TSMC's Largest Customer Accounts for 26 Percent of Revenues

[TheLostSwede]

You're not going to get an award for guessing who TSMC's biggest customer is, but based on details in TSMC's latest earnings report, its biggest customer stands for no less than 26 percent of TSMC's total revenue. That's up a whole percentage in 2021 over 2020 and as you most likely have already guessed, that company should be Apple. TSMC doesn't, for obvious reasons, reveal who their customers are, but it's no secret that Apple is spending a lot of money with the company. TSMC had a consolidated revenue of NT$1.587 trillion (US$55.73 billion) in 2021, or up 18.53 percent from 2020. The second largest source of revenue for TSMC might surprise some, at least based on the kind of information that the usual analysts tend to claim in their reports.

Although second place in terms of revenue only accounts for another 10 percent of TSMC's total revenue, we're still looking at some serious money here. However, as both Qualcomm and NVIDIA departed for Samsung in 2021, second place is said to be taken by AMD, which might not have been everyone's first guess. Unsurprisingly, 64 percent of TSMC's revenue is coming from companies in the USA, with Taiwan being the second largest source of revenue at 12.8 percent. As far as the PRC is concerned, revenue is said to be down by 29.6 percent and only makes up 10.3 percent of TSMC's revenues for 2021. This is largely due to the US sanctions against Huawei, according to the Taipei Times. The 7 nm node is still the big money maker for TSMC, which pulled in over NT$440 billion, followed by the 5 nm node at over NT$262 billion. However, the 5 nm node revenue grew by 188 percent in 2021, while the 7 nm node only had a revenue growth of 11.5 percent.



View at TechPowerUp Main Site | Source

 

[DeathtoGnomes]

is this breakdown from Taipei times? Its an interesting piece regardless.

 

[TheLostSwede]

is this breakdown from Taipei times? Its an interesting piece regardless.

Yes, link at the bottom of the story on the front page, as always.
Couldn't locate the full year report on TSMCs website.

 

[Daven]

AMD being the second largest and presumably growing customer makes the possibility of Intel trying to corner third party fab capacity a little less likely.

Also its quite interesting that 7 nm and lower nodes only make up 50% of TSMC revenue. Less dense nodes are still important.

 

[Steevo]

AMD being the second largest and presumably growing customer makes the possibility of Intel trying to corner third party fab capacity a little less likely.

Also its quite interesting that 7 nm and lower nodes only make up 50% of TSMC revenue. Less dense nodes are still important.

Less dense nodes become hardened for industrial/automotive/space use. -40F its important that the microprocessor in a engine controller works, same at +240F, or further development to -390F for space use etc....

 

[mechtech]

GF are probably kicking themselves for not staying cutting edge and intel for not staying cutting edge and marketing their foundries.

looks more profitable now just to etch silicon then be an actual cpu company.
(yes over exaggerating but still)

I wonder what revenue was from all foundries combined??

 

[TheLostSwede]

GF are probably kicking themselves for not staying cutting edge and intel for not staying cutting edge and marketing their foundries.

looks more profitable now just to etch silicon then be an actual cpu company.
(yes over exaggerating but still)

I wonder what revenue was from all foundries combined??

As it happens, the answer has been provided for you.

Top 10 Foundries Post Record 4Q21 Performance for 10th Consecutive Quarter at US$29.55B, Says TrendForce

The output value of the world's top 10 foundries in 4Q21 reached US$29.55 billion, or 8.3% growth QoQ, according to TrendForce's research. This is due to the interaction of two major factors. One is limited growth in overall production capacity. At present, the shortage of certain components for...

www.techpowerup.com

 

[tajoh111]

AMD being the second largest and presumably growing customer makes the possibility of Intel trying to corner third party fab capacity a little less likely.

Also its quite interesting that 7 nm and lower nodes only make up 50% of TSMC revenue. Less dense nodes are still important.

I would not be so sure about that. Nvidia does not have the revenue or spending of Intel, and they are spending about 7 billion a year on 5nm at TSMC which is probably just as much if not more than AMD will spend on 5nm considering their 7nm obligations and current spending. Intel could easily spend this and with Qualcomm coming back into the mix, AMD might not have the cashflow and revenue to compete considering their current obligations. AMD consoles obligations make them guaranteed and predictable income but it is takes a huge portion of their cashflow to do it.

