News Intel says foundry business won't break even until 14A in 2027

[Admin]

Intel expects its loss-making Foundry division to reach breakeven by 2027, driven by internal adoption of its 18A process, and contributions from packaging, and mature nodes.

Intel says foundry business won't break even until 14A in 2027 : Read more

 

[acadia11]

They better get to accelerating that TL!

 

[TerryLaze]

They better get to accelerating that TL!

What's TL?!

Also if the Fabs start producing 18A for the end of the year (2025) it's pretty optimistic to say 2027 ,unless they know that they will be producing like crazy.
They invested like 50bil in the new FABs, getting that back in two years, not in revenue but actual net income would be super.

 

[dalek1234]

Intel could be setting the tone here for cancelling 18A in near future, in favour of 14A; just like they did with 20A in favour of 18A

 

[phead128]

If it's only 2 years to become profitable, why did they fire Gelsinger.

 

[TerryLaze]

Intel could be setting the tone here for cancelling 18A in near future, in favour of 14A; just like they did with 20A in favour of 18A

That would be even more insane, making that much profit of off only one node and within 2 years...

 

[User of Computers]

This wasn't news 3 years ago when Intel first announced it, and it still isn't news. Foundry is expensive and there is very little in the way of outside volume coming Intel's way.

 

[rluker5]

Note that this is specific to the foundry side of the business. The side that is getting modernized.

 

[acadia11]

What's TL?!

Also if the Fabs start producing 18A for the end of the year (2025) it's pretty optimistic to say 2027 ,unless they know that they will be producing like crazy.
They invested like 50bil in the new FABs, getting that back in two years, not in revenue but actual net income would be super.

Timeline … I’m still believing Broadcomm may gobble them up. The real is how long that bargain basement pricing will last … if they are stating 2027 before they’ll have the partners utilizing 14A in enough capacity to right their FAB business it’s a problem. Remember they aren’t talking about 18A here and they better get some real life out of that investment.

 

[User of Computers]

if they are stating 2027 before they’ll have the partners utilizing 14A in enough capacity to right their FAB business it’s a problem. Remember they aren’t talking about 18A here and they better get some real life out of that investment.

That is not what they are saying. They are saying that it will be until 2027 that Foundry will break even. Bear in mind, of course, that this number includes internal volume for Intel products. Intel will make money on 18A regardless of which partners actually use the node, as Panther Lake, Clearwater Forest, and likely Diamond Rapids will be made on 18A or variants thereof.

 

[thestryker]

Intel could be setting the tone here for cancelling 18A in near future, in favour of 14A; just like they did with 20A in favour of 18A

No that isn't happening.

These situations are dramatically different from one another, because Intel 20A is to 18A what Intel 4 is to Intel 3. That is to say the latter are expansions of the former which had limited libraries and PDKs. 14A is an entirely new process which is not a direct derivative of anything that came before. If Intel were to cancel 18A for 14A then the foundry business is dead.

 

[thestryker]

They invested like 50bil in the new FABs, getting that back in two years, not in revenue but actual net income would be super.

They're not getting back the investment in that period of time. He's talking about the foundry running in the black instead of red.

It would be insane if they were making that kind of profit off the fabs in 2 years time (TSMC had around $35b profit in 2024).

 

[acadia11]

That is not what they are saying. They are saying that it will be until 2027 that Foundry will break even. Bear in mind, of course, that this number includes internal volume for Intel products. Intel will make money on 18A regardless of which partners actually use the node, as Panther Lake, Clearwater Forest, and likely Diamond Rapids will be made on 18A or variants thereof.

Have you missed the last 10 years? Intel as a supplier to Intel is a losing proposition, it’s not enough … and mind you with Panther Lake they still require TSMC. To have a viable foundry business they must become a major supplier of other people’s chips. Yes it made sense as a business model 30 years ago it doesn’t make sense today … AMD couldn’t compete with Intel until they built their own FABs in the 90s , they eventually spun off their foundry business to what became Global Foundries … the investment for today’s advanced manufacturing at the leading edge is too costly if your customer is essentially only yourself. Not to mention x86s pressure it’s feeling from ARM based chips and also the generally move specific use case ISA like GPUS and other ICs. Have you at all looked into why Intel is the position it got itself into?

Intels retailer business will make money but their foundry business will continue to lose heaps of money if they only supply to themselves.

