News Intel will keep using TSMC's services even when 18A is ramped up: 'It is a good supplier

[Admin]

Intel is set to continue outsourcing some of its products to TSMC even when 18A and other advanced nodes are fully ramped.

Intel will keep using TSMC's services even when 18A is ramped up: 'It is a good supplier : Read more

 

[bit_user]

"I think a year ago we were talking about trying to get [usage of TSMC] to zero as quickly as possible, but that is no longer the strategy," said John Pitzer, vice president of corporate planning and investor relations at Intel

I didn't expect they'd quit sourcing from TSMC overnight, but this announcement is a worrying sign.

"We think it is always good to have at least some of our wafers with TSMC. They are a great supplier. This creates healthy competition between them and Intel Foundry."

That doesn't make much sense, at a time when IFS is still contributing to their own bottom line. It's not as if IFS doesn't already face enough competition from them.

Such controllers usually rely on trailing nodes that Intel currently lacks (since Intel’s 14nm and 22nm process technologies were designed for CPUs and use proprietary IPs). So the use of TSMC and other foundries to make them is a good fit.

Intel used to make its chipsets on its own older nodes. They started outsourcing those sometime around 2019, I think, due to a major supply constraint on their 14 nm production.

 

[Gururu]

How are they going to pay TMSC they are bankrupt.

 

[helper800]

How are they going to pay TMSC they are bankrupt.

Are they actually bankrupt or are you using a turn of phrase? That would be news to me.

 

[phead128]

What's stopping potential customers from bypassing the middle-men to source from original TSMC to begin with. Intel has to be more reliable, cheaper, and better yields than TSMC to get customers to switch.

 

[JamesJones44]

I didn't expect they'd quit sourcing from TSMC overnight, but this announcement is a worrying sign.

If it is low volume products like their GPUs or tile components like the I/O tile (just using it as an example) it might make sense to use a 3rd party instead of building/expanding capacity which may go under utilized. However, if they start letting TSMC compete with IFS on price, performance, etc., which seems like that may be the angle, then I 100% agree it could be a very bad sign.

 

[2Be_or_Not2Be]

What's stopping potential customers from bypassing the middle-men to source from original TSMC to begin with. Intel has to be more reliable, cheaper, and better yields than TSMC to get customers to switch.

If IFS could produce enough quantity and the client's usage is US-based, then it could be worth a bit more to the client to have local production (shipping, logistics, etc.). IFS really needs to get rolling, and it doesn't need its own PR people to say things that lessen them as first-look contract producers.

 

[thestryker]

Intel used to make its chipsets on its own older nodes. They started outsourcing those sometime around 2019, I think, due to a major supply constraint on their 14 nm production.

Intel has never used TSMC for their chipsets as far as I'm aware (maybe mobile, I'm less tapped into that market). They started using older chipsets anywhere they could though and increased usage of older nodes. There were also some chipsets manufactured on Intel 16 (22nm) after they'd stopped using it.

 

[thestryker]

The biggest problem Intel has is a complete lack of a competitor for N7 (and its derived nodes). While Intel 7 is performance/density competitive enough it's very expensive and can't be used for external customers. That gap is one they don't have any plans to fill as the available nodes will be Intel 12 in ~2027 (the UMC node) and then Intel 3. This is the gap that Intel's GPUs/tiles, SoC and IO tiles have lived in*. I imagine this won't change any time soon as Intel would need available EUV capacity and it's possible using TSMC for these is more cost effective.

Intel also has chips being made at TSMC which were always made there from companies they've acquired. It never made sense in the past for them to port over to their own nodes so that also likely wouldn't change until Intel has EUV capacity. Of course the other problem with EUV capacity is that they've kicked the can down the road on their biggest EUV expansions. Servicing the highest profit margin parts is always going to come first in the fab game so while they may have plenty of capacity for internal and external customers it's believable they wouldn't for low margin parts.

*They did leverage Intel 7 for the GNR IO tiles, but I think that's more about keeping everything in house and the DUV fabs active than being the best move fiscally.

 

[TerryLaze]

I didn't expect they'd quit sourcing from TSMC overnight, but this announcement is a worrying sign.

It would be more worrying if intel would out of spite build FABs for old nodes just to not buy them from tsmc...maybe they could look into getting them from glofo.

What's stopping potential customers from bypassing the middle-men to source from original TSMC to begin with. Intel has to be more reliable, cheaper, and better yields than TSMC to get customers to switch.

People aren't tsmc users out of fandom, they don't use them because they love them, so IFS just has to be as reliable, cheap, and yield-y , even just the ballpark and not even just as good.
Throw in the tariffs and now many companies are restricted in how much product they can get out of tsmc, even more than before, and tsmc will have the US expenses that will increase their prices even more, again.

