Onshoring is a logical result of improved automation.
When labor costs lessen, transportation and logistics become the primary concern. Wouldn’t be surprised if 100 years from now most final good assembly is done locally.
Of course sometimes raw materials and inputs are more expensive to transport than the finished good, so there will always be cases where long distance transportation of finished products is still preferred
The highest costs in Semiconductor fabrication are the specialized machines needed. Most of them are at best dual sourced. Many are single sourced. The fact a technician in the US can make $100k when their counterpart in China would make $50k doesn't matter when the EUV Lithography machine is hundreds of millions of dollars.
Capex amortizable across the life of the fab and usually treated beneficially for tax purposes (though these fabs are so tax-special that may be meaningless). Opex is with you forever and affects the economics of the business much more profoundly.
The semiconductor industry is somewhat special in that it requires nearly constant investment in not just R&D but also in upgrading existing fabs to keep up with the cutting edge. And these are well into the billions of dollars.
This is not a problem for players who have dropped out of the race like GloFlo. But its definitely the case for TSMC, Intel and Samsung. Your last three serious players who aren't working for the PRC.
Who makes those expensive machines? Are they made in-house by companies like TSMC? Or are they purchased from others? And whats the geopolitics of that part of the supply chain (made in taiwan? made in china? made elsewhere?)
The machines are made by other companies. Lithography is the most expensive now and its made by ASML in Netherlands is the only solution for EUV lithography. But there are many steps in manufacturing and a machine for each. The major equipment manufactures are spread across US, Europe and Japan. In the US the three big equipment manufacturers are Applied Materials and Lam Research and KLA.
The costs are probably driven by extreme engineering requirements (temperature, pressure, corrosive chemicals, low contamination) and R&D costs to meet these requirements. The production volumes are not high either. Its not like they are building millions of these machines.
Asianometry on YouTube is a good source for this kind of topic.
"Tax incentives" like writing off Capex have been there for the taking. So yes, the improved tax incentives with bipartisan support translates to stability for investors.
the cost reductions are necessarily worse in the places where costs are lower. I think you didnt do that math right. You might be right that the logistics for shipping silicon are still so good that this does not reach the amdahl limit, but it is not a consequence of your argument.
The benefits of JIT manufacturing outweigh everything else though. The waves of oversupply and undersupply in the semiconductor world are something that only JIT can fix, and transportation costs are only low if you go against JIT and try to bulk ship items.
While semiconductor fabs are highly automated, the machines are complex enough and do extreme enough things that they break a lot, requiring human intervention. MTBF for many of the machines used can be measured in hours, it's not like you can set up and go home. TSMC has around 65k employees overall - a cursory search did not yield their makeup by job breakdown, although from 1997 through 2003 around 50% of their employees were 'factory' workers via [0].
I'd expect high-margin items like semiconductors to be more amenable to onshoring anywhere - certainly Intel has almost all of its fabs in highly developed countries, but I think this is hardly a cost efficiency move - rather it's a hedge against the failure of globalization and the risk of the US losing access to advanced semiconductor technology in case of strife in Taiwan. Russia has already lost this access. TSMC took in about $56B in '21, of which about $28B was 'cost of revenue', the amount they needed to spend on supporting this revenue (marketing, legal, production, e.g., not building factories or R&D). There's certainly room for increased production costs while maintaining a profit.
It’s onshoring from the perspective of an American company that contracts out to TSMC fabs, such as Apple, AMD, nVidia, Marvell, Qualcomm, or other major TSMC clients. Sony is also a TSMC client so if TSMC built fabs in Japan, Sony would be onshoring.
Hsinchu, Taiwan, R.O.C., Nov. 9, 2021 - TSMC (TWSE: 2330, NYSE: TSM) and Sony Semiconductor Solutions Corporation (“SSS”) today jointly announced that TSMC will establish a subsidiary, Japan Advanced Semiconductor Manufacturing, Inc. (“JASM”), in Kumamoto, Japan to provide foundry service with initial technology of 22/28-nanometer processes to address strong global market demand for specialty technologies, with SSS participating as a minority shareholder.
