TSMC Execution [2024]

Well, I believe the narrative is, like other industries, manufacturers which used to have factories in the US, moved to other countries to lower their costs. This is true for many companies in the US, but not very much so for logic fabs in the US, because the largest logic fab company in the US was (and still is) Intel. They have some presence in Israel and Malaysia but not very much elsewhere. This, in a sense, creates a illusion that it's possible to manufacture logic chips in the US with reasonable gross margin and cost. This is, IMHO, unfortunately not true.

If the objective is just to make TSMC and Samsung to build fabs in the US, that's really not difficult and it's already happening, due to the CHIPS act. Trump's point, AFAIK, is that the CHIPS act is not a good deal because it's handing out subsidies, where tariff will achieve the same without handing out government money. The government can even make money because of the extra taxation. However, this will increase chip prices in the US, and thus increases inflation. Note that the US is not the only market of high end chips. Many other countries are also buying chips. Thus, if the US actually raise tariff on chips, yes, these companies will make chips in the US, for the US market only, at a much higher cost and thus higher price (as long as it's less than the tariff raised). Other countries, on the other hand, will keep getting cheaper chips manufactured in other countries. This will end up making the US less competitive.
 
They already have some under construction and Arizona will start producing A16 for Apple and Zen 5 for AMD.

Well yea, but these still only represent a fraction of TSMC's total output, though. Hence why I said 'more' fab plants, cuz we'll need quite a few if we're to exclusively source chips only from American-made plants, speaking nothing of any packaging issues or whatever that is still handled by Taiwan and China.

It will be quite a long while before that is even possible, all while American consumers will deal with notably raised prices. It's frankly a despicable policy in so many ways, speaking nothing of the further politics involved with our allies.
 
Well yea, but these still only represent a fraction of TSMC's total output, though. Hence why I said 'more' fab plants, cuz we'll need quite a few if we're to exclusively source chips only from American-made plants, speaking nothing of any packaging issues or whatever that is still handled by Taiwan and China.

It will be quite a long while before that is even possible, all while American consumers will deal with notably raised prices. It's frankly a despicable policy in so many ways, speaking nothing of the further politics involved with our allies.

But it is a progress. There is also the news that they will produce 2nm chips in the US.

 
Intel stock rose 4% on Wednesday morning after investment banker R.W. Baird sparked fresh discussion about a potential strategic tie-up with Taiwan Semiconductor Manufacturing Company. Baird analyst Tristan Gerra reveals that discussions emerging from Asia's supply chain explore Intel spinning off its semiconductor fabrication unit into a joint venture with TSMC. Under this proposal, TSMC would provide critical semiconductor engineers and technical expertise to help manufacture advanced 3 nanometer and 2 nanometer chips in the U.S., supported by federal subsidies from the CHIPS Act.
Edit: Wait, this isn't an Intel thread, but it's 50% relevant anyway.
 
Last edited:
It seems like the market is buying into this story as INTC is currently up 15% while TSM is currently down 3.5%. If we look at this story from TSMC's angle (i.e. why would TSMC want this) it could be quite interesting. What we know for sure is that TSMC just commited to invest more than US$100 billion in the US. I'm not sure if it's clearly stated that they'll be building more fabs but to me it looks like managing some Intel's fabs is also one of the investment option.
Now, back to what does this buys TSMC... I think depending on how it works this could be an option to fast track the fab building process. It'll certainly be faster to take over someone else's fab than building one. However, how this is handled will be the big question. Are they going to replace most of Intel foundry's employees with TSMC ones? That'd be politically difficult. On the other hand, if they are just being "trained" or "integrated" into the new company, I suspect that huge culture conflicts will certainly occur and I personally am not very optimistic about this idea.
 
A step to being a complete monopoly? Aside from Samsung are there anyone else left with cutting edge nodes?

The key question here is whether these Intel fabs are going to be a net positive or a net negative. If it's a net negative then it's not going to benefit TSMC to become a complete monopoly. While it can be surprising but sometimes it's cheaper to just squeeze your competitors out than buying them.

One interesting note could be related to this is that I've read somewhere about TSMC CEO Wei talking about investing in the US. He specifically noted that one of the target is American talents. He acknowledged that the cost can be much higher in the US, but American talents are very good and are difficult to replicate elsewhere. From this point of view, taking over Intel foundry's empolyees could be beneficial but as I said the cultural differences (which I mean corporate culture) will be difficult to overcome.
 
It seems like the market is buying into this story as INTC is currently up 15% while TSM is currently down 3.5%. If we look at this story from TSMC's angle (i.e. why would TSMC want this) it could be quite interesting. What we know for sure is that TSMC just commited to invest more than US$100 billion in the US. I'm not sure if it's clearly stated that they'll be building more fabs but to me it looks like managing some Intel's fabs is also one of the investment option.
Now, back to what does this buys TSMC... I think depending on how it works this could be an option to fast track the fab building process. It'll certainly be faster to take over someone else's fab than building one. However, how this is handled will be the big question. Are they going to replace most of Intel foundry's employees with TSMC ones? That'd be politically difficult. On the other hand, if they are just being "trained" or "integrated" into the new company, I suspect that huge culture conflicts will certainly occur and I personally am not very optimistic about this idea.
The 15% bump was supposedly due Intels new CEO who apparently is against selling out (even foundries). Investors apparently pretty much forced the board to take him back on the board too, which he had left last August, suggesting they might not want to sell things out either, questioning the current boards views.
(This is from ex-Intel friend)
 
Jensen Huang outright denied NVIDIA being asked to be part of such venture
"Nobody's invited us to a consortium," Huang said. "Nobody invited me. Maybe other people are involved, but I don't know. There might be a party. I wasn't invited."
 
I could believe
Producing wafers at TSMC Arizona is only 10% more expensive than in Taiwan

I could believe the daily OpEx is only 10% higher, but (I think?) the majority of the labour costs are going to be on the construction/CapEx side, so I’m not sure if they are taking that into account correctly - they might be, or they might be intentionally misleading for political reasons, I honestly can’t tell without having access to the full data. And it’s always a bit subjective because you need to make assumptions about the amortisation period and utilisation.

I agree with their assessment it only gets better from here as they build a pool of experienced workers though, the 1st fab is obviously the hardest/most expensive to get right, and the fact yields are competitive is a great achievement.

NVIDIA is probably going to build a lot of GB300/Blackwell Ultra in Arizona to reduce risk and build some political goodwill, irrespective of cost and whether the tariffs actually apply. That doesn’t really work for Rubin in 2026 which is on 3nm though.
 
Maybe they know something we don't and Intel 18A has really caught up in yields.

If their trade secret moat gets smaller, having cutting edge fabs farther from home is less of a problem.
 
Back
Top