Chip Prices Face Possible Surge as Electricity Prices in Taiwan Set to Rise by 30%?

[LabRat 891]

Also, neither country have regular earth quakes or lie next to a major fault line... Or typhoons.

Yup. There's definitely more to factor-in on the engineering side. However, (competently ran) New Generation Reactors* negate/mitigate many of the issues.
*Which, Japanese and US mil-industrial contractors have made great contributions towards.

Even Japan got over their nuclear phobia, and Taiwan is in the same boat of being a high tech nation with little to no natural resources.

Sadly, the issue is seemingly less the technology, and more the organizations that run the reactors. As I recall, TEPCO had been found grossly incompetent on multiple fronts (regarding the 2011 disaster).
From what I can gather, that's not an issue limited to Japanese Business Culture and Regulatory ineptitude, either. IIRC, even into recent years, there are USMIL nuclear accidents still being declassified.

 

[TheLostSwede]

Even Japan got over their nuclear phobia, and Taiwan is in the same boat of being a high tech nation with little to no natural resources.

Government ≠ the people.

 

[LabRat 891]

Government ≠ the people.

Jokes aside...
You're not wrong, whatsoever.

 

[NoneRain]

This will result in 15-20 bucks increase.
Source: voices in my head

 

[evernessince]

The logic present in the article title is nonsensical. In order for an electricity price hike to impact chip prices electricity would have to hold a significant portion of the cost of producing chips. That is not the case though, not even remotely.

TSMC spends 2.97% of it's revenue on SG&A (Selling general and administrative expenses) which includes rent, utilities, and other overhead. By extension this means electricity is a fraction of that 2.97% figure.

Saying it accounts for 1% of total SG%A would be generous. Now factor in the 30% price increase and that brings you to a mere 1.3%. or a 0.3% price increase in the absolute worst case scenario.

So no, chip prices will not "surge" as the title puts it based off a 0.3% or less cost increase. That's far below inflation and could easily be considered an inflationary adjustment for the power company to maintain the same level of profitability.

 

[Gucky]

It's not just the wafer though, energy prices impact the whole facility. That beong said I don't expect it to be a noteworthy increase.

That was actually in that paper I got the 500kW from. Their calculation was for the WHOLE facility with a capacity of 20000 wafers per month.
It was around 9.3GW (9.300.000kW) of energy per month. So Instead of 930k$ they pay 1,2mil$, or instead of ~45$ per wafer, it is now around 60$ just for electricity.

It is a paper of 2019:

https://www.sciencedirect.com/science/article/pii/S2351978920300573/pdf?md5=21e48a5f5624e132eb3bc207742bdf35&pid=1-s2.0-S2351978920300573-main.pdf

Since I am not that knowledgeable, I can't tell how close it is to the actual powerusage of TSMC, but I guess that seems pretty close...

Other sources say that TSMC released data that shows that they use around 1kWh per cm² of wafer (also a few years old) and that TSMC uses ~20TW per year for ~30million 8" (equivalent) wafers.
That amounts to ~666kW per wafer. So that 30% would mean an increase of around 20$ per 8" wafer...

Even if they pay double of electricitycost per wafer, that would still be nothing compared to what AMD or Nvidia pays for the whole wafer.... The increase per AD102 chip for example would be less then 3$.

But I know, I know. TSMC puts a 200% markup on that 3$, then Nvidia puts up a 200% markup and then the seller puts up a 200% markup and we land on a 50$ increase for the model on the shelves...

 

[TheLostSwede]

That was actually in that paper I got the 500kW from. Their calculation was for the WHOLE facility with a capacity of 20000 wafers per month.
It was around 9.3GW (9.300.000kW) of energy per month. So Instead of 930k$ they pay 1,2mil$, or instead of ~45$ per wafer, it is now around 60$ just for electricity.

It is a paper of 2019:

https://www.sciencedirect.com/science/article/pii/S2351978920300573/pdf?md5=21e48a5f5624e132eb3bc207742bdf35&pid=1-s2.0-S2351978920300573-main.pdf

Since I am not that knowledgeable, I can't tell how close it is to the actual powerusage of TSMC, but I guess that seems pretty close...

Other sources say that TSMC released data that shows that they use around 1kWh per cm² of wafer (also a few years old) and that TSMC uses ~20TW per year for ~30million 8" (equivalent) wafers.
That amounts to ~666kW per wafer. So that 30% would mean an increase of around 20$ per 8" wafer...

Even if they pay double of electricitycost per wafer, that would still be nothing compared to what AMD or Nvidia pays for the whole wafer.... The increase per AD102 chip for example would be less then 3$.

But I know, I know. TSMC puts a 200% markup on that 3$, then Nvidia puts up a 200% markup and then the seller puts up a 200% markup and we land on a 50$ increase for the model on the shelves...

Also as pointed out, TSMC has set up contracts with various power producers in Taiwan and do in fact buy up all the renewable energy they can get their hands on in Taiwan, which at one point was 100%. As such, they'll pay a different, negotiated rate with their providers and won't rely on on the national power company in Taiwan (which is always on the verge of going bust anyhow, hence the increase).

 

[mechtech]

“A super-consumer is any entity that has drawn over 5 billion kWh over the past two years”

if it’s an individual meter that’s a crap ton of power. Just have to segregate a few buildings and give them a meter to get under…..if that’s a loophole??