TSMC US is booked out into late 2027, what do you think will these companies do that can only get orders in for 2028 and later?!
They are going to wait three years for their products to become obsolete just because they don't want to use intel?!

 

[Gururu]

Are they actually bankrupt or are you using a turn of phrase? That would be news to me.

No they are not, I just wanted to jump in on the bashing.

 

[Thunder64]

It would be more worrying if intel would out of spite build FABs for old nodes just to not buy them from tsmc...maybe they could look into getting them from glofo.
People aren't tsmc users out of fandom, they don't use them because they love them, so IFS just has to be as reliable, cheap, and yield-y , even just the ballpark and not even just as good.
Throw in the tariffs and now many companies are restricted in how much product they can get out of tsmc, even more than before, and tsmc will have the US expenses that will increase their prices even more, again.

TSMC US is booked out into late 2027, what do you think will these companies do that can only get orders in for 2028 and later?!
They are going to wait three years for their products to become obsolete just because they don't want to use intel?!

What IFS needs is a history of success with external customers. They don't have that. Therefore, companies will use TSMC and pass on additional costs to consumers.

 

[PixelAkami]

No they are not, I just wanted to jump in on the bashing.

Cringe.

 

[MXM0]

This isn't news. Intel has always stated that this is the purpose of the chiplet design since 12th gen. The IO and GPU chiplets, in particular, don't need to be 18A, as yet, so they may as well use cheaper fabs, and use the 18A fabs for the CPUs.

A consumer "Panther Lake" or "Nova Lake", will very likely have a 18A Intel fabbed CPU connected to a N4, or N3, "Celestial" GPU, and other chiplets made at TSMC, or heck maybe TI, or Samsung, for all we know.

Saving money is the whole point, and 18A space is premium space. The days of single chip packages are long since over.

 

[acadia11]

ok this is a bit confusing considering Intel is focused on becoming a more viable leading edge provider? I’m assuming if it’s older node tech it’s less troubling but if you want to be a foundry and can’t produce your own chips it a just not a good look. At this point Intels saving grace is their massive amounts of IP and being an American company.

 

[jackt]

hmmm...

 

[MXM0]

ok this is a bit confusing considering Intel is focused on becoming a more viable leading edge provider? I’m assuming if it’s older node tech it’s less troubling but if you want to be a foundry and can’t produce your own chips it a just not a good look. At this point Intels saving grace is their massive amounts of IP and being an American company.

I think you're misinterpreting what I stated. Intel is using 18A for the CPU. Even nVidia is still using N4/N5 for their GPUs. No one is going to shrink a GPU down to 18A anytime soon, and if someone can, and wants to hand Intel huge money to do it, it'd be nVidia.

Integrated GPUs are *not* premium parts. There's not a good reason to waste 18A wafer space on it, plain and simple. Intel is being *very* smart by NOT doing that. They are much better off selling those wafers to nVidia and Broadcom.

AMD processors are ALSO built this way, using cheaper process nodes for some of the chiplets -- in fact, the iGPUs are exactly that.

 

[Notton]

TSMC doesn't only make chips.
They have CoWoS.
IDK if TSMC changed their position on CoWoS, but last I heard they don't want to integrate chips made elsewhere with CoWos.
My guess is because there's some deeply guarded IP secrets in gluing all the chips together.

 

[acadia11]

I think you're misinterpreting what I stated. Intel is using 18A for the CPU. Even nVidia is still using N4/N5 for their GPUs. No one is going to shrink a GPU down to 18A anytime soon, and if someone can, and wants to hand Intel huge money to do it, it'd be nVidia.

Integrated GPUs are *not* premium parts. There's not a good reason to waste 18A wafer space on it, plain and simple. Intel is being *very* smart by NOT doing that. They are much better off selling those wafers to nVidia and Broadcom.

AMD processors are ALSO built this way, using cheaper process nodes for some of the chiplets -- in fact, the iGPUs are exactly that.

We are discussing two different things. I’m referring to Intel wanting to supply fab services not for their own CPUs but for the market. The article mentions TSMC is better suited to be a supplier for trailing - edge older die technology, I’m finding it interesting they are lagging in trailing edge capabilities and still vying to be a fab service supplier that’s the confusion? It would be like global foundries hiring TSMC to provide fab services. It doesn’t make sense in my mind of the long term vision of Intel at least stated?

The point as a company Intel 18A may be great but Intel can’t be the only customer to make it a success. And it’s being said they don’t have the capability in trailing edge tech … so where is the revenue going to come from for Intel as fab company. It makes sense now why the fab side of the business was losing so much money.

 

[Trake_17]

"We think it is always good to have at least some of our wafers with TSMC. They are a great supplier. This creates healthy competition between them and Intel Foundry."

