You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Let's check from another side, do you think intel delivered 3N3Y and will have 2N in 2024? If we check it based of HVM ready with high volume, IMO it "is not" and "will be not" but good enough. For breakeven, you need to have effective capacity, good yielded wafers and cost structure. If intel bets on 14A using Hi NA EUV, it will be very very challenging to be breakeven in 2027 and also in 2030 with 3 more year preparation.
I don't know what wafer pricing Intel uses for their product groups. Is it list price? Or is it preferred pricing? Will Intel provide that level of transparency?
The foundry business is very competitive and I believe that IFS will be competing more with Samsung than TSMC. Samsung is very aggressive on pricing. Do you think customers will pay a TSMC like premium for Intel wafers? Or will Intel have to compete with Samsung pricing?
IFS was losing 4B in 2022(using their new financial model), even if they have in-house customer(Intel products). It's quite surprising because Intel products teams are the single largest leading edge silicon users in the world. Intel products revenue was almost 60B per year. Their modelled foundry revenue was 27B, which is already much larger than Samsung's(sometimes surplus, sometimes deficit).
They need to be really great to achieve break-even in 2028. Foundry lost 6.95B in 2023. Their model says that 2.1B deficit came from "Lower product profit driven by lower internal revenue". Since Intel products' revenue decreased about 10B(57B => 47.6B) during 2023, we can say that Intel foundry will break even if their products sell 70~80B per year, or they fill 30~40% of their capacity with external foundry customers. Maybe more cost-saving and increased utilization help them a bit more.
Followings can help achieving break-even
- Get 'all' of products revenue from AMD(22B??)
- Make mobile customers 'magically' jump to Intel in 2026~2027
- drastic cost cuts in every places
But we know that Intel is also relying on external silicons, so they need to boost more products revenue to break-even. With more cost cuts. No wonder some expects 2030 timeframe..
If Intel had to do all its Intel product 4, 3, 20, 18A, 14A tiles in-house, there wouldn't be enough capacity for foundry. So they had to outsource to TSMC. But it's doubtful they have the capacity for the same customer load as TSMC. There was a report that one customer did pay ahead for capacity in Arizona: https://www.tomshardware.com/news/intel-accelerates-arizona-fab-buildout-to-regain-lead-in-2024
If Intel had to do all its Intel product 4, 3, 20, 18A, 14A tiles in-house, there wouldn't be enough capacity for foundry. So they had to outsource to TSMC. But it's doubtful they have the capacity for the same customer load as TSMC. There was a report that one customer did pay ahead for capacity in Arizona: https://www.tomshardware.com/news/intel-accelerates-arizona-fab-buildout-to-regain-lead-in-2024
Let's check from another side, do you think intel delivered 3N3Y and will have 2N in 2024? If we check it based of HVM ready with high volume, IMO it "is not" and "will be not" but good enough. For breakeven, you need to have effective capacity, good yielded wafers and cost structure. If intel bets on 14A using Hi NA EUV, it will be very very challenging to be breakeven in 2027 and also in 2030 with 3 more year preparation.
Since Intel 7/12 was reasonably mature before Gelsinger’s 5N4Y proclamation, and given 4/3 and 2/1.8 are single nodes, and a node doesn’t count until HVM begins, as of today, I’d say the score is 1N3Y, and the next score will be 2N4-5Y.
I don't know what wafer pricing Intel uses for their product groups. Is it list price? Or is it preferred pricing? Will Intel provide that level of transparency?
The foundry business is very competitive and I believe that IFS will be competing more with Samsung than TSMC. Samsung is very aggressive on pricing. Do you think customers will pay a TSMC like premium for Intel wafers? Or will Intel have to compete with Samsung pricing?
historically the price was allocated cost from manufacturing. The allocated cost from process development went to product groups also. So OM for product groups was poor due to inefficient manufacturing and development.
Under the new model, the product groups are charged fair market price. Simplistically, this is TSMC price. This price includes technology development and manufacturing. So now the OM for product groups is much better. OM for manufacturing is negative. We have a spreadsheet breaking out the impacts. Note: this changes both cost of sales and RnD costs as manufacturing and LTD are treated as as a foundry.
