AMD Exec Says Intel IFS is Destined to Fail

[delong.height]

It takes minimum $20 of annual foundry capex (and growing) to stay on leading edge. $20b in revenue is a fraction of what is needed. I would guess ~$50b in revenue would be bare minimum to support a leading edge foundry business.

there will be government subsidy, prepayment from customers, and joint partnership with companies like Brookfield. And 20B is a fraction of what is needed. but IFS is still an Intel company. So, it will get the support from Intel for sure.

 

[Daniel Nenni]

It takes minimum $20 of annual foundry capex (and growing) to stay on leading edge. $20b in revenue is a fraction of what is needed. I would guess ~$50b in revenue would be bare minimum to support a leading edge foundry business.

IDM Foundries like Intel and Samsung leverage internal products for the CAPEX justification. Otherwise, no it would not be feasible to be a leading edge pure play foundry in this day and age. TSMC really is the last of its kind.

 

[count]

IDM Foundries like Intel and Samsung leverage internal products for the CAPEX justification. Otherwise, no it would not be feasible to be a leading edge pure play foundry in this day and age. TSMC really is the last of its kind.

there will be government subsidy, prepayment from customers, and joint partnership with companies like Brookfield. And 20B is a fraction of what is needed. but IFS is still an Intel company. So, it will get the support from Intel for sure.

Agree, but government isn't going to give Intel the missing $30b every year. What I'm saying is that the way things are trending, there isn't going to be enough revenue from leading edge for 3 players to be successful.

These means one of a few outcomes. Samsung will fall off leading edge, or IFS will fail, or maybe as someone suggested Samsung Foundry and IFS will enter some kind of strategic alliance.

Personally I think it makes more sense for Intel to spin off IFS, maybe sell it to Samsung along with a wafer commitment and focus on design.

 

[Daniel Nenni]

Agree, but government isn't going to give Intel the missing $30b every year. What I'm saying is that the way things are trending, there isn't going to be enough revenue from leading edge for 3 players to be successful.

These means one of a few outcomes. Samsung will fall off leading edge, or IFS will fail, or maybe as someone suggested Samsung Foundry and IFS will enter some kind of strategic alliance.

Personally I think it makes more sense for Intel to spin off IFS, maybe sell it to Samsung along with a wafer commitment and focus on design.

True but remember it is not just leading edge. It is < 14nm since UMC and GC failed at FinFET so there is room for 3. Personally I think Samsung and Intel run more on ego than CAPEX. At the Samsung foundry events they always remind us what they were first at even if they didn't yield. Intel does that now as well. They have quite a compelling packaging story.

The semiconductor industry on a whole knows there needs to be multiple fab sources so I think IFS will get plenty of chances, like Samsung has, absolutely.

 

[hist78]

True but remember it is not just leading edge. It is < 14nm since UMC and GC failed at FinFET so there is room for 3. Personally I think Samsung and Intel run more on ego than CAPEX. At the Samsung foundry events they always remind us what they were first at even if they didn't yield. Intel does that now as well. They have quite a compelling packaging story.

The semiconductor industry on a whole knows there needs to be multiple fab sources so I think IFS will get plenty of chances, like Samsung has, absolutely.

The semiconductor industry needs three or more leading edge foundries definitely. In order to achieve this goal IFS must be spun off from the Intel to become a truly independent company. The reason is simple: the IDM Intel can't afford to have IFS in house.

A fabless Intel can do very well with its great design, products, and diversified fab partners (TSMC, Samsung, GF, Tower, and the new independent IFS). At the same time the independent IFS can go freely to pursue whatever revenue/profit and whatever technology/services good for them.

Granted there is a transition period to guarantee the supply and revenue stream for both the new fabless Intel and the new independent IFS.

I suspect Intel and Pat Gelsinger are working toward this direction. It will potentially create two or at least one strong player(s) in the semiconductor industry. If they insist using the IDM model and to stay as one IDM Intel, they may end up to kill Intel itself.

 

[count]

True but remember it is not just leading edge. It is < 14nm since UMC and GC failed at FinFET so there is room for 3. Personally I think Samsung and Intel run more on ego than CAPEX. At the Samsung foundry events they always remind us what they were first at even if they didn't yield. Intel does that now as well. They have quite a compelling packaging story.

The semiconductor industry on a whole knows there needs to be multiple fab sources so I think IFS will get plenty of chances, like Samsung has, absolutely.

SMIC apparently making 7nm chips. I think China foundry will eventually figure out how to do EUV with or without AMSL, even if they aren't leading edge they may become the lowest cost option for 14nm to 3nm within 3-5 years.

 

[delong.height]

The semiconductor industry needs three or more leading edge foundries definitely. In order to achieve this goal IFS must be spun off from the Intel to become a truly independent company. The reason is simple: the IDM Intel can't afford to have IFS in house.

A fabless Intel can do very well with its great design, products, and diversified fab partners (TSMC, Samsung, GF, Tower, and the new independent IFS). At the same time the independent IFS can go freely to pursue whatever revenue/profit and whatever technology/services good for them.

Granted there is a transition period to guarantee the supply and revenue stream for both the new fabless Intel and the new independent IFS.

I suspect Intel and Pat Gelsinger are working toward this direction. It will potentially create two or at least one strong player(s) in the semiconductor industry. If they insist using the IDM model and to stay as one IDM Intel, they may end up to kill Intel itself.