 

[TerryLaze]

They're not getting back the investment in that period of time. He's talking about the foundry running in the black instead of red.

If they are talking about covering the running costs then they would have said it in that way.
Breaking even is a very specific term, and they said that intel hopes that the foundry unit will break even, meaning for main intel the foundry will be a zero cost within that time.

It would be insane if they were making that kind of profit off the fabs in 2 years time (TSMC had around $35b profit in 2024).

So a potential of 70bil in 2 years...

Have you missed the last 10 years? Intel as a supplier to Intel is a losing proposition, it’s not enough …

Have you missed the last ten years?!

To have a viable foundry business they must become a major supplier of other people’s chips.

Why?!
The foundry side gets the same amount of money no matter who makes chips there.
On top of that intel can make profits from selling the chips and so gets the money from both sides.
The only question is if intel has enough things to produce to keep the foundry busy all the time. Which is a completely different thing.

the investment for today’s advanced manufacturing at the leading edge is too costly if your customer is essentially only yourself. Not to mention x86s pressure it’s feeling from ARM based chips and also the generally move specific use case ISA like GPUS and other ICs. Have you at all looked into why Intel is the position it got itself into?

What position are they in right now?!
Also they have long partnerships with arm and risc-v and can produce for both these ISAs.
Intel has a GPU division and can make GPUs for servers and desktop.
They can also make GPUs for nvidia and amd, nvidia at least loves money and has an issue with making enough GPUs to sell.

 

[thestryker]

So a potential of 70bil in 2 years...

For TSMC absolutely, but no chance at all for Intel.

If every machine Intel has was running 24/7 they couldn't come close to matching TSMC's capacity.

If they are talking about covering the running costs then they would have said it in that way.
Breaking even is a very specific term, and they said that intel hopes that the foundry unit will break even, meaning for main intel the foundry will be a zero cost within that time.

They're absolutely talking about operational costs and not covering past losses. Foundry was in the red to the tune of ~$2.3b this quarter and was ~13.4b last year. They are not turning those losses into profits high enough to cover last year's losses and the rest of the losses this year (no new products on Intel nodes until the end of the year) in the following 2 years.

 

[TerryLaze]

They're absolutely talking about operational costs and not covering past losses. Foundry was in the red to the tune of ~$2.3b this quarter and was ~13.4b last year. They are not turning those losses into profits high enough to cover last year's losses and the rest of the losses this year (no new products on Intel nodes until the end of the year) in the following 2 years.

Those aren't losses in that sense.
That's the money intel spend on building the foundries.
Once the FABs are finalized intel won't have to spend 10-15 bil per year on them anymore.

 

[thestryker]

Those aren't losses in that sense.
That's the money intel spend on building the foundries.
Once the FABs are finalized intel won't have to spend 10-15 bil per year on them anymore.

Seeing as construction in Ohio is still ongoing and the fab isn't expected to go live until around 2030 now any of those costs being folded into Foundry financials aren't going anywhere. Intel currently has utilization rate issues due to shifting to contract services. The only client part being made on an EUV node (until PTL on 18A) is MTL refresh. Fab 52 is spinning up for volume production, but is making no revenue. D1X has two High-NA machines generating no revenue and whichever machines are allocated for 18A are also generating no revenue.

Intel's foundry costs aren't particularly high with ~$1b operating expenses and ~$6b cost of sales (this would be the area any construction costs should appear), but with with low revenue (~$4.7b) that doesn't really matter. With the launch of 18A that revenue should go up a fair bit, but we won't see any of that on the books until Q4 at the earliest. Getting the UMC node finished is also extremely important for DUV utilization which may also be part of the 2027 expectation to be running in the black.

 

[dalek1234]

...If Intel were to cancel 18A for 14A then the foundry business is dead.

That's my concern

 

[Gururu]

I think the leadership now is probably more the under promise/over deliver type, so I think 2027 is pretty good news. Plenty of time to carefully purchase stock.

 

[User of Computers]

Intel as a supplier to Intel is a losing proposition, it’s not enough

Is it? I'd call a fab division that works so well that it can start producing chips for other people a pretty good proposition for the products side of the company. Having an internal leading-edge fab be in complete lockstep with your design process is any designer's dream. Clearly the arrangement worked out pretty well, until the fab team got too aggressive for their budget.

 

[thestryker]

Clearly the arrangement worked out pretty well, until the fab team got too aggressive for their budget.

