And then, you know, just within that, given the plans, you know, at least the the contingency plans to maybe not do 14 a, I mean, I guess, this suggested if 14 a dies, does the foundry strategy die with it? And can you run a sustainable business just on internal volume with 18A and increasing outsource if 14A doesn’t do what it needs to do?
I agree this was a very good question to which LBT mumbled some generic answer only for Dave to intervene and answer.
- If 14A development is cancelled, Intel would have to outsource its CPU tiles to TSMC on notebooks, desktops and server (why? because AMD & Qualcomm are going to be on A14 for their chips by then, so if Intel wants to compete, then they have to)
- Right now, only a niche LNL & low volume of ARL-H & ARL-S are on TSMC, Xeons are still on Intel nodes. Rest are all Intel 7 or Intel 3 based notebooks. Intel said they are outsourcing about 30% of their wafers to TSMC. So, we can say that outsourcing is going to be >>30% wafers if 14A is dead in 2028.
- So now 30% of wafer outsourcing led to fab underutilization and foundry operating margins tanking from -20% in 2021 to -60% in 2024. Intel is saying 18A ramped to full volume and bringing most of the outsourced wafers back to Intel will lead them to breakeven in 2027 on Intel volume alone + some contribution from Intel 16, 12 & adv packaging.
- Foundry node trajectory - Intel 16 is ready, Intel 12 ready in 2027. Intel 6 likely ready in 2028+. Intel 18A/18A-P ready in 2025/2027.
- Idle fabs when 14A dies (2028/2029), Most of Intel 7, Intel 14 & Intel 3 because no one else can use it anymore. Let's assume Intel 3 will have uses for Intel's own products for base tiles etc. Intel 18A\18AP will have partial use as well but since most of the volume was Intel in 2027, there is a big hole in 18A wafer volume.
So, they write down all the idle fab on Intel 7 & 14, maintain underutilized Intel 4/3 fab (1 fab in ireland) and underutilized 18A fabs (fab 42/52) and outsource >>30% of wafer volume to TSMC and expect everyone to believe that economics work? So >3 leading edge EUV wafer fabs underutilized, atleast 1 Intel 6 fab (prev 7nm) & 1 Intel 16 fab (prev 14nm).
Counter argument from David Zinsner - we will win more customers for 18A along the way to make up for the outsourced Intel volume, would that customer be as large as Intel to replace most of Intel's outsourced wafer volume? If not the 18A fabs will remain underutilized. [SO, Intel does think they have capability to win customers on a node after its own use 2 year down the line to recover R&D costs - why that thinking does not extend to 14A - no clue]
So, a bunch of fabs that are underutilized (losing money as they are fixed costs), and a significant portion of your wafer volume is outsourced to TSMC. Gee! I wonder what that situation reminds me of? The answer is Intel in 2024 but only the outsourcing plan is a stop gap.
LBT should make up his mind if he wants to be committed external foundry or take Intel fabless. This strategy as described now sounds more like one leg on a boat and the other on the shore and I don't see how that works financially unless they can win considerable external customers which begs the question then why not win customers for 14A too (even on wave 1 or 2 or 3 to recuperate the R&D cost)?
