Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/chip-globalization-is-over-and-sanctions-work-says-tsmc-founder.17611/page-2
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Chip Globalization Is Over and Sanctions Work, Says TSMC Founder

Agree.

But engineer salaries cost have very little impact to Fab's competitiveness. UMC engineers probably only get 70% of TSMC engineers but gives UMC no advantage.

In my opinion, cost difference came from so many different small factors which ultimately makes huge cost difference.

For example, 3-6 months difference to build a new Fab. 3-6 months difference to improve yield for new nodes. For daily operation, almost all TSMC supplier are 24h stand by to fix the problem. Even the same 24h stand by duty, TSMC's culture (peer pressure) may force you to work extra harder to solve a little problem today instead of wait for tomorrow. Actually former TSMC top executive Konrad Young predicted that those TSMC employees moving from Taiwan to US. Afterworking with local American coworkers, they will find out there are more important stuff in life other than work so the culture will change immediately and will no longer work as hard as they do in Taiwan.

Intel Foundry Services have to dance with many customers at one in the future. Those small factors may result in a even larger cost difference.

Once we consider all the cost factors such as productivity/efficiency, material and equipment cost, salary and benefits, taxes/pension/health insurance, and the availability of the necessary resources, making leading edge semiconductor in US is a very expensive proposition. It's not just 7% to 17% cost difference estimated by Scotten can explain.

Below is my old post about the productivity comparison between TSMC and Intel for 2021. For 2022 the difference between TSMC and Intel is getting even bigger. Intel's 2022 net income dropped to $8.02 billion while TSMC's increased to $34.07 billion.

***

If we measure the employee productivity by net profit, TSMC is significantly stronger than Intel.

2021 TSMC average net profit per employee, based on 65,000 employees and US$21.354 billion net profit for 2021:
US$ 328,523

2021 Intel average net profit per employee, based on total 121,000 employees and US$19.87 billion net profit for 2021:
US$ 164,214

Source: https://semiwiki.com/forum/index.ph...’t-fix-the-talent-bottleneck.15998/post-52515
 
Last edited:
I greatly respect Chang, but I hope he's wrong on free trade. Globalization seems to be on life support with grave consequences in tow, but cooler heads may prevail.

It's increasingly more probably that cool heads will not prevail, and we will have another major war with mass draft, and double digit population loss.

This is an absolutely real possibility for which we need to prepare.

In 1991, nobody would've sided with dying China, or Russia, because they knew they had zero chances militarily, as demonstrated by the Gulf War, and absolutely none economically. The entire notion of Eastern Bloc felt apart. Countries were jumping off the sinking ship one after another.

Now, the situation look eerily like nineteen thirties. Countries openly bet on the enfeebled West losing, and jumping off our ship this time. Because the West chose to wait until the last moment, what could've been a walk in the park subjugation of Germany, and 1-2 more countries in 193X, ended up a huge global conflict with dozens of countries involved by early forties, and being a much harder fight. US can easily win a one front war, possibly a two front war, but winning against 3+ counties without a global coalition is absolutely unreal no matter what we will do.

WW2 did not redefine the world order, but what followed after did.

The great redrawing of map, fall of empires, and redefining of alliances happened after WW2 not because of winners, but exactly because of lack of clear winners.

Similarly, the current conflict may not end with just vanquishing of Russia, or China, but will go on for years after, as more, and more dictators will try their luck, knowing the West would be exhausted by prior conflicts.
 
Is this 10% cost difference you mentioned just for the 12 inches wafer material or the overall manufacturing cost?

Scott did this analysis in 2021:


This has been discussed quite a bit since and it still stands, especially now that the AZ fab has been expanded. Remember, this is a mature high yielding process and not a greenfield fab per say. And this was before the CHIPS Act and the political onshoring surge.

Bottom line: TSMC knows what they are doing, do not believe otherwise even if it comes from 91 yo Morris Chang. He is net worth is North of $2B so he can say anything he wants. Unfortunately his legacy is suffering and will continue to do so until he says something that gets him cancelled completely, my opinion.
 
Globalization isn't over at all, or even in decline. Trade and investment in or from China is declining. Western countries and companies are looking for alternatives to dependencies on China. China does not equal globalization.