Intel has stated they will have to beat competitors price to win market share. So you can guess impact to margins.
As I mentioned in our summary note . Intel HAD to do this (Charge TSMC price and make foundry eat loss). otherwise, the product group would not choose Intel wafer fabs over TSMC. Intel had freed up product groups to choose on their own years ago.
As for whether margins will be break even. It is not clear how this could happen before 2029/2030 using GAAP accounting .... put the numbers in a spreadsheet and check it out (we did) We can check costs over the next 1-2 years and we will know for sure.
we can say that Intel foundry will break even if their products sell 70~80B per year, or they fill 30~40% of their capacity with external foundry customers
We can see this in another way. In 2023, Samsung electronics reported that about 6B dollars were spent to buy Application Processors(from Qualcomm...etc). This means that Intel foundry will still lose money even if they get 20%(Samsung's smartphone share) of AP foundry share. In other words, they need to get more than half of mobile foundry market to break-even.
In other words, Having world's largest fabless(Intel products) AND getting half of AP foundry share does not give IFS net profit, but just break-even. So Intel's struggle is not really about release of 18A, 14A....etc nodes. We can't expect intel getting +20~30% margins per wafer compared to their competitions. It's more like Intel fighting some residual fats inside. In many cases, these fights takes really long time(corporate cultures, working processes...) and TSMC have been doing this for more than 30 years.
Chiplets really are going to change the foundry business. 2nd and 3rd source wafer manufacturing just got a whole lot easier, once the masses figure out how to connect them together and package it all up. It may even change the profitability of foundries. It is a lot easier to manufacture chiplets versus complex single die chips. Throwing chiplets to Intel and Samsung foundry to keep them in the game seems reasonable, absolutely.
Kind of insane that intel is as close to TSMC as they are. I expected it to be a larger gap given intel lacks the free money printers (aka old depreciated fabs). While 2021 is the most extreme year, to think that back in 2021 intel wafer demand (on a dollar basis) was like 30-40% of all of TSMC’s 8 and 12 inch technologies put together. Throw ontop of that the N5 revenue contribution becoming TSMS’s biggest during late 2021, and the higher revenue per wafer of N5 vs i7. Let’s just say that the 30% external is a whole heaping lot of wafers/revenue for TSMC (especially since it is mostly leading edge high revenue stuff for intel).
By intel’s own math that is up to 6 mods worth (or nearly a fully built out Ohio). With numbers like that intel products might be nipping Apple’s heels as TSMC’s largest customer.
Will losing at least a third of this business ruin TSMC? No; not even close. But this definitely feels like it would be a more impactful loss than say splitting the iPhone 6 business with Samsung.
As for Semiman’s point on needing more than half the mobile market and flat costs to break even. That logic only holds if intel nodes keep being sold close to cost. If intel gets back to even just roughly trading blows with TSMC PPA wise then intel can charge intel products more for said wafers. If they also can get to rough parity (or even advantage) on a COGS basis then margins per wafer will skyrocket. By doing this you can have revenue (and most importantly the operating income) to get to breakeven with far fewer wafers sold. The intel presentation mentioned this with most of getting to breakeven no longer being on a more expensive process that is just worse, and bringing back products on the expectation that said processes will be PPAC competive (no winning all of QCOM/Mediatek required ).
Very useful chart (perhaps we could do with more of these in some of the comment threads).
Back of a fag packet calculation:
So Intel:TSMC is currently around 1:4.
Apparently, Samsung foundry (internal + external, excluding memory) is around $20bn annually - so $5bn quarterly and 1:4 to TSMC.
Unless you're a profitable niche supplier or have a strong technology lead, that's probably too low a market share to be really competitive. For either Samsung or Intel.
But we know that 30% of Intel silicon (Intel said so recently; I assume by value ?) currently goes through TSMC. So pulling that back in house would add around $1.3bn to the Intel bar and take the same off TSMC and bring the ratio up to around 1:3.
But how large do Intel's total foundry sales need to be to become both a clear #2 and profitable ? Don't know/haven't done the work there, but let's guess it's 50% of TSMC. In that case, they'd need to make up around $3bn more in external foundry sales (I'm doing this all on 2024 numbers). While assuming that all the Intel silicon can be pulled back from TSMC.
).