TSMC vs Intel and Samsung 2024 Guidance
- Thread starter fansink
- Start date
If you are talking about Intel Foundry their event is on 2-21 so we will find out more then. Personally I do not expect IFS to see a significant increase in wafer revenue for years to come. It is a very difficult and time consuming task. 18A foundry designs will start this year and it will be 2-3 years before customer product HVM?
I don't remember Samsung giving foundry guidance but I do know that Samsung 5/4 is doing much better but Samsung 3 is still a non starter. In my opinion customers will continue to lead with TSMC N3 and N2 then follow with Samsung and IFS. It is in the industry's best interest that we have three leading edge foundries, right?
The thing about semiconductor revenue is that it is a combination of ASP and unit count. My guess is that TSMC's uptick this year is half and half. Nvidia is the big outlier here. I would guess their ASPs are through the roof while Intel and AMD are battling it out in the trenches. The other non leading edge foundries will be much more challenged on ASPs than during the shortages, absolutely.
fansink
Well-known member
I'm asking about net revenue, where TSMC's net revenue was ~$68.9B, INTC 2023 net revenue was ~$54.2B, and Samsung's 2023 net revenue was ~$51.0B.
Why do this comparison? Just curious.
Net corporate revenues aren't a useful comparison point. It would be like comparing Walmart to Amazon and Walgreens. Sure they both make some revenue selling batteries. But Walmart sells general consumer goods and groceries, Walgreens is mainly a pharmacy, and Amazon owns is all online the largest cloud service provider and something of a media empire. As a result looking at corporate revenues isn't a useful comparison for anything, least of all who is the best battery retailer. Back to semis it depends on what your goal is. If your goal is to talk about manufacturing scale, then you need to intel's new fab-co numbers starting at EOQ1, and to somehow reverse engineer Samsung foundry revenues by removing the LSI business's share of the revenue and Samsung logic's share of the CAPEX.
MKWVentures
Moderator
Intels fab co num
Intels fab co numbers include internal work (which is 99% of the fab work). AND Intel plans to use "price" as the transfer cost to BU... not cost AND they are going to have a somewhat arbitrary "allocation of process and product development costs" ... some to BU some to Fab Co. End Results is that I would expect the new reporting to be way more convoluted than todays reported. Reminder external Fab foundry wafers sales in 2023 are <400M. Intel is selling mask making, tools, and packaging primarily so far. I would think they would have to report external sales .... but I am hearing they will not for strategic reasons. And they have already stated they will show losses in manufacturing.
fansink
Well-known member
For the same reasons that most investor wants to know, which is to better understand their prospects and predictability.
Notice how I said if you are talking about "manufacturing scale". If we are talking scale/amortization, then internal wafers are 100% as relevant as external wafers. If we wanted to compare the foundry business only, then I agree you would want to look at external revenue (which should also conveniently remove their 2D packaging to make it more like vs like compared to a TSMC, GF, or UMC) for both IFS and SF. As an addendum to the foundry health, Dan has also brought up the argument that in intel's case counting internal business is far more relevant than with Samsung as intel has consistently shown that they will outsource parts. So even if intel never generates a dime in revenue from IFS (which I know is already not true but roll with me here), intel increasing revenue off of internal demand alone is a good thing because intel can capture TSMC's margin for themselves and lower their wafer costs while doing it.
They said the TD's costs will be eaten by Fab-co. As opposed to the old way of doing things (allocating CAPEX, process/packaging R&D, and manufacturing O-costs to the BUs COGS).
I don't think so. So right now intel has their corporate statement and then they zoom in to show more specific financials for the various BUs. If intel is lumping MSO, IFS, TD, and IMS into one BU and charging other intel BUs for their goods/services, then there is no change to intel corp's cashflow. Intel cannot claim their revenue is now double off of this transition because Fab and Prod co are still the same entity. Since external IMS/foundry sales is real cashflow, intel will have to report it. So my assumption is that you will still see IFS on the corp-wide piechart and that will be their $200M per quarter (or whatever the number turns out as). Then they zoom in and show CCG revenue and op income like they normally do. Then the other BUs, until they get to Fab-Co and show the same set of numbers. Intel leadership also said they will keep us updated on deal value as things get signed to give a better idea of the value/growth of the IFS pipeline beyond what they will be obligated to report on their corp wide earnings.
