Opinion: To make IDM 3.0 a success, Intel must make other companies IDM

[delong.height]

In the grand narrative of the semiconductor industry, Intel and TSMC have long represented two distinct business models: the vertically integrated Integrated Device Manufacturer (IDM) and the specialized Foundry. The IDM model, with its deep synergy between design and manufacturing, has stood on its own, while the Foundry model, with its neutrality and economies of scale, has come to dominate the market. However, based on a series of recent strategic moves by Intel and SoftBank, a new strategic concept that merges the best of both models is emerging. I speculated it as IDM 3.0, and it could be key to Intel’s return to prominence.

IDM 3.0 is not merely an extension of Intel CEO Pat Gelsinger's "IDM 2.0." It's a more radical and disruptive innovation. The model is designed to transform Intel's core strength as an IDM—Design-Technology Co-Optimization (DTCO)—from an internal-only asset into a high-value service for external clients. This strategic shift aims to solve the two biggest problems of the traditional IDM model: massive capital expenditure and a lack of customer trust.

---

Analyzing the Strengths and Weaknesses of Traditional Models​

To understand IDM 3.0, one must first review the characteristics of the traditional models.

The traditional IDM (Intel) thrives on its end-to-end control. This vertical integration allows Intel to tailor its products for maximum performance, and its self-sufficiency provides a powerful strategic advantage, especially during supply chain disruptions. However, the IDM model has clear drawbacks: it requires enormous R&D and fab construction costs, and its closed nature can deter clients due to a competitive conflict of interest.

The Foundry (TSMC), by contrast, rules the market with its neutrality and economies of scale. As the "Switzerland of the semiconductor industry," TSMC doesn't compete with its clients, earning the trust of nearly all fabless companies worldwide. By serving hundreds of companies, TSMC effectively spreads its massive capital expenditures and achieves a strong scale advantage.

---

IDM 3.0: The Ultimate Hybrid of Both Models​

The IDM 3.0 model I've outlined seeks to combine the technological depth of the IDM with the commercial breadth of the Foundry. This is more than just a simple foundry service; it's a comprehensive strategy based on a restructuring of capital, technology, and business models.

• Capital and Governance Integration: A core component of this model is allowing strategic investors like Arm, fabless companies, and SoftBank to invest in Intel Foundry Services (IFS) with shared governance rights. This joint governance structure transforms the relationship from a simple transaction into a strategic partnership of shared interests. With a seat on the board, clients can mitigate intellectual property concerns.
• 
  
• Open DTCO: While TSMC currently lead in certain advanced nodes, Intel's accumulated DTCO capability as an IDM remains unique. The IDM 3.0 strategy would offer this capability as a high-value service to the Arm ecosystem. By collaborating closely with clients, Intel could optimize the Arm architecture on its advanced process nodes, leading to performance, power efficiency, and yield improvements beyond what traditional foundries can offer. This model sells not just a manufacturing process, but a performance optimization solution.
• 
  
• Business Model Redefinition: As Intel has already implemented, its product divisions (like CPU and GPU) are now treated as external clients, paying for foundry services and committing to wafer orders. This internal marketization provides a credible reference for external clients. More importantly, IDM 3.0 adopts the profit allocation logic of the IDM model: IFS would use rebates to return a portion of the "manufacturing profit" to key clients. This mechanism would boost client profitability, creating a mutually beneficial ecosystem and, through economies of scale, lowering Intel's own costs.

---

Strategic Trade-offs and Practical Considerations of IDM 3.0​

This model comes with practical complexities and requires careful strategic trade-offs:

1. Balancing "Partners" and "Neutrality": With limited resources, IDM 3.0 prioritizes a pragmatic strategy: focus on the most critical clients. This mirrors the influence Apple and Nvidia have on TSMC. The model acknowledges Intel’s financial and growth constraints by deeply binding with key players in the Arm ecosystem, securing crucial initial orders and capital while accepting a loss of "neutrality" for the broader market.
2. 
   
3. Sustainability of the Rebate Mechanism: Rebates are not simple subsidies but performance-linked incentives. They would only be shared after a process is mature and its profitability is stable. Over the long term, the sustainability of this model must come from the foundry business's own profitability. By increasing production volume and yield, Intel can lower its unit cost, creating room for rebates.
4. 
   
