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!
I think it is more instructive to look at what level of wafer starts would be needed. Using $5B per year as the requirement based on Zinsner's recent comments and assuming that 1/2 of this is coming from packaging, UMC and Tower that leaves $2.5B from external 18A orders. Using the reported cost of $30K per TSMC 2nm wafer and knocking off 20% to make it attractive and you need 8680 wafer starts per month.
14A really needs to win some designs and that means it needs to deliver PDK 1.0 well before TSMC A16. For Intel Foundry to break even 14A needs to be in HVM? 2027 seems a bit early. I say set the break even goal post at 2030 and beat it.
I think Intel realizes this as they can't follow TSMC's release and be successful given they recently pushed out their 0.5 PDK.
Do you know when we should expect to see TSMC's PDK 1.0? I'm also curious as to why you think that Intel has to be "well before" rather than just at the same time.
I think it is more instructive to look at what level of wafer starts would be needed. Using $5B per year as the requirement based on Zinsner's recent comments and assuming that 1/2 of this is coming from packaging, UMC and Tower that leaves $2.5B from external 18A orders. Using the reported cost of $30K per TSMC 2nm wafer and knocking off 20% to make it attractive and you need 8680 wafer starts per month.
If Intel can't demonstrate to any significant external customers, that 18a is a viable alternative path to TSMC's N2, then it's safe to assume that 18a is not performing as "some of us" have been lead to believe.
Further supporting that 14a will be delayed, and as @Daniel Nenni recently stated, "If Intel wants to be in the foundry business 14a cannot be delayed"
I'd go a bit further than Nenni, and say, that if 18a has failed to attract any significant external customers, then 18a is in trouble, and everything after is suspect.
This sounds like, if I understand LBT from commentary here and elsewhere, a “resetting of expectations” by DZ. Seems widely reported that the lens through which PG presented the Foundry roadmap and progress just was too, let’s say, optimistic.
But I want to be real here and get down to brass tacks: are these statements today actually materially different from what DZ has said previously?
If Intel can't demonstrate to any significant external customers, that 18a is a viable alternative path to TSMC's N2, then it's safe to assume that 18a is not performing as "some of us" have been lead to believe.
That is one interpretation, but not the only one. Another way to look at is based on comments Daniel Nenni has made is that by the time Intel released their PDK 1.0 the design cycle had already chosen TSMC since their PDK was released earlier and was more mature.
Further supporting that 14a will be delayed, and as @Daniel Nenni recently stated, "If Intel wants to be in the foundry business 14a cannot be delayed"
I'd go a bit further than Nenni, and say, that if 18a has failed to attract any significant external customers, then 18a is in trouble, and everything after is suspect.
If poor performance was the only possible explanation I might buy that, but there are other reasons for slow external 18A adoption and until we see how Panther Lake performs I don't believe an objective observer would make that extrapolation.
That is one interpretation, but not the only one. Another way to look at is based on comments Daniel Nenni has made is that by the time Intel released their PDK 1.0 the design cycle had already chosen TSMC since their PDK was released earlier and was more mature.
No argument on the can't afford a delay part of this. Intel needs to deliver the 14A PDK no later than the TSMC A16 PDK at the very least.
If poor performance was the only possible explanation I might buy that, but there are other reasons for slow external 18A adoption and until we see how Panther Lake performs I don't believe an objective observer would make that extrapolation.
Intel’s 18A required on-time delivery and strong performance to win external customers, who’d then endorse it, attracting more to order 14A and revive Intel’s foundry
They failed
TSMC’s lead is unassailable unless Intel pulls a miracle, like yesterday
Nah. There is an art to comms as a leader of a massive company. It's a mix of fiduciary duty, internal morale boosting, customer sales, and Street sales. You of course can and should come to your own conclusion on your interpretation of it (as you make readily apparent).
Intel’s 18A required on-time delivery and strong performance to win external customers, who’d then endorse it, attracting more to order 14A and revive Intel’s foundry
PG did say he bet the company (being Product + Foundry) on 18A. Seems half of that has not and likely will not play out. Increasingly clear why he got the boot.
