News Producing wafers at TSMC Arizona is only 10% more expensive than in Taiwan: TechInsights

[Admin]

Despite comments suggesting high costs in the U.S., analysts found that producing chips at TSMC's Arizona fab costs less than 10% more than in Taiwan.

Producing wafers at TSMC Arizona is only 10% more expensive than in Taiwan: TechInsights : Read more

 

[thisisaname]

He is talking up the costs in order to justify charging a lot more.

 

[Notton]

Sure, the chip may only cost 10% more, but it still has to be packaged/bonded to a substrate, and that part of the process isn't up and running yet in AZ.

 

[Marlin1975]

"Only"?
10% is a pretty big number. Many business's run on less than 10% profit margins.

 

[phead128]

It's still shipped back to Asia to be packaged. This is why TSMC Arizona represents a small fraction of total capacity.

 

[ManDaddio]

With as many chips as tsmc sells I really don't understand why they keep talking about how much it cost to build fabs. They make billions of dollars in profit. Everything should be paid off within a couple years.
Sure I understand the cost of running the place is constant but still.
Give me a break.

 

[ManDaddio]

"Only"?
10% is a pretty big number. Many business's run on less than 10% profit margins.

Tsmc is not just any other business. 10% is nothing for them.

 

[isofilm]

Why should we trust your analysts, over the foundry's numbers, I'm SURE you analysts have included EVERY expense in their analysis.

 

[urdrwho10]

Despite comments suggesting high costs in the U.S., analysts found that producing chips at TSMC's Arizona fab costs less than 10% more than in Taiwan.

Producing wafers at TSMC Arizona is only 10% more expensive than in Taiwan: TechInsights : Read more

Is this 10% the total manufacturing cost or just a portion?

Direct materials+ direct labor+ overhead costs

 

[DS426]

The analysts noted "less than 10%", so how much are we talking actually? 1% is less than 10%.

TSMC doesn't have a cost problem -- it has a production capacity problem on their most advanced nodes.

Final numbers today probably are over 10% though as not all of the packaging can be done in the U.S. at this time, as others have noted.

 

[rluker5]

It is notable that labor is less than 2% of the cost in this case. This is clearly on the low side for manufacturing, but there are a lot of industries that have relatively high (material+equipment)/labor cost ratios.
Some seem to miss this.

 

[phead128]

TechInsights is a Canadian consulting firm, does Canada even have any semiconductor fabs? What do they really know.

 

[spongiemaster]

What do they really know.

Ice Hockey?

 

[Kalalicious]

Why should we trust your analysts, over the foundry's numbers, I'm SURE you analysts have included EVERY expense in their analysis.

For what it's worth, found this in a report on SeekingAlpha regarding TSMC's Q3 2024 earnings call (required having a free account for me):

Yet, TSMC's margins have been expanding at a rapid pace, despite the increasing mix shift towards the 3nm process node alongside rising electricity costs in Taiwan. The company delivered a substantial Q3 earnings beat, with gross margins exceeding the upper range of previous guidance by more than two percentage points at 57.8%.

Pardon my French, but that's a crap-ton of margin...

I initially was looking for the news regarding Intel losing a 40% discount from TSMC. My point being there that if the can afford to offer Intel a 40% discount, they've gotta have a sizeable margin already in place.

Anyway, yeah, they're margin is huge. Of course, I also believe they're reinvesting a very large portion back into the company for production, process improvements, r&d, etc.

 

[usertests]

"Only"?
10% is a pretty big number. Many business's run on less than 10% profit margins.

Even if they pass the cost onto the buyers, the big AI accelerators and datacenter CPUs have gigantic profit margins, and even today's ballooning wafer costs are low-ish compared to the final product.

For example, if you can get around 50 good 600mm^2 dies on a hypothetical next-next-next-gen $100,000 wafer (TSMC N2 is estimated to be $30,000), you might be making $1,000,000 in revenue or more from those dies.

The consumer/prosumer products are more likely to be squeezed, since 50x 600mm^2 dies sold at Titan/*90 MSRP pricing of around $2,000 would only be $100,000. Smaller chiplets can get more use out of the die. AMD can easily get 700-800 usable 70-75mm^2 chiplets out of a wafer for maybe $250,000 in consumer CPU revenue.

