TSMC May Increase Wafer Pricing by 10% in 2025

[Scotten Jones]

TSMC started N5 risk production in April 2019 at its Fab18 in Tainan Taiwan. That means around April 2024 or a little bit latter certain part of TSMC N5 equipment had completed its 5-year depreciation schedule. Any newer equipment or equipment put into service later will definitely reach the end of 5-year depreciation schedule on a later date.

I can tell you for a fact that through Q1 of 2024 none of the 5nm equipment is fully depreciated. My costs match their financials within about 2% and I have everything still depreciating. When Q2 comes out I will check again.

 

[Scotten Jones]

By this logic did intel 4/3 tooling in D1 that was not used for intel 7 production only start depreciating once intel sold it's first MTL chip to an OEM, rather than when they turned those tools over to the development team's pilot line? My understanding from my engineering economics course in college was that R&D tooling counts as "in production" as soon as you start running value creating experiments. If my understanding is correct, why wouldn't that apply to F18's N5 pilot line pre first revenue generating wafer out?

Intel D1 is really complex, there is mod 1, mod 2, and mod 3, different nodes in different phases of development/production in each. R&D costs and depreciation likely all hit the R&D line but when a mod starts running initial production then it would start being charged to cost of goods sold.

 

[Scotten Jones]

For tax purpose, typically once an equipment is "placed in service", the depreciation will start as soon as possible. It's not required that the equipment needs to generate revenue or be running in order to record the depreciation expense. I think TSMC will start the equipment depreciation as soon as its risk production started. Additionally for internal accounting purpose the equipment depreciation may start even earlier to truly reflect the useful life of the machine. This may be different from the depreciation reported to the taxing authorities.

I think it is the opposite, you don’t want to have depreciation expenses hitting your P&L dragging down profitability until you have offsetting revenue.

 

[hist78]

I can tell you for a fact that through Q1 of 2024 none of the 5nm equipment is fully depreciated. My costs match their financials with about 2% and I have everything still depreciating. When Q2 comes out I will check again.

Depreciation of an asset begins when it is available for use. Whether it generates revenue is not important and not required. A machine can be idle in the factory but still allowed to incur depreciation expense. This kind of accounting principles and tax rules are the same in Taiwan or in the US. TSMC has a very high gross profit margin each year (49% in 2019). It will be foolish for TSMC not to record depreciation expense when it started using an EUV machine in April 2019. TSMC (and any "normal" companies) wants to have more depreciation expense to reduce its tax labilities.

Obviously any taxing authorities wish a company to report as little depreciation expense as possible

 

[hist78]

I think it is the opposite, you don’t want to have depreciation expenses hitting your P&L dragging down profitability until you have offsetting revenue.

IMO, Intel is not a "normal" company in terms of financial integrity and discipline. Intel changed its fab equipment depreciation schedule from 5 years to 8 years in order to make the appearance of its finance situation better. But I believe they still need to follow the accounting principles to record depreciation when the machine is available for use. It will be a scandal if Intel intentionally delayed the depreciation for a significant amount time. A well-known CPA firm will probably flag it as a serious issue.

 

[hist78]

In my experience they start depreciating equipment right around when they start reporting revenue, within one or two quarters.

Revenue has nothing to do with whether a machine's depreciation starts or not.

 

[hist78]

By this logic did intel 4/3 tooling in D1 that was not used for intel 7 production only start depreciating once intel sold it's first MTL chip to an OEM, rather than when they turned those tools over to the development team's pilot line? My understanding from my engineering economics course in college was that R&D tooling counts as "in production" as soon as you start running value creating experiments. If my understanding is correct, why wouldn't that apply to F18's N5 pilot line pre first revenue generating wafer out?

I don't think Intel dares to delay depreciation start time until it sold a processor made by that particular machine. Intel did extend the its machine depreciation schedule from 5 years to 8 years but Intel can't change the accounting principles enforced by auditors and SEC. The accounting principle is very clear: "Depreciation of an asset begins when it is available for use".