I don't really see it. From what I see, I see Western companies not moving away from China, but doubling down on it, and they will continue to do so, until an extremely dramatic regulatory measure will come.

I'm right now witnessing one of Foxconn's clients attempt to move to India, with one dramatic failure, after another. The appetites for the move have certainly cooled down, as much as their premature attempt to move to Vietnam. Any thoughts of complete move to India are off the table for them at least for a few years.

1679165097720.jpeg
 
Last edited:
Is this 10% cost difference you mentioned just for the 12 inches wafer material or the overall manufacturing cost?
Overall cost. Substrate cost has no reason to be any different.

Below is my old post about the productivity comparison between TSMC and Intel for 2021. For 2022 the difference between TSMC and Intel is getting even bigger. Intel's 2022 net income dropped to $8.02 billion while TSMC's increased to $34.07 billion.
I have already pointed out the flaws with this methodology. Namely intel and TSMC are very different companies. It would be like comparing J&J to P&G. They both make chemicals but how they do it and what their worker mix are, is very different. It is also not consistant with your earlier argument that intel was expanding oversees to escape high costs in the US.

Once we consider all the cost factors such as productivity/efficiency, material and equipment cost, salary and benefits, taxes/pension/health insurance, and the availability of the necessary resources, making leading edge semiconductor in US is a very expensive proposition. It's not just 7% to 17% cost difference estimated by Scotten can explain.
As I already stated fabs are so heavily automated that even with a way over exagerated 5x cost there is a marginal impact on wafer costs due to how few employees there are per wafer out. There is NO data or engineering reason to suggest that wafer costs should be 2x or even 1.5x what it is in the ROC. Furthermore the same arguments that apply to the US vs ROC also apply to the ROC vs PRC/SE Asia. Lastly as I have also stated before I totally reject claims that American workers "lazy". Morris Mentioned that in the USA if a tool breaks they will leave it overnight because Americans apparently = Lazy. I do not know of a single semiconductor fab or even chemical plant where this is the case. Literally every process engineer job I have ever applied for always wanted to know if I was okay working nights when it was necessary. If all of the numbers and evidence I have presented is still insufficient then I don't know what to say other than convincing you is a hopeless endeavor.
 
Two thoughts:

1. For US$100K annual salary (including benefits), TSMC and other top tier Taiwanese tech companies will certainly get a very capable and multi-years experience engineer who graduated from a top Taiwanese university.

For US$100K annual salary (including benefits), Intel will have some chances (if Intel is lucky) to recruit an entry level engineer from a top University.

2. Intel engineer's salary or US engineer's salary is generally much higher than Taiwan's. According to some surveys, the number can start with $130K or more.


In Taiwan it's extremely uneven, and compensation mechanics there are unfathomable to Americans.

There, for example, is a very strong desire for a "seniority pay" in the literal meaning of the word. You can be completely unceremoniously asked about your age at the recruitment interview, or your family situation.

Another is title hunting. Nearly the same jobs with different titles can show up to 2x difference for a more prestigious sounding role. Those titles are usually given as rewards for good networking.

The pay of people nearing VP level skyrocket, and can be easily 4-5 times of the most well paid engineer following the normal career trajectory. So management, non-management divide is very visible. A manager expects n-times the pay of his subordinates as something given.
 
Globalization isn't over at all, or even in decline. Trade and investment in or from China is declining. Western countries and companies are looking for alternatives to dependencies on China. China does not equal globalization.
I hope you’re right. You’re definitely a little more optimistic than me. I’m also thinking of:

- Brexit
- Middle East seems to be drifting more towards less trade
- The US seems to embracing populism and appears to be abandoning globalization (Bretton Woods is the reason we have globalization)
- Germany starting to age out of productivity

It seems like Mexico or India may be next to step up as China loses some major partners.. but no guarantees of course.
 
I hope you’re right. You’re definitely a little more optimistic than me. I’m also thinking of:

- Brexit
- Middle East seems to be drifting more towards less trade
- The US seems to embracing populism and appears to be abandoning globalization (Bretton Woods is the reason we have globalization)
- Germany starting to age out of productivity

It seems like Mexico or India may be next to step up as China loses some major partners.. but no guarantees of course.
I didn’t say China would be easy or timely to replace or even augment. China didn’t develop overnight, and won’t be replaced soon. My point is that using low-cost countries for manufacturing will continue with or without China. The stakes and potential profits are high, so globalization will continue and one way or another China’s position will be eroded unless they stop being adversarial.
 
Most of the shrinkage was American companies becoming less pervasive. This logic also is not coherent either. If the US was 2x ROC, then why on earth would Germany, Israel, Ireland, or Japan be equal to the ROC? Also why does TSMC build in the ROC and not in Malaysia or Korea? The labor force is even cheaper there than in the ROC. Additionally the case you haven't noticed intel's AZ fab is way bigger than Ireland and a good bit bigger than Israel. If the US was a cost disaster why would intel's flagship fab be in AZ? California is a different beast, but there is a very good reason why everybody shuttered their CA fabs. As for Micron they make comedity products where price is the number 1 quality. That 7-17% difference in price is what could make the difference between 10% and 30% marketshare. In spite of this they still keep their US fabs open. Why you may ask? Because the cost difference is small enough that even in the hyper competitive DRAM/NAND industry, the cost difference is small enough that building a new fabs in SE Asia to replace their US/JP/ROC fabs wouldn't make any sense. The IRR for this project over continued use of a US fab would be on the time scale of centuries.

TSMC does not equal the whole fab world,Western companies did/are building lots of fabs in SE Asia. I don't remember any SE Asian gov giving huge subsdise to build these fabs
 
Overall cost. Substrate cost has no reason to be any different.


I have already pointed out the flaws with this methodology. Namely intel and TSMC are very different companies. It would be like comparing J&J to P&G. They both make chemicals but how they do it and what their worker mix are, is very different. It is also not consistant with your earlier argument that intel was expanding oversees to escape high costs in the US.


As I already stated fabs are so heavily automated that even with a way over exagerated 5x cost there is a marginal impact on wafer costs due to how few employees there are per wafer out. There is NO data or engineering reason to suggest that wafer costs should be 2x or even 1.5x what it is in the ROC. Furthermore the same arguments that apply to the US vs ROC also apply to the ROC vs PRC/SE Asia. Lastly as I have also stated before I totally reject claims that American workers "lazy". Morris Mentioned that in the USA if a tool breaks they will leave it overnight because Americans apparently = Lazy. I do not know of a single semiconductor fab or even chemical plant where this is the case. Literally every process engineer job I have ever applied for always wanted to know if I was okay working nights when it was necessary. If all of the numbers and evidence I have presented is still insufficient then I don't know what to say other than convincing you is a hopeless endeavor.

I don't know why we need to exclude the cost and productivity comparison between TSMC and Intel. Indeed there are significant difference between them in terms of business model and products/services. But under Intel IDM 2.0 strategy, Intel treats TSMC as a partner, a supplier, and a competitor while TSMC treats Intel as a partner, a customer, and a competitor at the same time. It's very logical and meaningful to understand TSMC's and Intel's cost and the reflection on their productivity and financial statements.

From the comparison people can learn why TSMC and Intel got where they are standing today and the challenge and opportunities they are facing. For the past several years a lot of the discussion on semiconductor industry (for good and bad reasons) are pointing to two semiconductor companies: Intel and TSMC. Why don't we have some discussion about them?

Intel with its IDM business model should have commanded (at least Intel claimed to have) a wider profit margin than a pure play foundry. From design to manufacturing to test/assembly to sales/marketing, a pure play foundry (like TSMC) is only part of the whole semiconductor supply chain. TSMC supposedly should not make more profit than Intel who can add profit at each step from begin to the end.

Using the 2021 data, TSMC has less employees and less revenue but made more total net profit and more net profit per employee than Intel. I wonder why? I don't think it's 7%, 10%, 15%, or 17% TSMC manufacturing cost advantage can explain. That's why I think these numbers are optimistically underestimate the situation. Although if you don't think Intel vs TSMC is a fair comparison, you can pick and choose your own semiconductor "manufacturing" companies to do your own analysis.


2021: TSMC 65,000 employees, revenue US$57.225 billion, net profit US$21.354 billion
TSMC average net profit per employee: US$ 328,523

2021: Intel 121,000 employees, revenue US$79.024 billion, net profit US$19.87 billion
Intel average net profit per employee: US$ 164,214

Net Profit = Revenue - All cost (materials, depreciation, salary, tax, interest, etc.)

