N3 and beyond. TSMC's sexiest, most cutting edge process
is N5 - the one that Apple calls in its marketing 5 nanometers. This process entered
high volume manufacturing in 2Q 2020 at Fab 18 in Tainan. Through 2020, it was majority
booked by Apple for its A14 and M1 chips. I talked about this in another video. But
to match Apple's annual iPhone cadence, TSMC has a half-step strategy. They debut a new
variant of their node every year. For instance, N5 for the A14 this year. Next year, they will make refinements and
release a "plus" version. A15 will be made on that N5+. The year after that,
N5+ then evolves into an N4 process, which means further refinements while
maintaining backwards compatibility. Apple moves on to the next major
step - an entirely new node, N3. This follows what TSMC did with N7, which
evolved into N7+ and then N6 after that. It allows customers to deliver improved product
year over year without much additional work. As mentioned, the next big jump
from N5 is N3, colloquially called 3 nanometers. And like I mentioned
earlier, it will stay with the same FinFet device architecture that helped
TSMC deliver the last few generations. I have seen some rumors floating
around that N3 was going to be delayed. There was a Digitimes report that said that N3
was being particularly challenging for both TSMC and Samsung. And that leaves some risk that
the Apple A16 might not be fabbed on N3. If that is the case, the backup plan would
probably be to use the evolved N4 node. But during the conference call,
management says that N3 is on track. The Digitimes report might be wrong. It
likely is considering their track record. But it can also be that TSMC is trying to make
up for the lost time with a dump truck of money. TSMC hasn’t missed a deadline for
a node delivery in many years. So they are pushing super hard to hit that 2022
high volume deadline for the iPhone launch. As for what happens after N3, the future is
hazy. At the TSMC Technology Symposium in September 2020, the company gave a brief
look at the future roadmap beyond N3. N2 appears to be the next big node.
The company appears to be in the early stages of a massive N2 Fab in Hsinchu.
That is going to cost a lot of money. A lot of financial analysts were watching this.
Intel CEO Bob Swan had said that the company would look at the possibility of outsourcing its
high end logic chips to a third party foundry like TSMC or Samsung. Such a decision would be
announced at the Q4 earnings call in Jan 2021. There were a lot of smoke signals coming up
that Intel would be indeed signing on to such an outsourcing agreement. The stock market
ran up TSMC's stock price in anticipation of it happening. Bloomberg ran an article saying that
TSMC and Intel had been in talks, with no result. Then Bob Swan stepped down and new CEO Pat
Gelsinger took up the seat. Intel's first CTO, Gelsinger has an engineering background and in the Q4 call emphasized Intel's commitment
to its engineering and manufacturing: > “I am confident that the majority of
our 2023 products will be manufactured internally ... At the same time, given the
breadth of our portfolio, it’s likely that we will expand our use of external foundries
for certain technologies and products.” The stock fell 9%. An interesting response by the
market. There is a real argument for semiconductor foundry work to be done on American soil. One can
say that the market is ignoring those national security arguments and being short term. But
one can also care for those arguments and just instead be saying they don't think Intel
should be the company leading that charge. Putting those questions aside, the Intel and
TSMC relationship is going to put itself on hold for a little bit as management reviews
the status of their 7 nanometers process (roughly equivalent to TSMC's N5). Tim Culpan for Bloomberg wrote an
analysis of the results that I liked. In it, he cites one reason why TSMC would move
so aggressively with the capital expenditure would be to corner the market
and lock in supplier agreements. Tactics like this would be taking a page out of
the Apple supply chain handbook. Apple leverages its cash reserves and size to secure supply for
its products ahead of practically everyone else. That is why you were able to buy the new iPhones
this holiday season with decent wait times while other consumer electronics companies
like Sony struggled to fulfill demand. And Apple sells something like 75 million
units a quarter during the holidays. I've touched on TSMC's business tactics in
prior videos I've done on Chartered and SMIC. My feeling coming out of that is that they will not
hesitate to throw their weight around. Like Apple, they are going to try to muscle
their way to the front of the line so to get whatever supplies
they need to fulfill demand. And based on supplier ASML’s recent quarterly
earnings call, TSMC's supply chain right now seems to be in a bit of a dizzy spin. ASML
- TSMC's critical EUV supplier - is only capable of delivering a limited number of said EUV
machines in a single year. And its big customers' projections had moved around a lot throughout
2020 - causing a backup in the supply chain. ASML's CEO and president
Peter Wennik said in the call: > I think our capacity ... in the Netherlands, in
Veldhoven, is to build 50 systems ... Now we have to build those systems out of modules, which
we don't produce. It's in the supply chain ... He then talks about two customers
moving their spend around. > What happened last year in Q2 and Q3 ...
our key foundry customer came back and said, listen, our key customer
for N3 is now blacklisted. So we cannot ship. So we need to adjust
our 2021 outlook for EUV systems. Peter is probably referring to TSMC and
Huawei/HiSilicon. Huawei has long been a leading edge customer alongside Apple. They got banned
from using EUV tech by the Trump administration. > Which was followed by another customer and said, well, we're going to delay the roadmap ...
pushed back one year, which actually led to a situation where we actually reduced
the number of planned system 2021 for EUV This customer is probably Intel, and
it matches up with how Intel in July 2020 announced that their 7 nm process
(which uses EUV) got delayed 6 months. But at some point between Q2 2020 and Q4
2020, TSMC and the rest of the foundry industry realizes that they are way
under capacity. Like I said earlier, astounding demand in everything that
has to do with semiconductor chips. So TSMC and I presume other foundries
in the industry go back to ASML to put those orders back on the menu. But ASML
can't just ramp up on demand because just seven or eight months earlier they told
their entire supply chain to ramp down. Peter did hint that Intel's delay let them
transfer Intel's EUV orders over to TSMC and Samsung. Despite that, my guess is that for the
next few quarters (per the CEO there is about 20 months between module production
and EUV tool installation), EUV machine supply is going to be super tight. Project this out to the rest of the
foundry supply chain - it's not just ASML, you know - and you get the feeling of a
massive oil tanker trying to turn on a dime. You can get things done right, done fast, or
done cheap. Pick two out of the three. TSMC is apparently picking done right and done fast.
Damn the cost, they're going to buy everything. For what it is worth, Wennik warns us about the
folly of using those TSMC capital expenditures to project ASML's 2021 revenue. This is
due to the supply issues mentioned above: > Yes. TSMC gave a range, $25 billion to $28
billion. Hey, great. We plan our business based on what they ask us. And as you know,
TSMC has been asking us in 2020, on several occasions to ship very different numbers for
2021 ... So I don't think you can draw any direct conclusion from the TSMC capex numbers.
Directionally, yes, but not in absolute terms. I am reminded of a story
published back in April 2001 when Morris Chang still ran the
company as its founder and CEO. In it, TSMC was a few generations behind
Intel but doing their best to catch up. 2001 is a year after the tech bubble crash
and the electronics space in general was suffering. But regardless, TSMC is pushing
ahead with investments. This means not only building more foundries but also investing
in being able to provide better services to their customers. Chang saw an opportunity
and poured in the money to service it. I get the sense that TSMC is a company that likes
to take big swings at things. Unlike Chartered and SMIC, they have the power of the purse and
they leverage it like few other companies can. A famous investor once said that their
entire job is to stand at the batter's box and wait for a fat pitch that they can swing at
as hard as they can. Hit one big home run and it does not matter that you missed a bunch of little
things. I think TSMC thinks a fat pitch is coming.