Huh? This is re-shoring by political influence (probably pressuring Taiwan/TSMC) and financial incentives. Both TSMC and Samsung[1] are buildings Fabs in the US. Both Taiwan and South Korea are dependent on US military for their own existence. This is anti-competitiveness at its best.
I’ve heard the claim repeated by an Alcor employee. But they may have been repeating a myth. (Alcor definitely exists, and claims it chose Phoenix for the aforementioned reasons.)
Probably also hedging for geopolitical risk? Not sure how only having facilities in a country at least somewhat likely to be invaded in the next few years affects the stock price but it probably doesn't hurt to put one in the middle of a country very unlikely to be invaded and very likely to want to purchase chips built domestically.
Nuclear + Solar seems like it would be a pretty ideal mix in a hot climate, especially given that peak demand for air conditioning coincides with solar maximum output, and Nuclear could handle some of the base/overnight load.
It is but you also need controllable components with the ability to quickly change output performance when building solar as your peak source. Usually this means gas turbines, in some cases this could also mean battery.
Arizona has the reputation of being a desert, but in reality it is a state that is half-desert and half-mountain, and has a tons of water. Enough to waste on ubiquitous golf courses and alfalfa farms. There is a canal system taking the water from the mountains to the drier south where people live.
But the population pressures are realigning who pays what for the water now, so I expect the alfalfa farms are going to go, and chip fabs will take their place.
The thing you don't mention is that Arizona is wasting other people's water -- nearly 40% is sourced from rivers originating in other mountainous states
I don't mention it because it's not true. The idea that if a river originates in one state, then that state owns all the water from the river is not how ownership of river water works. And in America, rivers often form the border of states, so there are always competing claims about which state gets which percentage of the river. These are generally settled through treaties and agreements among states, and those agreements were made also for the Colorado river. Arizona is only taking its share of the water as per the treaty, not someone else's share.
And it's share of water was already insufficient a decade ago, which is why the water table has run dry in many rural areas, forcing people to truck in bottled water or pack up and leave.
> Arizona has the reputation of being a desert, but in reality it is a state that is half-desert and half-mountain, and has a tons of water.
The mountainous areas are 'high deserts'. Most of Arizona's water is underground. It's being pumped out faster than it's replaced. Old wells go dry all the time, because newer wells are deeper.
"The Sonoran Desert's bi-seasonal rainfall pattern results in more plant species than any other desert in the world. The Sonoran Desert includes plant genera and species from the agave family, palm family, cactus family, legume family, and numerous others." - https://en.wikipedia.org/wiki/Sonoran_Desert#Flora
> There is a canal system taking the water from the mountains to the drier south where people live.
The Central Arizona Project pumps water straight east from Lake Havasu (738 ft elevation) to Phoenix (1086 ft elevation), then south to Tucson (2388 ft elevation).
The Salt River Project [SRP] buffers water from the rivers that run through the Phoenix area, and diverts the rivers into canals for delivery to the Phoenix area. Before the dams, the Salt and Verde rivers ran year-round. SRP modernized/expanded the Hohokam's ancient canals.
This may not be true today. But when I first arrived in US in AZ, I was told Phoenix boasts highest numbers of golf club concentration in entire US, despite high water requirements of a golf club and being in middle of desert.
I suspect the ongoing Western US drought would have worsened the water situation.
from the wikipedia link - they use sewage/greywater for cooling:
> The power plant evaporates the water from the treated sewage from several nearby cities and towns to provide the cooling of the steam that it produces.
This treated sewage is the water that Vegas now provides for general use, and what AZ will need just for its people in about 5 years. Wasting it on chip production is going to go over slightly better than the Saudis wasting groundwater on alfalfa.