 

[nghanayem]

IMO, Intel is not a "normal" company in terms of financial integrity and discipline. Intel changed its fab equipment depreciation schedule from 5 years to 8 years in order to make the appearance of its finance situation better. But I believe they still need to follow the accounting principles to record depreciation when the machine is available for use. It will be a scandal if Intel intentionally delayed the depreciation for a significant amount time. A well-known CPA firm will probably flag it as a serious issue.

What are you on about!? There is nothing dishonest about using a different depreciation schedule it is in the law. As long as you stick to it the whole time all is legal. Intel used to use a 4yr depreciation model for most of it's existence. Were TI, GF, AMD, TSMC, Samsung, Micron, and SK cheating because they were making their numbers look better than using a 4yr. Of course not, it just changes how your cost changes overtime, you get to the same place eventually. Yeah you could do 5 or 8 years and your burden rate is lower in years 1 and 2 than the 4yr case. But you are still paying more in subsequent years. Given how expensive buildings are they (note the building not the land) are allowed to be 27.5 or 29.5 depending on what kind of building it is. Intel also did say that only some of the equipment would use the longer period. My guess without seeing the books is EUV tools given how expensive it is and the near time cash shortage at intel until these EUV nodes can get spooled up. It seems like a shrewd business decision to use a longer deprecation period to afford fab expansions in the short-term and accept slightly lower profits during the mid-term and same profitability during the long-term. In the world of intel nodes lasting 4-6 years a 5 or even 8 year depreciation schedule makes no sense. But if nodes will now run for decades like they do at TSMC, why not if it helps intel get past the hardest part of their foundry transformation.

Note the integral of the left plots are the same ($100)

 

[BruceA]

TSMC uses five year depreciation for equipment as disclosed in their annual report. Depreciation doesn't start until equipment is in production, I am not confusing anything, 100% of TSMC's 5nm equipment is still depreciating!

The other thing to keep in mind is they started phase 1 and then brought up phase 2, and phase 3. In each case they put some capacity on-line and then added to it over time. If you put 10k wpm on line, and then 6 months later add another 10k wpm, each group depreciates on a different time line. I track every fab by phase and by capacity over time.

Phase 1 and 2 will still be depreciating at the end of this year, at the end of next year roughly half the equipment will be depreciated in each. For phase 3 all the equipment will still be depreciating at the end of next year. Plus eventually Arizona will come on-line with new 5/4nm equipment that will start depreciating then.

“None of the "N5 Fab 18a" equipment is depreciated yet, your "mostly depreciated" statement is wrong.”

I stand by my comment I am closer to being correct than your “None” seriously. When I said mostly depreciated a tool say 100M cost in P1 is now 3/4 of the say deprecated, “mostly” and stands at 20M of assets. NONE means still 100M, I guess we are interpreting the words differently. As to the cost it means TSMC is going to see the depreciation roll off very soon as a factor of total cost, of course they have the burden of Fab18b N3 starting up, but N5 will reach N7 Fab15 like structure soon!

 

[Scotten Jones]

Revenue has nothing to do with whether a machine's depreciation starts or not.

Analyzing their reported financials I can tell when they start reporting the depreciation and they start reporting it around when they start reporting revenue. You can express all the opinions you want but that is what the numbers say they actually do for their production equipment.

 

[Scotten Jones]

“None of the "N5 Fab 18a" equipment is depreciated yet, your "mostly depreciated" statement is wrong.”

I stand by my comment I am closer to being correct than your “None” seriously. When I said mostly depreciated a tool say 100M cost in P1 is now 3/4 of the say deprecated, “mostly” and stands at 20M of assets. NONE means still 100M, I guess we are interpreting the words differently. As to the cost it means TSMC is going to see the depreciation roll off very soon as a factor of total cost, of course they have the burden of Fab18b N3 starting up, but N5 will reach N7 Fab15 like structure soon!

There are two ways to looks at this.