Consoles are a lower margin product and represents a huge portion of AMD's spending. For example if 2.5 to 3 billion is spent annually to pay for console chips which is not a stretch considering consoles take up 25 millions chips in a year( assuming a $120 cost to make), it would mean AMD has 4 to 5 billion tops to spend on 5nm. AMD is not in a position yet where it can double it's TSMC spending to keep up its current position. Intel and Nvidia are better equipped because of their margins, cashflow and revenue(less so for Nvidia).

Consoles were nice for AMD to have when they were broke, but when AMD had competitive or superior products, they have become an anchor for revenue or growth. If were not in the console business or just servicing one of the console makers, imagine how many more ryzen 3 or RDNA 2 chips they could have sold at higher margins.

 

[noel_fs]

@TheLostSwede writting is very convoluted and confusing, atleast for my adhd

 

[Daven]

I would not be so sure about that. Nvidia does not have the revenue or spending of Intel, and they are spending about 7 billion a year on 5nm at TSMC which is probably just as much if not more than AMD will spend on 5nm considering their 7nm obligations and current spending. Intel could easily spend this and with Qualcomm coming back into the mix, AMD might not have the cashflow and revenue to compete considering their current obligations. AMD consoles obligations make them guaranteed and predictable income but it is takes a huge portion of their cashflow to do it.

Consoles are a lower margin product and represents a huge portion of AMD's spending. For example if 2.5 to 3 billion is spent annually to pay for console chips which is not a stretch considering consoles take up 25 millions chips in a year( assuming a $120 cost to make), it would mean AMD has 4 to 5 billion tops to spend on 5nm. AMD is not in a position yet where it can double it's TSMC spending to keep up its current position. Intel and Nvidia are better equipped because of their margins, cashflow and revenue(less so for Nvidia).

Consoles were nice for AMD to have when they were broke, but when AMD had competitive or superior products, they have become an anchor for revenue or growth. If were not in the console business or just servicing one of the console makers, imagine how many more ryzen 3 or RDNA 2 chips they could have sold at higher margins.

If I was a betting man (and I am since I have AMD stock) smart money is on Epyc and Instinct. These product lines probably make the lionshare of fab capacity as well as margins for AMD. These also happen to be the markets techpowerup readers have the least knowledge. Gaming is great. I love to game. But the computing world is dominated by enterprise acquisitions.

And although I am not sure, it is possible that MS and Sony are TSMC customers for console chips since technically the silicon belongs to them. At the very least, AMD probably charged an upfront NRE in addition to fee per chip.

 

[DrCR]

@TheLostSwede writting is very convoluted and confusing, atleast for my adhd

Were you intending to be ironic*?

*Now that we are in the 20s, hopefully it’s safe to use that word on the internet.

 

[Wirko]

Also its quite interesting that 7 nm and lower nodes only make up 50% of TSMC revenue. Less dense nodes are still important.

Here's a breakdown of TSMC's revenue by node, actually all nodes are growing:

https://twitter.com/i/web/status/1481652648155643904

The 28 nm node is said to be the last one that's cheap (oh well) to design for - that has someting to do with the switch from planar FETs to FinFETs. An interesting analysis was published two years ago, I can't seem to find anything more recent, but it clearly shows that the costs of developing a chip are rising sharply with each new node:

Figure 1. Chip Design and Manufacturing Cost under Different Process...

Download scientific diagram | Chip Design and Manufacturing Cost under Different Process Nodes: Data Source from IBS [2]. from publication: Chiplet Heterogeneous Integration Technology—Status and Challenges | As a heterogeneous integration technology, the chiplet-based design technology...

www.researchgate.net

It's still best and safest to choose 28 nm or even some larger node if you're not designing a chip that will sell in tens of millions.

GF are probably kicking themselves for not staying cutting edge

Maybe. But they haven't stopped investing, and the path they have chosen can still be very profitable for several years to come - RF, analog, power, automotive, gallium nitride, government contracts. After that, they can buy used 7 nm equipment from Samsung. TPU alone reported many times what GF is up to:

TechPowerUp

www.techpowerup.com

 

[timta2]

@TheLostSwede writting is very convoluted and confusing, atleast for my adhd

What specifically? It looks good to me and I'm a native English speaker.

 

[tajoh111]

If I was a betting man (and I am since I have AMD stock) smart money is on Epyc and Instinct. These product lines probably make the lionshare of fab capacity as well as margins for AMD. These also happen to be the markets techpowerup readers have the least knowledge. Gaming is great. I love to game. But the computing world is dominated by enterprise acquisitions.