They could literally truck Evian water in if they wanted to. Water is extremely cheap compared to just about every substance on earth, and it’s hard to imagine that water costs can be a problem for this type of operation.
You're assuming that the severe undervaluing of natural resources in the US will simply continue. Try taking a field trip to a place (including the US), where people can't drink the water that comes out of their tap, and you'll get a better idea of where this leads. In no universe can TSMC use bottled water, much less Evian, and be competitive.
If you’re referring to the Palo Verde Nuclear Generating Station, it’s downstream of water treatment plants and is effectively evaporating the water from the effluent stream [1], in other words it’s cooled by 100% recycled water. Probably a significant amount of the evaporated water returns to the local area.
I think that depends how much build-out you have to do in order to install the kind of capacity required for a $40B semiconductor fab, which would hypothetically need some kind of major energy storage infrastructure or backup source in the event that solar is the main source of electricity.
> Large semiconductor fabs use as much as 100 megawatt-hours of power each
hour, which is more than many automotive plants or oil refineries do. [0]
If a fab is running at that consumption rate for twenty hours every day for one year, it will consume approximately all of Arizona's 735,000 MWh of annual utility-scale solar, wind, and geothermal net electricity generation. [1]
Contrast that with the annual hydroelectric generation of Itaipu Dam in South America, at 79,440,000 MWh in 2019. [2]
I prefer it as written, because the additional information of timeframe for measurement indicates that the author is not talking about momentary/peak power.
I'd still much rather see "100 MW continuous" over "100 megawatt-hours of power each hour". And the former actually tells you a lot more than the latter, as with the latter it could be bursty (e.g. much more than 100 MW momentary draw), which would be a lot harder to supply. Fortunately for everyone the power draw is actually steady.
Which indicates the author knows nothing about how semiconductor fabs use power. Power usage is stable with a few percent...and 100MW is really on the low side for this fab. If all 6 phases are completed as they are expected to be...it's more like 250MW.
Much of the power is used for keeping air clean with temperature and humidity controlled in a very tight range. More energy is used for water cooling, waste treatment and other facilities. The rest of the power is used for production tools...which when a fab is running are generally consistent users. The power usage will scale with wafer loading, but again advanced fabs like these will run at near 100% capacity for years at a time.
Your EIA source is for generation in the month of august only. Total solar generation in Arizona is much higher than that. Very roughy 1,500x nameplate capacity = total annual generation in Arizona. 5,600 mw capacity in 2021, 8,400,000 mwh annual generation.
All fabs have battery backup regardless of power source, although with solar you would need a much larger one indeed. Either way, the scale of solar power generation easily supports these new fabs.
So all power infrastructure would be less than 5% of the cost of the plant, and that’s overbuilding solar needs by 4x (meaning even on the crappiest day power wise you’ll have more than enough) and having 16 hours of backup power which is also overkill.
Here is a solar plant in California, which broke ground over a decade ago, that produces 550 MW of power, for $2.5B. Build even a small version of one of these and its 100 MW power needs are met.
The 550 MW "nameplate capacity" is peak. The capacity factor of 26.6% means that it actually only produces about 146 MW on average. Nuclear power plants generally have a capacity factor above 90%.
Electricity usage is much, much lower at night in Arizona because everyone turns AC off due to temperature drop. Plenty of baseload generators that can be better utilized at night with increased manufacturing.
No, you would need a large storage facility as well.
A nameplate capacity of 550 MW with a capacity factor of only 26% almost certainly means there are a lot of times when it isn't generating any power at all, or at least none to speak of.
Right, but if those align with the times where the plant is not used at all (night), does it matter? That being said, I have no idea what the normal operational schedules are for these types of facilities.
When labor costs lessen, transportation and logistics become the primary concern. Wouldn’t be surprised if 100 years from now most final good assembly is done locally.
Of course sometimes raw materials and inputs are more expensive to transport than the finished good, so there will always be cases where long distance transportation of finished products is still preferred