If this statement isn't a lie, it likely means that they aren't planning to continue owning their own foundry. In the event that they cease to own a foundry, then competition in the foundry space is good for Intel. If they plan to stay in the foundry business, then this statement is just a lie. If it's a lie, you gotta ask what they're hiding, which really can only be a lack of faith in their own foundry, which is an awful sign. Either way, Intel's long term business prospects are looking poor.

 

[acadia11]

We are discussing two different things. I’m referring to Intel wanting to supply fab services not for their own CPUs but for the market. The article mentions TSMC is better suited to be a supplier for trailing - edge older die technology, I’m finding it interesting they are lagging in trailing edge capabilities and still vying to be a fab service supplier that’s the confusion? It would be like global foundries hiring TSMC to provide fab services. It doesn’t make sense in my mind of the long term vision of Intel at least stated?

The point as a company Intel 18A may be great but Intel can’t be the only customer to make it a success. And it’s being said they don’t have the capability in trailing edge tech … so where is the revenue going to come from for Intel as fab company. It makes sense now why the fab side of the business was losing so much money.

Hence my statement their saving grace is massive amounts of IP and being American.

 

[TerryLaze]

What IFS needs is a history of success with external customers. They don't have that. Therefore, companies will use TSMC and pass on additional costs to consumers.

Ahhh, make products more expensive so that people buy them more....makes perfect sense.

I’m finding it interesting they are lagging in trailing edge capabilities and still vying to be a fab service supplier that’s the confusion?

That's why chiplets exist, intel can produce the small nodes for a company and that company can source the older nodes from wherever they like and intel can put them all together.

The point as a company Intel 18A may be great but Intel can’t be the only customer to make it a success. And it’s being said they don’t have the capability in trailing edge tech … so where is the revenue going to come from for Intel as fab company. It makes sense now why the fab side of the business was losing so much money.

Intel has been doing well money wise up until now so the fabs are worth their money for intel.
Also the fab side was not losing money, intel was, and is, and will be, investing money into the fab side.
I mean when you do your taxes and you bought something for your business don't you declare that expense in your tax form?
That's what's happening here, they are putting their expenses in their tax form to get tax refunds or whatever their reasoning might be.

 

[priyansh1]

Completely agree but have a look, Intel has made it clear that it will continue to rely on TSMC's services even as it ramps up its own 18A node technology. Despite Intel's plans to enhance its in-house capabilities, the company recognizes TSMC as a valuable and dependable partner in the semiconductor manufacturing space.

Intel’s decision to keep working with TSMC comes as no surprise, considering the strong track record the foundry has in producing high-quality chips. While Intel is pushing ahead with its own 18A technology, which promises significant improvements in performance and power efficiency, TSMC’s expertise and proven ability to handle advanced semiconductor processes remain crucial for Intel's overall strategy.

This partnership reflects the growing importance of having a multi-supplier approach in the semiconductor industry. By working with both TSMC and its own fabrication facilities, Intel ensures it can meet demand and stay competitive. It also highlights TSMC's leadership in manufacturing, particularly with its cutting-edge technologies like 5nm and 3nm processes.

In short, Intel’s continued relationship with TSMC shows that, even as it advances its own tech, TSMC remains a key player in the semiconductor landscape, and the two companies will likely continue benefiting from this collaboration moving forward.

 

[Thunder64]

Ahhh, make products more expensive so that people buy them more....makes perfect sense.

Not my fault if you don't like to hear the truth.

 

[acadia11]

Ahhh, make products more expensive so that people buy them more....makes perfect sense.

That's why chiplets exist, intel can produce the small nodes for a company and that company can source the older nodes from wherever they like and intel can put them all together.

Intel has been doing well money wise up until now so the fabs are worth their money for intel.
Also the fab side was not losing money, intel was, and is, and will be, investing money into the fab side.
I mean when you do your taxes and you bought something for your business don't you declare that expense in your tax form?
That's what's happening here, they are putting their expenses in their tax form to get tax refunds or whatever their reasoning might be.

uhm … I think you need to look back at their books this isn’t a matter of just book keeping but a result of bad decisions, poor execution, and delays. It’s precisely because they didn’t invest in new equipment until late in the game that while it reduced their capex expenditures increased their opex expenditures considerable compared to their competitors and resduced their margins. And Intel has never been setup to be large supplier to other companies for fab services … so if they aren’t producing for themselves or outsourcing as has been the case since alder lake their foundry business is a bust. You don’t lose 7 billion in 1 year and say oops tax write off.

The combination of reduced pc market, pressure from AMD in data center space and the lack of competitive products in the AI space is why Intel is where it is. With all that said they still power a huge swath of the global computing power and own massive amounts of IP. Intel can’t just be replaced and demand can’t just be picked up by competitors but their foundry needs a complete overhaul or spinoff. The news here is at odds with companies direction if it’s trying to right the ship.