MKWVentures
Moderator
Intels cost are usually higher than TSMCs price. Therefore The "price" to allocate is subjective (trust me on this)
I am not sure they will actually say what their external wafer sales revenues are..... I have heard they will not break it out. Lets see what happens
Obviously corp cash flow and P&L is unchanged. the issues is that they will show BUs make more money and manufacturing is money losing business for Intel.
I would focus on revenue recognition as allowed by law and not "lifetime deal value". What is intel IFS revenue plan over the next 2-5 years. I have a quantitative model for their plan, but Intel needs to announce it publicly
Just an opinion.... its going to get interesting when it all unfolds.
If we say that the west is 10-15% more expensive than TW and N7 wafers are sold at a 50-60% margin, then that means that the intel 7 structural cost+operational efficiency would have to be greater than 30-46% of N7's cost for intel's cost to be greater than TSMC's price. Does that sound reasonable, because I don't think so. With a cost delta that big intel's cost per FET advantage in the 2000s and 2010s would have been meaningless. A CPU on TSMC 40nm would cost way less than an intel 22nm one, and yet their old margin structure was too good for that to be anywhere close to true. Now if we want to talk about operating income wise I could see it. Pat and Ann have both mentioned in interviews "The blank check to complete 5N4Y". That isn't something you say if 5N4Y was a cheap endeavor. We also all saw the compressed margins N3 start up had on TSMC in 2023. Intel is supposedly going to be doing the same thing for intel 3 AND 20A this year. Combine that with the low margins of 2D packaging, lots of volume going external for MTL, and a fab startup rate that hasn't seen since the 300mm transition. IMO those will be the main costs, not intel wafers costing WAY more than what TSMC sells them for.
Wafer revenue, couldn't tell you. But I would be somewhat satisfied with the external IFS business since that is more then we get from Samsung who obscures their revenue with IC design and CAPEX with that of the memory business. Obviously IMS kind of mucks up the numbers right now, but presumably as the business scales that will be less of a factor (as well as intel having sold of substantial stake in the firm reducing their cut from IMS).
Isn't what you ask what they already provide? Their revenue recognition of 2D packaging + IMS + EMIB for amazon (assuming they still use intel) was like 300-whatever mill last quarter. Obviously this says nothing about the announced deals since chip design takes years, so intel literally can't be recognizing any wafer revenue or the most recent round of advanced packaging revenue at this point. Regardless investors would like to know that "Hey is this a $300M per quarter business in the next 5 years?". The best way to ease these doubts would be to say what the value of contracts signed is, no? In 5 years I suspect that will not be something to care about by as much though. In the late 2020s I doubt any investor wants to hear "Hey I know revenue isn't really materiel, buuuuuut our pipeline looks really good though.". By that time we need to be seeing that $10B pipeline being either fully or mostly delivered, with that next wave of however many billion already on the balance sheet, AND more internal wafer outs.
Agreed. Interesting to see how it starts and how it charts from now to 2030 when intel should be making good money on their depreciated assets, actually having a foundry portfolio/ecosystem, and hopefully be positive GM and free cashflow positive for fab-co. If they can do those things, in my books that is a smashing success considering it took GF like 15 years, dropping out of the leading edge, and a chip shortage just to get to profitability. On top of this they were even paid to take IBM's fabs/engineers and gifted Chartered semi (the world's second largest foundry at the time if I recall correctly) from their oil baron owners.
Last edited:
MKWVentures
Moderator
on most processes/products Intel cost is higher than TSMC price. It is more expensive to make at Intel than it is buy from TSMC. Intel margins in manufacturing are negative when wafers are sold at TSMC price.
tooLongInEDA
Moderator
Hold on - if Intel costs are higher than TSMC price (I'm taking your word on that - I've got nothing else to go on here), isn't that effectively saying that TSMC's costs are at least 40% (and probably more) lower than Intel's ?
If that is the case - and if there isn't a way to close that gap down to something much smaller - how is Intel going to pick up any significant external foundry business ? Or are we expecting Intel to squeeze 40% off its finished wafer costs ?
I would cut this equation in half. Intel's fabs without EUV and fabs with EUV. Opening up the non EUV fabs to foundry customers is brilliant because they will never be upgraded to EUV. The same can be said about mature EUV nodes when High-NA EUV is ready for HVM. I seriously doubt EUV fabs will be upgraded to HNA EUV. As long as Intel can keep these fabs running profitably it is a good business plan and a big plus for the greater good of Intel.