From a balance sheet perspective if you put $1B of equipment on line, the first year your write off $200M and have $800M left, the second year you write of $200M and have $600M left, etc. So in that sense by the end of the third year it is more depreciated than not.

But from a P&L perspective you spend $1B and your write off 20% of $1B every year for 5 years so nothing is really depreciated until the end of 5 years. It is semantics to some extent but from a wafer cost perspective the depreciation situation for a specific equipment doesn’t improve at all until the end of 5 years.

TSMC is still writing off 20% per year of 100% of their 5nm production investment to date. That is my point and it is confirmed by their financial statements.

Plus as I previously pointed out they brought equipment on-line over time as they ramped up, only the very earliest equipment will stop depreciating any time "soon". Plus phase 7 is brand new and Arizona will be soon.

 

[Scotten Jones]

What are you on about!? There is nothing dishonest about using a different depreciation schedule it is in the law. As long as you stick to it the whole time all is legal. Intel used to use a 4yr depreciation model for most of it's existence. Were TI, GF, AMD, TSMC, Samsung, Micron, and SK cheating because they were making their numbers look better than using a 4yr. Of course not, it just changes how your cost changes overtime, you get to the same place eventually. Yeah you could do 5 or 8 years and your burden rate is lower in years 1 and 2 than the 4yr case. But you are still paying more in subsequent years. Given how expensive buildings are they (note the building not the land) are allowed to be 27.5 or 29.5 depending on what kind of building it is. Intel also did say that only some of the equipment would use the longer period. My guess without seeing the books is EUV tools given how expensive it is and the near time cash shortage at intel until these EUV nodes can get spooled up. It seems like a shrewd business decision to use a longer deprecation period to afford fab expansions in the short-term and accept slightly lower profits during the mid-term and same profitability during the long-term. In the world of intel nodes lasting 4-6 years a 5 or even 8 year depreciation schedule makes no sense. But if nodes will now run for decades like they do at TSMC, why not if it helps intel get past the hardest part of their foundry transformation.
View attachment 2094

Note the integral of the left plots are the same (

Intel is certainly not the only company to do this. Micron extended depreciation on their DRAM equipment to 7 years in 2016 and then to 7 years on their NAND equipment in 2020.

 

[Scotten Jones]

I don't think Intel dares to delay depreciation start time until it sold a processor made by that particular machine. Intel did extend the its machine depreciation schedule from 5 years to 8 years but Intel can't change the accounting principles enforced by auditors and SEC. The accounting principle is very clear: "Depreciation of an asset begins when it is available for use".

"Depreciation of an asset begins when it is available for use".

That is not consistent with my understanding and not consistent with my analysis of their financials. When they buy an asset it goes into assets in progress and stays there not depreciating until they put it on-line.

 

[Scotten Jones]

IMO, Intel is not a "normal" company in terms of financial integrity and discipline. Intel changed its fab equipment depreciation schedule from 5 years to 8 years in order to make the appearance of its finance situation better. But I believe they still need to follow the accounting principles to record depreciation when the machine is available for use. It will be a scandal if Intel intentionally delayed the depreciation for a significant amount time. A well-known CPA firm will probably flag it as a serious issue.

As I said elsewhere Intel is not the only company to change their depreciation period, I can pretty much guarantee they checked with their auditors before doing that.

I have no idea where your intentionally delayed depreciation comment comes from.

 

[fansink]

As I said elsewhere Intel is not the only company to change their depreciation period, I can pretty much guarantee they checked with their auditors before doing that.

I have no idea where your intentionally delayed depreciation comment comes from.

Thank you very much Mr. Jones, for sharing your meticulous data and insights here at SemiWiki, and elsewhere, for these many years.

I always look forward to your posts and articles.

Cheers

 

[hist78]

As I said elsewhere Intel is not the only company to change their depreciation period, I can pretty much guarantee they checked with their auditors before doing that.

I have no idea where your intentionally delayed depreciation comment comes from.

I was trying to point out Intel can't delay depreciation as you described because that's against GAAP Principles.

 

[hist78]

"Depreciation of an asset begins when it is available for use".