It was likely never going to work for the DUV to EUV changeover. Intel relied on sunsetting nodes pretty quickly, but keeping old nodes for a generation or two for lower margin parts. The problem with the shift to EUV is that there are different fab requirements (as in physical) and brand new machines for everything. Building a new fab is basically required for volume production to switch over, but then they'd end up with idle fabs which is a huge money loser.

Had they tried to keep going without opening up the fabs for contract work they'd have had to outsource all leading edge production during construction.

Having an internal leading-edge fab be in complete lockstep with your design process is any designer's dream.

It is right up until something goes wrong. Intel's designs had been tied to nodes because of this which is why there were so many Skylake revisions. Intel's nodes were all designed using internal tools which meant things also changed node to node more than if they'd been using industry standard (see Cypress Cove as an example of the mess that happened backporting).

For all of Intel's EUV nodes they're using industry standard tools and processes. This doesn't mean they can't optimize the node for what's best for Intel's product team. All it really means is that designs are far more portable than in the past so the design team isn't hitched to a specific Intel node.

 

[acadia11]

Is it? I'd call a fab division that works so well that it can start producing chips for other people a pretty good proposition for the products side of the company. Having an internal leading-edge fab be in complete lockstep with your design process is any designer's dream. Clearly the arrangement worked out pretty well, until the fab team got too aggressive for their budget.

Yeah so did kerosene before we invented electricity. It won’t work today as a business model, because it’s too expensive to develop leading edge technology today at the cadence of today’s node changes as well as the complexity of today’s chip design with only one customer in mind. It’s not a matter of budget it’s matter leading edge is much more expensive today as are the tools, materials, solutions because you are working on much much smaller scale … the engineering to make ICUs at today’s leading edge transistor sizes are prohibitively costly. That’s why there are only 3 leading edge companies in the world today. Intels foundry business is not sustainable if Intel is the only client . Actually, I’m not sure why the debate the reasons for Intels financial issues are well known.
And not to mention Intels share of the market from products standpoint is smaller … and that x86 or general purpose CPUs are giving way to specialized processors built for specific use cases … example being AI GPUs… the bottom line Intel has not proven to be a serious partner FAB yet.

 

[TerryLaze]

Actually, I’m not sure why the debate the reasons for Intels financial issues are well known.

Yes we all know the reason, intel is building FABs....
and they are building them because they know that that is still going to be cheaper for them than to keep giving their money to tsmc.

Also:
They just revealed their new AI GPUs the other day at computex.
They could fill all their FABs with this card, sell them way cheaper than nvidia, and still make a giant profit...

View: https://www.youtube.com/watch?v=vZupIBqKHqM

 

[acadia11]

Yes we all know the reason, intel is building FABs....
and they are building them because they know that that is still going to be cheaper for them than to keep giving their money to tsmc.

Also:
They just revealed their new AI GPUs the other day at computex.
They could fill all their FABs with this card, sell them way cheaper than nvidia, and still make a giant profit...

View: https://www.youtube.com/watch?v=vZupIBqKHqM

No, they are building them because they have a new leading edge process node 18, they are building them because they lacked the ability in their current FABs through a lack of standardization and configuration to act as a proper external partner FAB,and to attempt to remove the need TSMC for future builds. However, that move in itself is because Intel didnt invest early on the latest tools and doesn’t simultaneously get the long term benefit out of process node investment from trailing edge techs because they didn’t use the process nodes long enough

Yes we all know the reason, intel is building FABs....
and they are building them because they know that that is still going to be cheaper for them than to keep giving their money to tsmc.

Also:
They just revealed their new AI GPUs the other day at computex.
They could fill all their FABs with this card, sell them way cheaper than nvidia, and still make a giant profit...

View: https://www.youtube.com/watch?v=vZupIBqKHqM

i started a long winded response , and suggest you read some prospectus on Intel, also understand their lateness and pour decisions that lead to delayed switch to EUV machine tools and then finally why Intel regardless of product as a FAB manufacturer would fail to recoup the CAPEX expenditure of keeping up as leading edge node FAB provider if they were the main client. The math simply doesn’t work … on the retail side yes Intel pays TSMC but that’s half the story understand why they had to go to TSMC in the first place.. Do your own research. The bottom line Intel foundary relying on Intel as its main client is a failing business model in today’s CHIP industry.