And although I am not sure, it is possible that MS and Sony are TSMC customers for console chips since technically the silicon belongs to them. At the very least, AMD probably charged an upfront NRE in addition to fee per chip.

That is wishful thinking thinking consoles chips are not eating into AMD wafer supply which is not a surprised since you have a stake in AMD being profitable. However look at the past and AMD have always included their console chip in their revenue stream. It was what kept them afloat during the worst of times.

If AMD making a set amount per chip while SONY AND MS made the chips, that would be a licensing deal and AMD would not nearly have as much of an advantage in obtaining the console chip contract. That is Nvidia would be happy to give the console maker a contract since it would not eat into their wafer supply and it would help them dominate the market further. The reason Nvidia did not pursue it because it is a low margin revenue stream. Licensing technology is a not a low margin revenue stream. It's pure profit.

If Epyc and Instinct were actually taking the lionshare of Fab capacity, their margins would be much much higher. Have you looked at the price of EPYC and Instinct lines? How big are the chips for EPIC and Instinct for their price? AMD's margins are the lowest of the three and this would not be the case if their professional lines were taking up most of their manufacturing ability.

AMD likely has to service those console contract or a penalty will happen. It's why in terms of output, console system have actually been selling at a higher rate than usual(compared to PS4 and xbox one) and supply is pretty healthy considering the shortage of chips. However Ryzen 3 and RDNA2(particularly this one), have been generally short on supply. Look at the size of a console chip and the shear volume of them being sold to console makers and it should be pretty clear AMD should be manufacturing more RDNA 3 and Ryzen 3 chips and manufacturing less console chips but they can't.

 

[TheLostSwede]

Here's a breakdown of TSMC's revenue by node, actually all nodes are growing:

https://twitter.com/i/web/status/1481652648155643904

The 28 nm node is said to be the last one that's cheap (oh well) to design for - that has someting to do with the switch from planar FETs to FinFETs. An interesting analysis was published two years ago, I can't seem to find anything more recent, but it clearly shows that the costs of developing a chip are rising sharply with each new node:

Figure 1. Chip Design and Manufacturing Cost under Different Process...

Download scientific diagram | Chip Design and Manufacturing Cost under Different Process Nodes: Data Source from IBS [2]. from publication: Chiplet Heterogeneous Integration Technology—Status and Challenges | As a heterogeneous integration technology, the chiplet-based design technology...

www.researchgate.net

It's still best and safest to choose 28 nm or even some larger node if you're not designing a chip that will sell in tens of millions.


Maybe. But they haven't stopped investing, and the path they have chosen can still be very profitable for several years to come - RF, analog, power, automotive, gallium nitride, government contracts. After that, they can buy used 7 nm equipment from Samsung. TPU alone reported many times what GF is up to:

TechPowerUp

www.techpowerup.com

You're not wrong.

“The average cost of designing a 28nm chip is $40 million,” said Handel Jones, CEO of IBS. “By comparison, the cost of designing a 7nm chip is $217 million, and the cost of designing a 5nm device is $416 million. A 3nm design will cost up to $590 million.”

Transistors Reach Tipping Point At 3nm

Nanosheets are likeliest option throughout this decade, with CFETs and other exotic structures possible after that.

semiengineering.com

What specifically? It looks good to me and I'm a native English speaker.

There was a couple of sloppy bits in the second paragraph, that I fixed for readability.

 

[watzupken]

GF are probably kicking themselves for not staying cutting edge and intel for not staying cutting edge and marketing their foundries.

looks more profitable now just to etch silicon then be an actual cpu company.
(yes over exaggerating but still)

I wonder what revenue was from all foundries combined??

I don’t believe any foundry is making losses at this point in time. Not producing cutting edge nodes doesn’t mean that they will lose money because a lot of electronics/ chips are still produced on older nodes, or rather, no need for cutting edge nodes. While the foundry may not earn as much as the likes of TSMC or Samsung’s cutting edge nodes, there is also substantial upfront cost involved to ready and deploy cutting edge nodes. In fact, there’s news on GF also on 100% capacity over the last 2 years and with aggressive plans to expand. Which says a lot about their profit.

 

[Vayra86]

Less dense nodes become hardened for industrial/automotive/space use. -40F its important that the microprocessor in a engine controller works, same at +240F, or further development to -390F for space use etc....

But is that the case for these nodes already? I don't think so. They serve a much better purpose in mass production still and they probably will all the way 'back up' to 22~28nm. We saw examples even in CPUs where CCX es were on smaller process than other chiplets in Zen.

The stuff used in space is decades old.