My observation is simple: cost will impact productivity and will impact the profit eventually. There are many other indexes or benchmarks that allow us to evaluate a company's competitiveness. I'm looking for your input.
 
In Taiwan it's extremely uneven, and compensation mechanics there are unfathomable to Americans.

There, for example, is a very strong desire for a "seniority pay" in the literal meaning of the word. You can be completely unceremoniously asked about your age at the recruitment interview, or your family situation.

Another is title hunting. Nearly the same jobs with different titles can show up to 2x difference for a more prestigious sounding role. Those titles are usually given as rewards for good networking.

The pay of people nearing VP level skyrocket, and can be easily 4-5 times of the most well paid engineer following the normal career trajectory. So management, non-management divide is very visible. A manager expects n-times the pay of his subordinates as something given.

"The pay of people nearing VP level skyrocket, and can be easily 4-5 times of the most well paid engineer following the normal career trajectory. So management, non-management divide is very visible. A manager expects n-times the pay of his subordinates as something given."

You brought up an interesting aspect regarding the compensation. I checked the 2021 data (2022 data hasn't been released yet):

Intel:
Total revenue: US$79.024 billion
Net Profit: US$19.87 billion
Intel CEO Pat Gelsinger total compensation: US$178.59 million

TSMC:
Total revenue: US$57.225 billion
Net Profit: US$21.354 billion
TSMC CEO C.C. Wei total compensation: US$14.4 million

I'm wondering if Pat Gelsinger or C.C. Wei ever went to their own board of directors to tell them it's not fair to be treated like this :)
 
Micron (all of their R&D in the US and some production in high cost nations like the US and Japan. Significant presence in the cheaper than ROC SE asia)
Employees in 2020: 40k
Net income: $8.7B
Ratio: $218k/person
Also they have less scale, no large Chaebol to back them, and no cheap loans/land grants from the ROK/PRC.

SK Hynix (all of their R&D in the ROK and a large manufacturing presence in the PRC (even cheaper than the ROC))
Employees in 2020: 36,854
net income: $1.9B
Ratio: $50.7k/person

SK vs Micron: Micron is 4.3x more efficient


UMC (mostly ROC with some PRC and JP fabs):
net income: $620M
Employees (2021): 19,741
Ratio 31k/person

GF (leading edge in the west trailing edge in singapore):
Net income Q4: $668M
Employees (2021): 14,600
Ratio $46k/person

GF vs UMC: GF is 43% more efficient.

So once again, there is no evidence that the US is 2x the cost of the ROC.
 
Last edited:
Besides salary/equipment/building cost that many people have discussed, there's one hidden cost/efficiency difference: The vendors' support.
----
>60% global logic chips manufacture (tsmc+UMC) are in a very small region. From Hsinchu (HQ of tsmc) to Tainan (N3/N5 main manufacture site) only takes 1.5h by high speed railway or 3h by car. So vendors can put tremendous resources in Taiwan to support all manufacture running smoothly: Every morning you can see tons of vendors' engineers wait in the lobbies of tsmc's fabs. These levels of vendors' support may only exist in Taiwan.
----
Meanwhile tsmc also aid many Taiwanese local vendors to provide cheaper products/service to lower tsmc's running cost. These Taiwanese vendors are not in the comparable scales as Applied Materials/ASML/TEL... So some of Taiwanese vendors are reluctant to set a branch at Arizona. As tsmc's 2nd Arizona fab is announced, more vendors should be willing to go to Arizona.
 
  • Like
Reactions: VCT
Morris Chang said something interesting, something to the effect, TSMC has 400 dance partners, while Intel dances alone.

TSMC grew big but not inefficient, in part because they have meaningful input from many external partners. Thats a key difference. I think TSMC will overcome any transient cost headwinds and operate successfully in the USA. They will solve the issues just as they always have.

My last thought is, Taiwan is like Hawaii. Islands are inherently expensive. I think eventually TSMC in Arizona will be cheaper than Taiwan.
 
My last thought is, Taiwan is like Hawaii. Islands are inherently expensive. I think eventually TSMC in Arizona will be cheaper than Taiwan.
An interesting analysis I've never heard before. I doubt it, but I won't soon forget your post.
 