They probably are going to do a Mobileye-like IPO, and I think that would be it. But a fully independent IFS won't be possible unless they entered into a joint partnership with Samsung and have sufficient capacity in the East as well. It simply cost more to make in US. Intel can afford it because they are IDM. But a fully independent IFS will not despite all the support from the government. Because companies, despite knowing that they should diversify their supply chain, they won't really take action until maybe last minute because this is so unpredictable as the case with Israel. It can happen tmr, or maybe it can happen five-year, ten years from now, or it may never happen.

 

[delong.height]

Agree, but government isn't going to give Intel the missing $30b every year. What I'm saying is that the way things are trending, there isn't going to be enough revenue from leading edge for 3 players to be successful.

I think that's right. It also illustrates how in 2000, there were 18 leading edge players in the space, but today it has been down to 3. If Intel chose not to join the race, we would only have 2 leading edges with one that's truly reliable.

Personally I think it makes more sense for Intel to spin off IFS, maybe sell it to Samsung along with a wafer commitment and focus on design.

I think Samsung has proven they can't be counted on for a consistent R&D excellence because their culture is so complex and maybe mistaken. The downfalls in Exynos and < 7nm process node advancement kind of illustrates that point. Plus they also have OEM in both mobile and PC segments, and memory as well. The resource must be really tight to be allocated among all these divisions. And Samsung has just been getting bad news after bad news. SK Hynix acquired Intel's memory division and being the biggest beneficiary in HBM markets. China has their own offering that's been eating their market share. The memory market is still difficult to navigate. Their Samsung phone has beaten down by Apple year after year. And they also lack years of experiences when it comes to any process node that is below 7nm. Selling IFS to Samsung and appoint Samsung as the foundry partner then count Samsung to be successful is like AMD signing GlobalFoundries as their sole foundry partner, the possibility for it to be a positive for both Intel and Samsung are highly unlikely, which might take both companies down, and making TSMC the only player in leading edge player in the world

 

[semiman]

I am quite curious about how this business landscape will unfold. Foundry business tends to start with mobile products, and then move to HPC with the given node.

- Samsung has some external mobile clients and has its own demands (smartphone) but lacks foundry volume
- Intel has competitive internal volumes but they've been focusing on HPC(CPU, GPU...etc) so they might have some disadvantages in the mobile market
- TSMC will just do fine like before

Intel can get No. 2 foundry easily if they include their internal products (CPUs) but that doesn't mean that they can get access to mobile customers(with a large IP ecosystem which can give Samsung an edge against Intel). And we have tons of things to standardize like chiplets and packaging (which might give more edges to intel, compared to Samsung?), more things integrated outside of the fab. What an interesting day...

 

[count]

Also I want to shoot down the idea that the foundry "needs" 3 players on leading edge. It needs 3 players in the same way they "need" 3 CPU architectures. In that it's not actually a need at all, chipmakers want 3 players so they can have more market power. Design has always looked down on manufacturing and can't stand the fact that when it comes to leading edge, it's a manufacturer that has the power. Even though TSMC has acquired that power by being the best partner the designers could ever hope to have.

 

[Daniel Nenni]

Also I want to shoot down the idea that the foundry "needs" 3 players on leading edge. It needs 3 players in the same way they "need" 3 CPU architectures. In that it's not actually a need at all, chipmakers want 3 players so they can have more market power. Design has always looked down on manufacturing and can't stand the fact that when it comes to leading edge, it's a manufacturer that has the power. Even though TSMC has acquired that power by being the best partner the designers could ever hope to have.

I agree, in my mind a monopoly is inherently bad but that is not always the case. One thing that I will add is that with the advent of chiplets it is much easier for customers to play foundries against each other. That really is a game changer for the foundry business that has spent the last 10 years building a moat so customers cannot 2nd and 3rd source.

Exciting times in the semiconductor ecosystem, absolutely.

 

[semiman]

I agree, in my mind a monopoly is inherently bad but that is not always the case. One thing that I will add is that with the advent of chiplets it is much easier for customers to play foundries against each other. That really is a game changer for the foundry business that has spent the last 10 years building a moat so customers cannot 2nd and 3rd source.

Exciting times in the semiconductor ecosystem, absolutely.

Chiplets are interesting in many ways. For example, Big foundries might not want smaller foundries to 'bankrupt' in the short run, because a lack of certain chiplet will result in a reduced demand of high-performance compute chiplets. Intel might not want N3 to fail because they need to use GPUs from TSMC. Samsung might not want TSMC CoWoS to fail this year, because that's the practically only way to integrate HBM(which has extremely high pin counts) to high throughput logics.

The boundary of cooperation and competition is blurring out.

 

[Barnsley]

Chiplets are interesting in many ways. For example, Big foundries might not want smaller foundries to 'bankrupt' in the short run, because a lack of certain chiplet will result in a reduced demand of high-performance compute chiplets. Intel might not want N3 to fail because they need to use GPUs from TSMC. Samsung might not want TSMC CoWoS to fail this year, because that's the practically only way to integrate HBM(which has extremely high pin counts) to high throughput logics.

The boundary of cooperation and competition is blurring out.

The big foundries have no concern for their suppliers , so it would be somewhat surprising if they didnt want smaller competitors to die