I personally have faith that the timelines of leading edge semiconductor design is very long, and so 18A having customers committed in 2025 to tape-outs in 2026/2027 was always a HUGE stretch goal, versus it being an abject failure.
Do I put my money where my mouth is and bet on late 18A customer commitments and this still entirely unproven (and undesigned) 14A PDK? Hmmmmm.
Intel’s 18A required on-time delivery and strong performance to win external customers, who’d then endorse it, attracting more to order 14A and revive Intel’s foundry
No Intel needs more than great technology. That is why their previous foundry efforts floundered. They need to build the support system around that technology to enable it that TSMC spent decades developing. Anyone with two brain cells to rub together could figure out 18A on it's own is not enough. It is a necessary, but not sufficient requirement. Intel still has much to learn about how to provide the kind of customer experience that wins customers and Lip-bu Tan has the experience to lead Intel down that path.
To meet Gelsinger's promises and unrealistic expectation?, Yep, they did. As a foundry? I'm not convinced the game is over just yet as much as you may want it to be.
Many would have said that about Intel at 14nm, but we know how that played out. Never say never.
And Intel doesn't have to push TSMC from the top of the hill to win. They just need to bring in enough volume to pay for leading edge development. There is plenty of demand to satisfy TSMC and still provide Intel what they need. Especially if Intel can stretch out the tail on their process nodes and bring in revenue on depreciated processes.
No Intel needs more than great technology. That is why their previous foundry efforts floundered. They need to build the support system around that technology to enable it that TSMC spent decades developing. Anyone with two brain cells to rub together could figure out 18A on it's own is not enough. It is a necessary, but not sufficient requirement. Intel still has much to learn about how to provide the kind of customer experience that wins customers and Lip-bu Tan has the experience to lead Intel down that path.
Intel lost after 14nm and Samsung lost Apple due to failure. So if TSMC fails, then Intel will be able to jump in. That and price leverage is why almost all companies are "looking" at Intel. Intel will have technologies.... we will see who buys it. At this point it looks like we will not see "significant" revenue until well after 2027.... so they question is whether the financials work. And remember revenue/profits get shared with Brookfield, Apollo, UMC, Tower based on the current agreements.
Lets see how the 2026 numbers look for 18A cost and foundry P&L.... and how those 18A prepayments are coming in.
@Daniel Nenni
"David also mentioned a push for more Intel content in future chips meaning less TSMC."
He also clearly said intel will outsource 20-30% going forward.
I can assure you that Intel products group likes working with TSMC. It is likely that Intel will outsource more to TSMC/Samsung than they sell to others through 2030.
Here's the thing, prior to the big Intel agreement that Bob Swan signed Intel was already outsourcing to TSMC for Altera, Mobileye, mobile stuff, etc... If I remember correctly it was 20-25% of their total chip shipments? Does anyone remember this?
So when Intel says they will outsource 20-30% what does that all include? Just CPU/GPU? Or all of the other stuff combined? We need some transparency here.
Here's the thing, prior to the big Intel agreement that Bob Swan signed Intel was already outsourcing to TSMC for Altera, Mobileye, mobile stuff, etc... If I remember correctly it was 20-25% of their total chip shipments? Does anyone remember this?
So when Intel says they will outsource 20-30% what does that all include? Just CPU/GPU? Or all of the other stuff combined? We need some transparency here.
I had heard it as 30% of the wafer area? Since there is no chip... only chiplets going forward that is misleading as well. and then there is the inactive base tile. (someone tried to claim the base tile was internal area).
I would recommend as area of chiplets total .... no passive base die... none of the dummy die.
You can also do some math on the cost of sales internal and external to help understand spending.
they will spin it how they want. but the newest volume chips (meteor, arrow, lunar) are 75% external. All datacenter CPUs are internal. I believe Intel 7 products (2023 products) are still the highest volume by far (raptor lake+sapphire rapids+ etc) and those are all internal.