That is not consistent with my understanding and not consistent with my analysis of their financials. When they buy an asset it goes into assets in progress and stays there not depreciating until they put it on-line.

It's a great discussion. My suggestion to anyone interested in this subject is to talk to their accountant friends (especially with experience in manufacturing sector) and search for related accounting principles and rules on Google or Bing. There are several related IRS publications too.

Here is the answer from ChatGPT:

Please note: IFRS (International Financial Reporting Standards) is used in Taiwan, including TSMC. GAAP (Generally Accepted Accounting Principles) is followed in the US.


"Yes, there are differences between IFRS (International Financial Reporting Standards) and GAAP (Generally Accepted Accounting Principles) regarding when to begin depreciation of equipment.

IFRS

Under IFRS, depreciation of an asset begins when the asset is available for use, which means it is in the location and condition necessary for it to be capable of operating in the manner intended by management. This is specified in IAS 16, "Property, Plant and Equipment."

GAAP

Under US GAAP, the depreciation of an asset also begins when it is placed in service. This means the asset is ready and available for its intended use, regardless of whether it is actually being used. The relevant guidance is found in ASC 360, "Property, Plant, and Equipment."

Key Differences

1. 1. Criteria for "Available for Use":
   
   • 
     
   • IFRS: The focus is on the asset being in the location and condition necessary for it to operate as intended by management.
   • 
     
   • GAAP: The focus is on the asset being ready and available for its intended use, which is slightly more flexible and can be interpreted as when the asset is placed in service.
2. 2. Specific Guidance:
3. 
   
   • IFRS: IAS 16 provides detailed guidance on determining when an asset is available for use.
   • 
     
   • GAAP: ASC 360 provides guidance but is generally considered to be somewhat less prescriptive than IAS 16 in defining when an asset is ready for use.

While both standards are fundamentally similar in that depreciation begins when the asset is ready for use, the exact criteria and interpretations can vary slightly between the two frameworks."

 

[hist78]

What are you on about!? There is nothing dishonest about using a different depreciation schedule it is in the law. As long as you stick to it the whole time all is legal. Intel used to use a 4yr depreciation model for most of it's existence. Were TI, GF, AMD, TSMC, Samsung, Micron, and SK cheating because they were making their numbers look better than using a 4yr. Of course not, it just changes how your cost changes overtime, you get to the same place eventually. Yeah you could do 5 or 8 years and your burden rate is lower in years 1 and 2 than the 4yr case. But you are still paying more in subsequent years. Given how expensive buildings are they (note the building not the land) are allowed to be 27.5 or 29.5 depending on what kind of building it is. Intel also did say that only some of the equipment would use the longer period. My guess without seeing the books is EUV tools given how expensive it is and the near time cash shortage at intel until these EUV nodes can get spooled up. It seems like a shrewd business decision to use a longer deprecation period to afford fab expansions in the short-term and accept slightly lower profits during the mid-term and same profitability during the long-term. In the world of intel nodes lasting 4-6 years a 5 or even 8 year depreciation schedule makes no sense. But if nodes will now run for decades like they do at TSMC, why not if it helps intel get past the hardest part of their foundry transformation.
View attachment 2094

Note the integral of the left plots are the same ($100)

With all the possible good intention from Intel, as you mentioned on extending equipment depreciation from 5 to 8 years, we have to look into to the reality and what kinds of problems can be imposed on Intel.

Under the rapid change of manufacturing technology and market demand, TSMC's strategy is to depreciate the equipment as quick as the tax laws and accounting rules allowed while Intel is doing the opposite way. Do we really believe Intel's assertion that the equipment is still be competitive during the 6th, 7th, and 8the year?

If the answer is "yes", then Intel will be facing TSMC's near zero depreciation cost competition because TSMC completes its equipment depreciation in 5 years.

If the answer is "no", Intel will be holding a lot of non-competitive or non-productive equipment for three additional years than TSMC. TSMC is going to recover all its equipment investment, buy newer and more powerful machines, and move on.