Anyway, as long as there's huge demand and the bubble hasn't popped yet, there should be no problem selling wafers at a slightly higher price. If demand evaporates for multiple years in a row, TSMC could be in the hot seat with all these fabs they've spent hundreds of billions on.

 

[thestryker]

"Only"?
10% is a pretty big number. Many business's run on less than 10% profit margins.

Nobody using an advanced EUV node is operating on margins anywhere near that low.

 

[Mr Majestyk]

"Only"?
10% is a pretty big number. Many business's run on less than 10% profit margins.

This has nothing to do with profit margin.

 

[JTWrenn]

So if they were not shipping it all back to Taiwan for the rest of the procedure would it be cheaper? Really curious on the breakdown here.

 

[phead128]

Ice Hockey?

Yes, Canadians need to stick with their ice hockey and maple syrup and leave semiconductor manufacturing to the big boys.

TechInsights is renown for reverse engineering and disassembly of iPhones and gadgets, they can glean insight into node info based on electron microscopes, but I don't think Ottawa, Canada is really a mecca for semiconductor manufacturing SMEs.

 

[JRStern]

Good news, especially if true.
And it makes a lot of sense, with the huge capital costs.
But then why would the company itself have previously cited much higher numbers?

 

[Pename]

Wait till China starts banning wafer material exports to US, cost will shoot thru the roofs.

 

[jkflipflop98]

Wait till China starts banning wafer material exports to US, cost will shoot thru the roofs.

Like what? China doesn't have anything that everyone else doesn't.

 

[hotaru251]

thats 10% to the customer...who then has to make their own cut and as hot as tech is they arent gonna raise their increase at a mere 10% when thehy know they will sell regardless....

 

[passivecool]

I recently listened, facinated, to the deep dive on tsmc at acquired.fm/episodes/tsmc-remastered (also spotify etc.)
If i remember correctly, they are running a 40% operating profit.
there is no one else on the planet who can compete. the podcast explains how they got there.
to the article., the hand full of EUV experts on the planet who can set up and run the machines necessary won't care where they have to fly to.

 

[systemBuilder_49]

I just want to provide some background - - -

1. Morris Chang (born 1930) IS THE VLSI FOUNDRY BUSINESS INVENTOR.
2. He came to the USA from Taiwan and attended Harvard, Stanford, and MIT.
3. Morris Chang became a USA citizen in 1962. Currently he works in Taiwan.
4. He invented the foundry business idea at TI in the 1980s, but they didn't want to become a foundry.
5. He returned to Taiwan and convinced the government to invest in his project.
6. He has worked for more than 50 years on his idea and deserves our great respect.

I was at the largest TSMC foundry in December (in Taiwan). Don't blame TSMC for the high costs of VLSI production. The high costs are a symptom that Moore's Law is ending. What happens at the end of Moore's Law is that foundry costs go up faster than VLSI gets better / cheaper / faster. And we are in that phase now. I truly believe that without TSMC, we'd be 1.5 generations behind in VLSI, everyone with a GPU would have an RTX3000 card or a RX6000 card, everyone with a CPU might have an AMD 5800x. Thank you, TSMC.

One of the main reasons that Intel is doomed is because VLSI costs are going up faster than ANY OF THEIR BUSINESSES ARE GROWING. Under these circumstances, you can hire the most brilliant engineers in the world and it won't matter. You must become a foundry and start taking outside customers. Morris Chang knew this in the 1980s. Intel is still too stupid to realize that it's their only choice, and they have dragged their ass for 20 years in recognizing this fact !!

TSMC needs MONSTER PROFITS to build a next generation fab. They need to save up $30B for their next-generation 2N fab, with annual operating costs of $3B. It's a gigantic gamble that nobody wants to finance. So yes, they have 40% profit margins, but what if I told you 38% of those margins were saved so they could continue to improve VLSI? Would you fault them for that ??

Each new generation of fab costs about 30-50% more than the previous generation. Costs at TSMC are going up exponentially. If the new fab is not exponentially better than the old one (with a HIGHER exponent), VLSI ends, it's kaput, no more improvement in VLSI, computers, GPUs, AI, nothing gets faster. Let that sink in. VLSI is currently mankind's greatest endeavor in history - our greatest struggle against Nature. Let that sink in ...