In regards to Intel getting leading edge foundry business, that will take some time. Customers must TRUST that Intel will deliver and that type of trust is earned. In my opinion foundry customers will test the waters with Intel 18A. If Intel does well on 18A I can see semiconductor whales using Intel 14A and below for big runners, absolutely.
hist78
Well-known member
Intel is complicated to the degree I feel they often doing things that are against their own words.
For example, Pat Gelsinger keeps saying Intel's cost is too high compared to its competitors. Consequently he did several things to cut the cost, including layoffs, extending depreciation schedule from 5 years to 8 years, implanting multi billion cost cutting initiatives, etc.
Yet, all Intel's new fabs are going to be built in those high cost areas: (2022 GDP per capita in US$)
USA $76,000
Israel $54,000
Ireland $104,000
Germany $48,000
While Intel's major competition are coming from: (2022 GDP per capita in US$)
Taiwan $33,000
Korea $32,000
Japan $34,000
The cost of hardware, machines, materials, services, and compensations will be impacted by the those various GDP levels.
tooLongInEDA
Moderator
Thanks. Yes, I'm recalling the depreciation change. Curious if that's now any different to TSMC or Samsung.
Note that Ireland's GDP is hugely skewed upwards by their tax arbitrage income (a huge amount of their national income is due to financial engineering) and actual per capita income is certainly not higher than the US.
MKWVentures
Moderator
Excellent observation! 1) Intel (pat, Stu, etc) think they can cut costs to get closer to TSMC. We shall see. second: Intel goal is not to make lots of money on Foundry, it is to get scale to allow them to continue to develop leading fab technologies (this is absolutely required... you can't lead in tech or cost without scale). I can't wait to go through the new accounting for Intel Manufacturing.
possible third: If you sell to the US government you can charge 2x+ the price and they will pay (enter favorite hammer, toilet seat joke here). TSMC does not and will not sell advanced logic to DoD. Intel will. What would happen if Biden made is so that government tech had to have Intel Fab'd parts . Just Hypothetically speaking
Side note: your comments on Ireland are also very good. You need a big excel table to show how the Ireland costs work out
hist78
Well-known member
Yes, I am aware the tax incentives and multi-national companies' revenue and IP licensing schemes that inflated Ireland's GDP. Nevertheless it does increase Ireland's per person income with the population of only 5.1 to 5.3 million.
I gathered some average salary info of the electrical engineer for the selected country in the following table. Sometimes people think semiconductor manufacturing is a capital intensive industry and labor cost has a very limited impact on a company's cost structure. I do not think it's that simple. Higher salary will push machines, tools, materials, consulting and service price higher and impacts a semiconductor company's revenue, profit, and competitiveness.
For example, ASML has 3600 employees working in Taiwan. All the products and services ASML provided to TSMC are tied to ASML Taiwan's cost, including labor cost. Same situation will be true for those products and services TSMC received in order to produce a chip.
IMO, Intel's policy to avoid setting up a fab in East Asia and to have all Intel fabs in high living cost/high income locations is not a winning strategy. It doesn't make sense in recruiting talents and gaining resources either.
Average Electrical Engineer Salary
Source: https://www.salaryexpert.com/salary/browse/countries/electrical-engineer
Source: https://www.numbeo.com/cost-of-living/rankings_by_country.jsp
Last edited:
When Gelsinger was discussing cost that are too high, he was discussing sales, marketing, planning, and design engineering. That's where the cuts were, and mostly in the first three areas. Do you have any evidence of layoffs in manufacturing or process development?
As for these other arguments, like GDP per capita being a direct indicator of fab equipment costs and wafer costs, that sounds ridiculous considering that Intel negotiates equipment costs at the corporate level, and the labor component of an operating fab is not a large factor in costs.
hist78
Well-known member
@blueone ,
Please read my original post. My point is that Intel's cost is too high and not competitive. It's good if you want to dig into a specific area to discuss or you want to expand it into a broader subject. But please don't just chop one segment out from my comments and tell me that I am ridiculous.
I have noticed that you often use some strong "words" in your comments. I hope you can refrain yourself from using them. Those words do not help to create a professional, healthy, and happy environment on this board.
Last edited: