Discussion Leading Edge Foundry Node advances (TSMC, Samsung Foundry, Intel) - [2020 - 2025]

ajsdkflsdjfio said:

I don't think it's inherently an Issue that Intel is unable to rival the capacities of TSMC in the next 5 years or become the No.2 foundry by 2030 in volume. All Intel needs to do is be able to begin to turn a profit on their foundry business in the next few years and stop the bleeding market share in their products, although this is still a significant ask for Intel. Yes they need to reinvest into technologies to make sure they don't fall behind TSMC, but innovation/advancements aren't directly tied to the amount of money you dump into something. If that were the case, Nvidia and AMD would not be in the position they are in now as they were once in the shadow of Intel yet were able to compete for decades and eventually surpass Intel. Intel does need more money to "keep up" but their investments in the past 4 years have brought their technology up to snuff and built much needed capacity going forwards.

Click to expand...

I believe it is. Mostly, I believe that the amount of time it would take Intel to:

1. Create an environment of cooperation for rivals in the industry
2. Create an industry standard tool set for customers to use
3. Create a support infrastructure for onboarding customers and new designs
4. Create a loyal user base of long term customers
5. Ramp up production on leading edge processes to support said customer base
6. Develop a completely different set of processes based around risk reduction and developing iron clad roadmaps that customers feel they could take to the bank (because they literally have to!)

This time is WAY longer than they have left IMO. I don't want this to be true as I am a fiercely loyal American, but it certainly looks to be true.

ajsdkflsdjfio said:

Worries about high-NA capacity are important, but the thing is TSMC themselves don't expect to use it in actuality until around 2030+, and regardless it probably won't be as widely used in as many fabrication steps as EUV is being used today, at least not for the first one or two high-NA EUV nodes. For reference, TSMC has around 200 EUV machines which they acquired gradually over the past 3-4 years. I doubt that in a couple years, when TSMC actually begins using high-NA in their nodes, that they will require more than a few dozen high-NA machines for mass production of whatever node they are producing at that time period. Intel already has purchased 6 of ASML's high-NA EUV machines in 2024, more than TSMC, so not only are they not behind but they also have plenty of time before high-NA capacity actually becomes a factor. This is important time for Intel to get their financials in order and reap some of the benefits of their heavy investments from 2021-2024.

Click to expand...

Still, the basic problem exists. The Fab game has become so exponentially expensive that only a fab with HUGE volume from a vast array of customers can play without going bankrupt. And said company has to have a strong culture of financial conservatism and risk avoidance.

ajsdkflsdjfio said:

As for the physics and state of the market, I think there are actually many factors that benefit Intel's strategy. For example, the AI boom doesn't only far reaching consequences on AI chip designers, but also semiconductor manufacturers. TSMC prices have skyrocketed in addition to probable tariffs going forwards, allowing more headroom for Intel to price their own offerings and also more draw for TSMC alternatives especially with Samsung Foundry performing so poorly. Also although semiconductor manufacturing is getting more expensive essentially limiting the leading edge endeavors to only but a couple major players, the slowing pace of innovations and the diminishing returns of advancements means that Intel doesn't need to as rapidly innovate to keep pace with TSMC nor are they absolutely required to have the bleeding edge node. TSMC 3nm has been out for a while, yet the majority of capacity and revenue has still been in the their 5nm class nodes with 15% vs 35% revenue percentage respectively.

Click to expand...

TSMC is very risk averse. 3nm has been out for a while now, and will likely become the largest revenue generator for TSMC over the next 5 years. N2 class will be a play thing in comparison.

And as you correctly point out, TSMC will continue to service customers with 5nm class processes for the foreseeable future.

All of this has come to pass from DECADES of planning. Planning that Intel is only now recognizing is needed.

I'm not saying Intel can't do it, I am saying they have an awful long way to go. Scoring a touch down with 18A is only the VERY beginning IMO. Much much more will be needed (see above) for Intel to regain profitability.

oak8292 said:

Where do you get that Intel has purchased 6 high NA EUV machines in 2024? ASML is only producing development machines, 5000s and doesn’t expect to ship a production 5200 until sometime in 2026. At least that is what ASML said in January 2025 during the quarterly report out. I think Intel has two 5000s.

Click to expand...

Some sources say 6, some say 5, most recently they say 5. Purchased does not mean received, they simply purchased the machines that ASML were set to produce in 2024, AFAIK intel only received their first machine ~half a year ago and just recently got it operational, not sure though.

oak8292 said:

You probably need to dig into some financial statements for Intel, the 2024 Annual Report is out. They are still spending money and have obligations to their SCIPs (loan sharks) in 2025.

Click to expand...

I understand this however Intel expects to cut costs in 2025 by 10 billion (to be taken with grain of salt ofc) vs 2024, unless their debt repayments somehow significantly increases in 2025 they are still going to be much better off in terms of spending. They are also set to receive the remaining 5.66 billion of the CHIPs act hopefully this year. With more government support and a much better outlook financially regardless, I fail to see how they are going to be in such dire straits. Perhaps their products revenue will be poor in 2025 given arrow lake's failure and basically no significant DCAI/CCG products in 2025 but that's a different discussion.

OneEng2 said:

I believe it is. Mostly, I believe that the amount of time it would take Intel to:

1. Create an environment of cooperation for rivals in the industry
2. Create an industry standard tool set for customers to use
3. Create a support infrastructure for onboarding customers and new designs
4. Create a loyal user base of long term customers
5. Ramp up production on leading edge processes to support said customer base
6. Develop a completely different set of processes based around risk reduction and developing iron clad roadmaps that customers feel they could take to the bank (because they literally have to!)

This time is WAY longer than they have left IMO. I don't want this to be true as I am a fiercely loyal American, but it certainly looks to be true.

Click to expand...

I understand this takes a long time, but I don't see why there would be some sort of timer for Intel to pull this off and arguably Intel is already underway on many of these goals although not on par with TSMC. Intel doesn't need their foundry business to rival TSMC to be able to succeed. They are also a design company and have been for the their entire history. So long as Intel has enough customers to utilize the capacity they are planning to bring online in the coming years, foundry will soon break even and later on profit and begin to pay off the investment required for them. I would've been more skeptical about intel having enough customers 3 months ago, before the president was elected who plans on having significant tariffs on asian tech products and is super fixated on american made products, and before 18a was confirmed to be as impressive as it is turning out to be. There are obviously more challenges ahead, but in my eyes two huge roadblocks have just been removed in terms of whether or not Intel will have enough customers to utilize their fabs. Not only is the draw for Intel foundry much greater than just a short while ago, the draw for Intel themselves is much greater.

OneEng2 said:

Still, the basic problem exists. The Fab game has become so exponentially expensive that only a fab with HUGE volume from a vast array of customers can play without going bankrupt. And said company has to have a strong culture of financial conservatism and risk avoidance.

Click to expand...

Regardless, I also think that HUGE TSMC like volumes aren't necessarily required for a company like Intel specifically to continue to deliver leading edge node advances. In the first place, the spending for TSMC is mostly on operational/investment related capex rather than R&D. Their R&D has doubled in the past 5 years and yet is "only" (still alot) 5 billion USD for 2024. This pales in comparison to the amount of money they invest into all the equipment, land, etc etc for the huge volume they produce. OFC more volume = more revenue = more profits given the same margins, but you don't necessarily need even close to TSMC like profits to continue to match the pace of leading node innovation. Money put into R&D is less correlated with successful innovation than one might think. In addition to this, Intel themselves are still going to be a design oriented company going forward financially speaking. Foundry was the necessary investment they needed for their own products, opening up foundry was just the necessary move in order to continue to pay for nodes far down the line. With foundry operational and perhaps superior in some ways, the design team finally has a good semiconductor base to produce products on, not having to choose between inferior internal nodes and lower margins/lower volume external offerings.

ajsdkflsdjfio said:

Foundry was the necessary investment they needed for their own products, opening up foundry was just the necessary move in order to continue to pay for nodes far down the line. With foundry operational and perhaps superior in some ways, the design team finally has a good semiconductor base to produce products on, not having to choose between inferior internal nodes and lower margins/lower volume external offerings.

Click to expand...

I would say that Intel was primarily a foundry company with a side of architecture. Intel was built on Moore's Law which meant cheaper transistors. Robert Coldwell and early Intel architect talked about the benefits of Moore's Law and Dennard's Scaling versus architecture in a Hotchips presentation. The architects main role was to stay out of the way of shrinks and the resulting performance improvements. This is the mentality that Pat brought back to Intel.

The foundry business today is cost and power efficiency. The N3B node at TSMC was too aggressive on shrinks and did not yield well. They backed off to a lower cost design. The N2 node doesn't have backside power because they aren't designing for frequency but they are designing for cost. The same equipment will be used for A16 with backside power but it will probably have lower wafer volume. Because the equipment will also be used for N2 it will also probably be lower cost than Intel's A18.

Tech in the 2000's did not age well. A computer a node behind lost value rapidly. A computer two nodes behind was essentially e-waste. This was also true with DRAM on the shelf. The cost of bits reduced so rapidly that older DRAM lost value rapidly. In today's market there are people holding on to computers for 10 years without regret. There is probable still demand for DDR4. A node advantage for TSMC or Intel is not very meaningful. Intel needs to have patience and get their costs under control. TSMC sold a lot of wafers while they were trailing by a node due to diligence about costs. It's hard to know when TSMC's transistors costs actually dropped below Intel's but since Intel was not a foundry it did not matter.

oak8292 said:

I would say that Intel was primarily a foundry company with a side of architecture. Intel was built on Moore's Law which meant cheaper transistors. Robert Coldwell and early Intel architect talked about the benefits of Moore's Law and Dennard's Scaling versus architecture in a Hotchips presentation. The architects main role was to stay out of the way of shrinks and the resulting performance improvements. This is the mentality that Pat brought back to Intel.

Click to expand...

From how I see it, regardless of whether foundry was the driving force of past Intel, they didn't seriously consider using foundry to produce external products nor did they use the foundry advantage to expand into other segments like accelerators/GPUs/mobile chips, at least not seriously or effectively. Their sole goal was to deliver on their processors, not selling wafers, foundry was a means to an end, instead of focusing on optimizing cpu design they focused on improving the nodes the cpus are built on. Fast forward to the past 10 years and node progress is slowing down and getting more expensive and they are losing marketshare so they cannot afford to create entire nodes only for internal use long term, unless they want to stagnate and or go bankrupt. Regardless their money has always come from selling products because they have a whole ecosystem built around selling CPUs, not selling nodes. While I believe the foundry will profit eventually, their big bucks are still in products and will continue to be in the future.

ajsdkflsdjfio said:

I understand this takes a long time, but I don't see why there would be some sort of timer for Intel to pull this off

Click to expand...

Because they will go bankrupt or be purchased if they don't start making money.

ajsdkflsdjfio said:

Intel doesn't need their foundry business to rival TSMC to be able to succeed.

Click to expand...

I disagree. Intel has chosen to remain an IDM. They believe (incorrectly IMO) that it is less expensive for them to develop cutting edge processes on cutting edge machines and make their own chips on that process as a core advantage. When Intel fell behind on process tech, they fell behind in profit and in performance. They lost their golden goose in DC, and started losing market share on the desktop and laptop too.

ajsdkflsdjfio said:

Regardless, I also think that HUGE TSMC like volumes aren't necessarily required for a company like Intel specifically to continue to deliver leading edge node advances.

Click to expand...

They are. The cost of development being spread over fewer wafers causes EACH wafer to cost the company more money than that same wafer costs TSMC who is making more wafers on it. This puts Intel in a fundamentally poorer competitive position due to financials.

oak8292 said:

I would say that Intel was primarily a foundry company with a side of architecture. Intel was built on Moore's Law which meant cheaper transistors. Robert Coldwell and early Intel architect talked about the benefits of Moore's Law and Dennard's Scaling versus architecture in a Hotchips presentation. The architects main role was to stay out of the way of shrinks and the resulting performance improvements. This is the mentality that Pat brought back to Intel.

The foundry business today is cost and power efficiency. The N3B node at TSMC was too aggressive on shrinks and did not yield well. They backed off to a lower cost design. The N2 node doesn't have backside power because they aren't designing for frequency but they are designing for cost. The same equipment will be used for A16 with backside power but it will probably have lower wafer volume. Because the equipment will also be used for N2 it will also probably be lower cost than Intel's A18.

Tech in the 2000's did not age well. A computer a node behind lost value rapidly. A computer two nodes behind was essentially e-waste. This was also true with DRAM on the shelf. The cost of bits reduced so rapidly that older DRAM lost value rapidly. In today's market there are people holding on to computers for 10 years without regret. There is probable still demand for DDR4. A node advantage for TSMC or Intel is not very meaningful. Intel needs to have patience and get their costs under control. TSMC sold a lot of wafers while they were trailing by a node due to diligence about costs. It's hard to know when TSMC's transistors costs actually dropped below Intel's but since Intel was not a foundry it did not matter.

Click to expand...

Exactly this.

Intel needs desperately to figure out EXACTLY THIS. They seem to be hell bent on being blind to the facts that are plain to see. They can no longer maintain market dominance through better foundry service than the rest of the world.

Today, AMD is competing (and besting in most instances) Intel while using a much less expensive node process. AMD could have easily made desktop and laptop Zen 5 on N3E (like it did Turin D), but the business cases wasn't there for them to do it. The process was too expensive.

Intel, on the other hand, used an even MORE expensive process (N3B) and still didn't unseat Zen 5.

For quite some time (decades) I have fundamentally believed that Intel was primarily a world class foundry that supported a few design teams. You don't have to be a particularly talented designer to beat everyone else when you have fundamentally better foundry service that catered to your design goals to fit like a glove.

Perhaps this was a weakness in the long run. Design got slopy over time being held up by great foundry lithography. I mean, P4 .... REALLY? Itanic? Of course, Core 2 was just a logical extension from PIII (which is what P4 could have been). And tiles? Well, better late than never I suppose, but now they are stuck with latency they don't yet know how to deal with or get around well. Not to say AMD didn't have a similar setback, but it was on Zen 2!

Well, here's hoping Nova Lake is to Intel what Zen 3 was to AMD .

OneEng2 said:

Because they will go bankrupt or be purchased if they don't start making money.

Click to expand...

You don't need to rival TSMC to make money off of foundry. They also are a chip design company, and their chip design groups mutually benefit from a working Intel foundry. Also don't really see how they are going to go bankrupt any time soon unless once again, the board decides to spend the same amount of money as Pat except without rebuilding a foundry from the ground up. For Intel to go bankrupt they would have to be lying through their teeth about 10 billion reduction in expenses, and Intel would somehow have to fail even harder in 2026 despite finally having a working foundry and having refocused on designing products, and somehow the US government fails to deliver any support despite having a huge supporter of Made in America as president.

OneEng2 said:

They are. The cost of development being spread over fewer wafers causes EACH wafer to cost the company more money than that same wafer costs TSMC who is making more wafers on it. This puts Intel in a fundamentally poorer competitive position due to financials.

Click to expand...

What costs are being spread over fewer wafers? R&D cost is the only thing that is independent from wafer capacity and therefore spread between N number of wafers. For reference R&D from TSMC was 5 billion in 2024, even after climbing twofold since 2019 where it was only around 2.5 billion. It is but a fraction of the costs TSMC or Intel typically have to pay to operate the rest of their company. 3-5 billion is nothing to sneeze at, but at the size of a company like Intel, if that can promote long term growth in the rest of their businesses as well as opening up future business opportunities in the form of Intel Foundry, then it could be worth it. Whether or not it pans out depends on Intel's execution, but to say that it is impossible simply because Intel doesn't sell as many wafers as TSMC just doesn't make sense.

With your argument it would have been impossible for AMD and Nvidia to compete with Intel the past, nor would it be possible for AMD graphics/accelerators to make a profit in the present market because Nvidia is 10x the size and volume of products therefore 10x the money to dump into development, but we all know that isn't true. There is the additional point that money put into R&D does not correlate with level of innovation nor with competitiveness in the market. You never have been able to look at the R&D costs of any semiconductor company and correlate that with trends of competitiveness and success.

As a side note Intel's R&D for 2024 is 16.55 billion USD, increased about 3 Billion from the average 2017-2020 R&D budget. How much of this is attributable to elevated node spending is unknown, but regardless the net increase was 3 billion. Regardless the R&D budget has been massively bloated for a number of years and an entirely separate issue that should be addressed.

OneEng2 said:

Because they will go bankrupt or be purchased if they don't start making money.

I disagree. Intel has chosen to remain an IDM. They believe (incorrectly IMO) that it is less expensive for them to develop cutting edge processes on cutting edge machines and make their own chips on that process as a core advantage. When Intel fell behind on process tech, they fell behind in profit and in performance. They lost their golden goose in DC, and started losing market share on the desktop and laptop too.

They are. The cost of development being spread over fewer wafers causes EACH wafer to cost the company more money than that same wafer costs TSMC who is making more wafers on it. This puts Intel in a fundamentally poorer competitive position due to financials.

Exactly this.

Intel needs desperately to figure out EXACTLY THIS. They seem to be hell bent on being blind to the facts that are plain to see. They can no longer maintain market dominance through better foundry service than the rest of the world.

Today, AMD is competing (and besting in most instances) Intel while using a much less expensive node process. AMD could have easily made desktop and laptop Zen 5 on N3E (like it did Turin D), but the business cases wasn't there for them to do it. The process was too expensive.

Intel, on the other hand, used an even MORE expensive process (N3B) and still didn't unseat Zen 5.

For quite some time (decades) I have fundamentally believed that Intel was primarily a world class foundry that supported a few design teams. You don't have to be a particularly talented designer to beat everyone else when you have fundamentally better foundry service that catered to your design goals to fit like a glove.

Perhaps this was a weakness in the long run. Design got slopy over time being held up by great foundry lithography. I mean, P4 .... REALLY? Itanic? Of course, Core 2 was just a logical extension from PIII (which is what P4 could have been). And tiles? Well, better late than never I suppose, but now they are stuck with latency they don't yet know how to deal with or get around well. Not to say AMD didn't have a similar setback, but it was on Zen 2!

Well, here's hoping Nova Lake is to Intel what Zen 3 was to AMD .

Click to expand...

Intel's fab is not just for less expensive.

ajsdkflsdjfio said:

You don't need to rival TSMC to make money off of foundry. They also are a chip design company, and their chip design groups mutually benefit from a working Intel foundry. Also don't really see how they are going to go bankrupt any time soon unless once again, the board decides to spend the same amount of money as Pat except without rebuilding a foundry from the ground up. For Intel to go bankrupt they would have to be lying through their teeth about 10 billion reduction in expenses, and Intel would somehow have to fail even harder in 2026 despite finally having a working foundry and having refocused on designing products, and somehow the US government fails to deliver any support despite having a huge supporter of Made in America as president.

What costs are being spread over fewer wafers? R&D cost is the only thing that is independent from wafer capacity and therefore spread between N number of wafers. For reference R&D from TSMC was 5 billion in 2024, even after climbing twofold since 2019 where it was only around 2.5 billion. It is but a fraction of the costs TSMC or Intel typically have to pay to operate the rest of their company. 3-5 billion is nothing to sneeze at, but at the size of a company like Intel, if that can promote long term growth in the rest of their businesses as well as opening up future business opportunities in the form of Intel Foundry, then it could be worth it. Whether or not it pans out depends on Intel's execution, but to say that it is impossible simply because Intel doesn't sell as many wafers as TSMC just doesn't make sense.

With your argument it would have been impossible for AMD and Nvidia to compete with Intel the past, nor would it be possible for AMD graphics/accelerators to make a profit in the present market because Nvidia is 10x the size and volume of products therefore 10x the money to dump into development, but we all know that isn't true. There is the additional point that money put into R&D does not correlate with level of innovation nor with competitiveness in the market. You never have been able to look at the R&D costs of any semiconductor company and correlate that with trends of competitiveness and success.

As a side note Intel's R&D for 2024 is 16.55 billion USD, increased about 3 Billion from the average 2017-2020 R&D budget. How much of this is attributable to elevated node spending is unknown, but regardless the net increase was 3 billion. Regardless the R&D budget has been massively bloated for a number of years and an entirely separate issue that should be addressed.

Click to expand...

AMD DIDN'T compete with Intel when it decided "Real men have Fans". It bled red for years until FINALLY realizing that they didn't have the economy of scale to support the massive R&D required by this business strategy. Only when AMD spun off its fans and partnered wit TSMC did the turn things around.

Intel was double stupid IMO since they had seen the strategy fail catastrophicly at AMD for the same reason it is failing today at Intel. And today at Intel those same factors have ballooned ten fold.

Pat is a very very smart engineer, but his business intuition is Crap IMO. The evidence was plain to see and he bullied ahead anyway. Intel was right to replace him. He would have driven them under for sure. In fact, it may still come to pass that only a radically different "Intel" is left standing after all the buyouts and restructuring are over.

ajsdkflsdjfio said:

You don't need to rival TSMC to make money off of foundry. They also are a chip design company, and their chip design groups mutually benefit from a working Intel foundry. Also don't really see how they are going to go bankrupt any time soon unless once again, the board decides to spend the same amount of money as Pat except without rebuilding a foundry from the ground up. For Intel to go bankrupt they would have to be lying through their teeth about 10 billion reduction in expenses, and Intel would somehow have to fail even harder in 2026 despite finally having a working foundry and having refocused on designing products, and somehow the US government fails to deliver any support despite having a huge supporter of Made in America as president.

What costs are being spread over fewer wafers? R&D cost is the only thing that is independent from wafer capacity and therefore spread between N number of wafers. For reference R&D from TSMC was 5 billion in 2024, even after climbing twofold since 2019 where it was only around 2.5 billion. It is but a fraction of the costs TSMC or Intel typically have to pay to operate the rest of their company. 3-5 billion is nothing to sneeze at, but at the size of a company like Intel, if that can promote long term growth in the rest of their businesses as well as opening up future business opportunities in the form of Intel Foundry, then it could be worth it. Whether or not it pans out depends on Intel's execution, but to say that it is impossible simply because Intel doesn't sell as many wafers as TSMC just doesn't make sense.

With your argument it would have been impossible for AMD and Nvidia to compete with Intel the past, nor would it be possible for AMD graphics/accelerators to make a profit in the present market because Nvidia is 10x the size and volume of products therefore 10x the money to dump into development, but we all know that isn't true. There is the additional point that money put into R&D does not correlate with level of innovation nor with competitiveness in the market. You never have been able to look at the R&D costs of any semiconductor company and correlate that with trends of competitiveness and success.

As a side note Intel's R&D for 2024 is 16.55 billion USD, increased about 3 Billion from the average 2017-2020 R&D budget. How much of this is attributable to elevated node spending is unknown, but regardless the net increase was 3 billion. Regardless the R&D budget has been massively bloated for a number of years and an entirely separate issue that should be addressed.

Click to expand...

The biggest issue for any foundry, Intel or TSMC is running equipment at full capacity. I believe that Brian K made it to CEO because of his operational ability to get more wafers through the Intel fabs. Intel needs to match equipment purchase with wafer demand. Similar to AMD needing to project their wafer demand and entering wafer supply contracts with TSMC to cover what they expect will be their demand. In the last couple of years both companies have had mismatches, AMD is short on wafers and it appears that Intel is long on equipment. These competitive projections need to be made well in advance based on expected demand.

Intel can go bankrupt if they don’t predict demand well. Foundries are massive money sinks. AMD paid for unused wafers with Global Foundries when they still had wafer supply agreements with GF after the spin off. Intel and Micron were taking under utilization charges on their Optane manufacturing facility when demand did not meet expectations.

When Pat returned Intel needed new fabs shells for EUV equipment. Fab 42 would fit EUV but it had 10/7 nm in it. The Fab 34 shell in Ireland was started awhile back for the first EUV node previously called 7 nm and now 4/3 nm. It had some charges for a slow ramp in the fourth quarter of 2024.

Intel does not need to beat TSMC in PPA to stay in foundry but they need to be close on wafer cost performance. This means matching up supply and demand for wafers and optimizing throughput. Architecture sells wafers but costs determine margin and in a Capex heavy business they can’t just match AMDs margins because they still have foundry Capex to finance.

i don’t want to get into politics but I will point out that even though the U.S. has dominated semi’s to date the U.S. population is less than 5% of the global population. U.S. semi companies have a lot of foreign customers and the semi industry is one of most dependent on globalization for economy of scale. Nationalism is not going to improve semi finances. ‘Security’ could be expensive.

OneEng2 said:

AMD DIDN'T compete with Intel when it decided "Real men have Fans". It bled red for years until FINALLY realizing that they didn't have the economy of scale to support the massive R&D required by this business strategy. Only when AMD spun off its fans and partnered wit TSMC did the turn things around.

Intel was double stupid IMO since they had seen the strategy fail catastrophicly at AMD for the same reason it is failing today at Intel. And today at Intel those same factors have ballooned ten fold.

Pat is a very very smart engineer, but his business intuition is Crap IMO. The evidence was plain to see and he bullied ahead anyway. Intel was right to replace him. He would have driven them under for sure. In fact, it may still come to pass that only a radically different "Intel" is left standing after all the buyouts and restructuring are over.

Click to expand...

AMD might not have the economy of scale to beat Intel, but Intel arguably did have the scale to not only maintain it's leadership but expand and compete with TSMC business by opening external, they just never wished to until they realized opening up fabs was the only possible way they could still compete in the node race. This radical change unfortunately came at a time when Intel was no longer on top both technologically and financially. For years there were able to maintain node leadership and fund that endeavor solely through selling their own products, it was only in recent years where that strategy was no longer competent due to a multitude of reasons other than just "it's too expensive nowadays for Intel to continue". They had the funding and cash to maintain node leadership in the past from 2012-2021 but they didn't, not because it was too expensive but because they got complacent and their internal performance faltered. The same thing happened with the design teams leading to the fiasco Intel was in the past 5 years.

A lot of the pushback against Intel foundry and their investments isn't that there is no potential for significant customers in the future(still too early to determine), or that their node development hasn't been successful, but rather that perhaps they cannot afford such large investments required to develop a proper foundry, fab capacity and all. There is nothing wrong with the strategy of Intel opening a foundry, it just hasn't payed off yet which it never was intended to. It hasn't attracted as many customers as Pat might've hoped for at this point, but the real test of Intel Foundry success was always planned to be 18A. Nonetheless it is impossible to declare Intel foundry as a failure as of yet, especially considering 25% and possible increasing tariffs on semiconductors fabbed outside the US, increasing information indicating 18A is not only superior to N3 but in some ways superior/on par with N2, and signs that Intel themselves are beginning to return much of their products volume (which is and will be the bulk of their volume in the near future) back to IFS, instead of resorting to buying relatively low volumes of more expensive external nodes from TSMC.

I also don't understand why you draw the comparison between AMD's situation and Intel's current situation, there is a huge difference of relative size and circumstances for the situation. AMD was much smaller compared to Intel than Intel is now vs TSMC, and most importantly AMD hadn't gone under massive restructuring and investment in order to expand and open up their foundry, they were gearing up to let it go. Their trajectory was still on the downwards when they were nearing bankruptcy, they literally had no existing plans or room to do anything else but divest. Intel is in the valley right now but are absolutely nowhere as close to bankruptcy as AMD was in 2007-2008. Intel has reached a very low point, but they did it at the same time as heavy investments that are just now starting to pay off. Their trajectory is not going to be downwards for a long time and importantly they 100% have the time to let their foundry strategy materialize.

Theoretically there is nothing that prevents a IDM company from also running a fab business just as well as a fab only company. Obviously in the real world it has some complications namely with management and accountability issues when the product design teams work with an internal foundry, but nonetheless it is not an impossibility especially through proper structuring changes and a more driven/efficient Intel. Ultimately any question of IFS/Intel succeeding with their new strategy comes down purely to questions company culture, management, and accountability. If you were to magically be able to make IFS and all of it's management and workers just as good as TSMC, there would be zero issue with Intel opening up a foundry, especially if they had done it 5-10 years earlier than when Pat did it, when they had way more leeway to pull off such an ambitious plan. Pat's plan was risky for the period where he was CEO of Intel, but he was trying to turn the titanic around when they were already nearing the iceberg. He may have made some mistakes but it was impossible that the ship wouldn't take some damage at that point. It's just unfortunate that Intel didn't have the foresight to spot rocks in their path ahead of time, and much of the time they didn't even have the capability to maintain their status quo and instead let themselves slip although the full consequences would yet to be seen.

There is also the non-zero possibility that as interest picks up for 18A, it may turn out that Intel doesn't have enough capacity in 2027-2030 for both significant internal and external wafer orders. This scenario imo seems to be pretty unlikely unless things go really well in Intel's favor, but if such a scenario happens it just further proves that it wasn't the plan of ambitious fab spending that was the issue, just a lack of capital and the failure of Intel products occurring at the same time as such an ambitious plan.

Either way one thing can be sure, Intel's success in the next couple of years is the only sure measure of whether or not Pat's plan was good or not, and whether the IFS is as impossible as you claim it might be.

OneEng2 said:

TSMC is very risk averse. 3nm has been out for a while now, and will likely become the largest revenue generator for TSMC over the next 5 years. N2 class will be a play thing in comparison.

Click to expand...

I don't have a reference handy but I'm pretty sure I saw statements from TSMC saying that N2 would be an even bigger node than N3, in the same way N3 was bigger than N5 and N5 than N7 - not just in revenue (which is obvious given that each one costs more than the last) but also eventually in wafer starts.

Obviously what you say will be true at first. N7 was still a bigger revenue generator than N5 until early 2023 (when N3B was launching) and it'll probably be early 2026 (i.e. when N2 is launching) before N3 overtakes N5. So probably a safe bet to guess that the N2 family (N2/N2P/A16) won't overtake N3 until around the time A14 is launching.

High Na early report
Intel: No barriers inserting High-NA EUV @ 14A node.

- 30k wafers exposed so far on 2 fully installed tools.

- Source power exceeding targets (1st time ever for new EUV tool).

- Reliability is 85%, considered to be very good for a new scanner debut.

- Overlay EXE > NXE is 0.6 nm, on target, with no penalty for stitched die.

- No barriers from Intel's perspective to introduce a larger mask size of 6x12" to avoid die stitching; this would improve productivity by 23-50%. ASML clarified the scanner hardware assessment for this has not been stated.

- They implied the mask absorber was changed for their first set of masks, but "it wasn't anything novel."

- Overall very positive results; the only question that wasn't answered was the cost-per-pass one.

Yep, finally we are getting external rumors about IFS 18A's yield which is 20-30% for PTL's compute tile with 4+8+4. You would think with Intel separating CPU tile from GPU (LNL @ N3B), IFS should be able to mass producing PTL's CPU tile easily and yet the yield is 20-30%.

Of course, the quote is not from Intel but I rather believe Ming Chi Kuo than Intel's BS marketing talk....We will definitely hear more from other OEMs in the future...

Tigerick said:

Intel's 18A Process Reportedly Shows "Disappointing" Yield Rates As They Are Now At 20%-30%, Making Mass-Production Impossible

Intel's 18A yields are now to be around 20%- 30%, as the company is now claimed to face a massive roadblock to switching to mass production.

wccftech.com

Yep, finally we are getting external rumors about IFS 18A's yield which is 20-30% for PTL's compute tile with 4+8+4. You would think with Intel separating CPU tile from GPU (LNL @ N3B), IFS should be able to mass producing PTL's CPU tile easily and yet the yield is 20-30%.

Of course, the quote is not from Intel but I rather believe Ming Chi Kuo than Intel's BS marketing talk....We will definitely hear more from other OEMs in the future...

Click to expand...

How clued in is he with Intel matters? He mostly talks about Apple from what I’ve seen.

Tigerick said:

Intel's 18A Process Reportedly Shows "Disappointing" Yield Rates As They Are Now At 20%-30%, Making Mass-Production Impossible

Intel's 18A yields are now to be around 20%- 30%, as the company is now claimed to face a massive roadblock to switching to mass production.

wccftech.com

Yep, finally we are getting external rumors about IFS 18A's yield which is 20-30% for PTL's compute tile with 4+8+4. You would think with Intel separating CPU tile from GPU (LNL @ N3B), IFS should be able to mass producing PTL's CPU tile easily and yet the yield is 20-30%.

Of course, the quote is not from Intel but I rather believe Ming Chi Kuo than Intel's BS marketing talk....We will definitely hear more from other OEMs in the future...

Click to expand...

Which has been debunked on semiwiki Ming Chi Kuo is a Apple Analyst.This is just taiwanese propaganda there have been all rumours regarding the yields are bad for 18A and also how does he say this on the basis of OEM/ODM they don't know the yield.

External customers are yet to tape out

He is right about many things but I very much suspect yields given as % from industry sources. Intel wouldn't tell OEM about yields; just quantities. I suspect it's the same information from earlier - about an external customer with some test chips, not PTL.

Not that it bodes well for 18A in any case but it seems like one of MCK's weaker posts.

511 said:

Which has been debunked on semiwiki Ming Chi Kuo is a Apple Analyst.This is just taiwanese propaganda there have been all rumours regarding the yields are bad for 18A and also how does he say this on the basis of OEM/ODM they don't know the yield.

External customers are yet to tape out

Intel 18A yields at 20-30%?

The first Panther Lake engineering samples, made with Intel/IFS’s 18A, are currently being tested by major PC ODM/EMS makers. My early 2025 industry survey showed 18A yields below 20-30%, so there’s still a lot of room to step up—which doesn’t bode well for Intel’s goal of hitting mass...

semiwiki.com

Click to expand...

Actually, as in many things, I would bet the truth is somewhere in between Intel marketing and his estimates or numbers. Time will tell.

Markfw said:

Actually, as in many things, I would bet the truth is somewhere in between Intel marketing and his estimates or numbers. Time will tell.

Click to expand...

My first question to every yield data is die size without die size you are giving me a meaningless number.
For a die near reticle size this translates to about 0.2 Defects/cm2 for a 22% yield and samr D0 for Panther Lake die of 114 mm2 translates to about 75% die size.

511 said:

My first question to every yield data is die size without die size you are giving me a meaningless number.
For a die near reticle size this translates to about 0.2 Defects/cm2 for a 22% yield and samr D0 for Panther Lake die of 114 mm2 translates to about 75% die size.

Click to expand...

I believe the first rumor was ~ 10% and it was based on near reticle size die. If that remains true for this rumor, then Intel is likely at a pretty decent yield (70-80%) for the PTL die. IIRC the 10% rumor translated into ~ 60% on a reasonably sized die.

Intel actually provided an earlier number on defects/cm2 which I trust WAY more than a yield claim without a die size. On the other hand, Intel also did their yield on SRAM vs logic or mixed logic, so there is really only so much you can get from these numbers without actual production empirical data.

Hahahah.
Hahahahhahahahahhaah hahahahahhahahaha.

Masterful gambit, saar.

Hitman928 said:

View attachment 117815

Click to expand...

foundry.contact@intel.com

^^^ Is it just me or is this unusual?

adroc_thurston said:

CHIPS Act dies because employees are fired – NIST CHIPS people are probationary - Semiwiki

- NIST to lose 100's of mainly CHIPS Act people - If no people are left to administer CHIPS Act it dies by default - Following USAID play book to kill an unwanted program - Using the

semiwiki.com

Hahahah.
Hahahahhahahahahhaah hahahahahhahahaha.

Click to expand...

That is a massive bombshell. How's Intel stock doing today? This is the worst single Intel news event I can remember for years.

Maybe Intel doesn't need the CHIPS money anymore coz its fate has already been decided by vultures or whatever?