TSMC does not equal the whole fab world,Western companies did/are building lots of fabs in SE Asia. I don't remember any SE Asian gov giving huge subsdise to build these fabs
Money is money, regardless of how it is accumulated. High cost countries need subsidies to be competitive. Low-cost countries in Asia provide effective financial equalization by fewer (if any) environmental laws, fewer regulations, lower local taxes, lower wage rates, lower building costs, cheaper healthcare, and often longer generally accepted work week hours (though I'm not sure how really effective this last point is). You don't need subsidies to attract manufacturing with all of these advantages.
 
Micron (all of their R&D in the US and some production in high cost nations like the US and Japan. Significant presence in the cheaper than ROC SE asia)
Employees in 2020: 40k
Net income: $8.7B
Ratio: $218k/person
Also they have less scale, no large Chaebol to back them, and no cheap loans/land grants from the ROK/PRC.

SK Hynix (all of their R&D in the ROK and a large manufacturing presence in the PRC (even cheaper than the ROC))
Employees in 2020: 36,854
net income: $1.9B
Ratio: $50.7k/person

SK vs Micron: Micron is 4.3x more efficient


UMC (mostly ROC with some PRC and JP fabs):
net income: $620M
Employees (2021): 19,741
Ratio 31k/person

GF (leading edge in the west trailing edge in singapore):
Net income Q4: $668M
Employees (2021): 14,600
Ratio $46k/person

GF vs UMC: GF is 43% more efficient.

So once again, there is no evidence that the US is 2x the cost of the ROC.

My thoughts:

1. I haven't verified each of your number but Globalfoundries 2022 number you stated is incorrect. Can you please check and calculate them again?

2. Globalfoundries has been losing money every year except 2022. It's not clear to me their net profit per employee number is a stable and long term situation.

3. All the companies you mentioned in the comparison have significant East or Southeast Asian fabs operations that have brought down their cost and increase their profit. Micron Taiwan fabs have about 10,000 employees, similar to Micron US headcount. Micron Taiwan and Micron Japan are also making advance memory products.

On the other hand, Intel has no fab operations in Southeast Asia. Intel fabs are in high cost locations such as Israel, Ireland, and US. It partially explain why Intel's net profit per employee is only half of TSMC's.

4. Globalfoundries and UMC do not have leading edge products and do not use the expensive EUV tools. Their cost structure and target market are very different from what TSMC and Intel are dealing with.
 
My thoughts:

1. I haven't verified each of your number but Globalfoundries 2022 number you stated is incorrect. Can you please check and calculate them again?

2. Globalfoundries has been losing money every year except 2022. It's not clear to me their net profit per employee number is a stable and long term situation.
That net income number was directly off of their balance sheet for Q4'22. However upon some digging apparently an asset sale happened in Q4. So save the sanity of tracking down UMC's numbers again I will just use GF's Q3
GF:
Net income Q3: $336M
Employees (2021): 14,600
Ratio $23k/person

UMC is 34.7% more efficient.

3. All the companies you mentioned in the comparison have significant East or Southeast Asian fabs operations that have brought down their cost and increase their profit. Micron Taiwan fabs have about 10,000 employees, similar to Micron US headcount. Micron Taiwan and Micron Japan are also making advance memory products.
On the other hand, Intel has no fab operations in Southeast Asia. Intel fabs are in high cost locations such as Israel, Ireland, and US. It partially explain why Intel's net profit per employee is only half of TSMC's.
Yes, however SK only has operations in even cheaper places than the ROC. Meanwhile Micron has a large presence in the US and in JP which are even higher cost than the ROC. If the cost of doing business in the US vs ROC was 2x, then the difference between US and PRC or ROK would be like 3x. If the cost gap was actually this big, you would see this reflected in your metric.

4. Globalfoundries and UMC do not have leading edge products and do not use the expensive EUV tools. Their cost structure and target market are very different from what TSMC and Intel are dealing with.
Exactly! Now your catching on. Equipment costs (not labor) dominate leading edge wafer costs. For GF and UMC all of their fabs are fully depreciated and they are on far less complex nodes. These circumstances would make labor a larger share of wafer costs. Furthermore given the commodity nature of these older nodes that wafer cost is even more important to actually fill up your fabs than it is for TSMC who has a much stickier ecosystem.
 
Back
Top