Extending equipment depreciation from 5 years to 8 years has several effects on Intel:

1. Reduce the the expense and possibly boost Intel's EPS or keep Intel's EPS not to fall too quick. It will be good for Intel CEO and CFO to reach their performance requirements at the expense of Intel's future. Intel spent $133.9 billion since 2000 or $83.6 billion since 2010 in stock buyback program in order to boost Intel's EPS. Extending years of depreciation and buy back shares bear the same Intel attitude: keep EPS afloat while avoid addressing Intel's fundamental problems.

2. Reduce the annual depreciation expense can cosmetically boost Intel's "book value". Better book value and EPS can help Intel's stock price and consequently help Intel CEO's and CFO's performance review.

3. Reduce the annual depreciation expense can make people believe certain Intel capital investments are finically viable while the actual situation might be going the other way.

4. According Intel's estimate, extending the equipment depreciation from 5 years to 8 years resulted $2.5 billion increase to the gross margin for 2023. But wait a moment, Intel 2023 GAAP net profit was only $1.7 billion! That means without this depreciation manipulation, Intel 2023 will be in a net loss.

Semiconductor business is a very challenging and difficult business. But one thing for sure, financial engineering will masquerade the true problems and pass the hot potatoes to the future CEO and CFO.

 

[tooLongInEDA]

With all the possible good intention from Intel, as you mentioned on extending equipment depreciation from 5 to 8 years, we have to look into to the reality and what kinds of problems can be imposed on Intel.

Under the rapid change of manufacturing technology and market demand, TSMC's strategy is to depreciate the equipment as quick as the tax laws and accounting rules allowed while Intel is doing the opposite way. Do we really believe Intel's assertion that the equipment is still be competitive during the 6th, 7th, and 8the year?

If the answer is "yes", then Intel will be facing TSMC's near zero depreciation cost competition because TSMC completes its equipment depreciation in 5 years.

If the answer is "no", Intel will be holding a lot of non-competitive or non-productive equipment for three additional years than TSMC. TSMC is going to recover all its equipment investment, buy newer and more powerful machines, and move on.

Extending equipment depreciation from 5 years to 8 years has several effects on Intel:

1. Reduce the the expense and possibly boost Intel's EPS or keep Intel's EPS not to fall too quick. It will be good for Intel CEO and CFO to reach their performance requirements at the expense of Intel's future. Intel spent $133.9 billion since 2000 or $83.6 billion since 2010 in stock buyback program in order to boost Intel's EPS. Extending years of depreciation and buy back shares bear the same Intel attitude: keep EPS afloat while avoid addressing Intel's fundamental problems.

2. Reduce the annual depreciation expense can cosmetically boost Intel's "book value". Better book value and EPS can help Intel's stock price and consequently help Intel CEO's and CFO's performance review.

3. Reduce the annual depreciation expense can make people believe certain Intel capital investments are finically viable while the actual situation might be going the other way.

4. According Intel's estimate, extending the equipment depreciation from 5 years to 8 years resulted $2.5 billion increase to the gross margin for 2023. But wait a moment, Intel 2023 GAAP net profit was only $1.7 billion! That means without this depreciation manipulation, Intel 2023 will be in a net loss.

Semiconductor business is a very challenging and difficult business. But one thing for sure, financial engineering will masquerade the true problems and pass the hot potatoes to the future CEO and CFO.

Some good points.

Really enjoying this discussion. Not something I expected to be saying about the accountancy aspects of the business ! But depreciation is such a huge part of the semi business cost structure that it's worth getting a better understanding.

 

[kvas]

Thanks for explaining what the accounting guidelines say about depreciation -- this seems like the right source to look at. However, reading this discussion, I'm thinking about an even more basic accounting principle: matching (expense should be reported in the same period in which the corresponding revenue is earned). I would interpret it as meaning that depreciation should be reported together with revenue, or at least it would make sense to start depreciation when first revenue is reported. Is this too much of a "book knowledge" understanding of accounting?