5. Complexity of Investment Returns: The primary goal of IDM 3.0 is to lock clients into Intel’s end-to-end services, from wafers to packaging, testing, and assembly. This deep integration makes switching costs very high, securing stable returns. The ROI should be measured not only by the foundry's direct profits but also by the indirect enhancement of Intel's own and its partners' product competitiveness.

 

[Paul2]

The enormous number of South Asian outsourcing shops already fill that niche very tightly. They have very good physical expertise with all major foundries, and enough hands to manually tweak designs for given node, and usage regime cheaply.

I've seen Taiwanese SSD chip designers turning to them thinking that they were pretty good already, then BOOM 40% power reduction under light load, and idle.

 

[user nl]

Strategic Trade-offs and Practical Considerations of IDM 3.0​

This model comes with practical complexities and requires careful strategic trade-offs:
The primary goal of IDM 3.0 is to lock clients into Intel’s end-to-end services, from wafers to packaging, testing, and assembly. This deep integration makes switching costs very high, securing stable returns. The ROI should be measured not only by the foundry's direct profits but also by the indirect enhancement of Intel's own and its partners' product competitiveness.

I think you just ended with the best argument why customers would not go to INTEL3.0. Why move to INTEL3.0, when they are much more flexible and cost-effective working with TSMC, without the crazy governance responsibility of running/operating the Foundry, as you propose for INTEL 3.0.

Scale and Trust are the key determining factors why TSMC has 10 times more wafers processed than INTEL2.0

Your proposal seems dead on arrival to me, in a world that flourishes in OPEN INNOVATION

 

[benb]

I think there is a reason for IDM but we haven't seen it yet. IDM grew out of raw necessity, when the industry was young, then foundries created sophisticated alternative options.

Today it feels like there are not enough options. I don't know if IDM is ripe to metamorphosis into a new and better option. But some other thing, perhaps a national consortium of design houses, or designers owning the fab process to some degree, will be the next evolution. Maybe what is really needed is a big cost reduction since that is the friction preventing change.

 

[count]

I think there is a reason for IDM but we haven't seen it yet. IDM grew out of raw necessity, when the industry was young, then foundries created sophisticated alternative options.

Today it feels like there are not enough options. I don't know if IDM is ripe to metamorphosis into a new and better option. But some other thing, perhaps a national consortium of design houses, or designers owning the fab process to some degree, will be the next evolution. Maybe what is really needed is a big cost reduction since that is the friction preventing change.

Everyone who says TSMC needs competition is ignoring all the competition coming from China... today it is only on training edge nodes >7nm... but they will catch up.

My understanding is that margins on legacy nodes are facing a lot of pressure from Chinese foundries.

Trust me there is going to be lots of competition, so much competition coming from China that people in the semiconductor world are going to call it unfair competition in 5-10 years.

 

[Paul2]

Everyone who says TSMC needs competition is ignoring all the competition coming from China... today it is only on training edge nodes >7nm... but they will catch up.

Most of ICs in use are >7. Conceding mainstream, and legacy without a fight is a massive blunder. Morris kinda knew it, but new get leaders at TSMC got too focused on top clients.

I heard once that Mainland companies are big buyers of Digitimes research. I can guess they can already be bigger than Taiwanese. Taiwanese all pile up onto a few fat clients, and regions, while mainlanders have to hustle, be look for them actively, which will be good for them in the future. Lazy salesmen problem.

 

[count]

Most of ICs in use are >7. Conceding mainstream, and legacy without a fight is a massive blunder. Morris kinda knew it, but new get leaders at TSMC got too focused on top clients.

Yes it's classic innovators dilemma. Intel made this mistake when they conceded the mobile market, and if TSMC concedes the legacy market it will be an equally catastrophic blunder as it opens the door to competition from the low end.

And anyone who claims China will never be able to catch up is not paying attention to the market.

People said China could never get 7nm, but they are here. People say China will never get EUV, but they are announcing new breakthroughs every quarter. They have money, talent, and most importantly lots of incentive to get there.

 

[tomatoma]

Well, everyone is underestimating China...
No, not just China, there is a tendency to underestimate newcomers with momentum.
The same can be said for Rapidus in Japan,
When they were 10 years late and when they created the 2nm process, those who underestimated them seemed to be quite surprised.
For the reasons mentioned above, when the TSMC N2 confidential information leak incident occurred, groundless and malicious rumors and damage to Rapidus' reputation spread in Taiwan. This may be because they looked down on Rapidus. They may have assumed that "they can't even make a 2nm process," which is why such rumors spread...
Those involved in the Japanese semiconductor industry felt tremendous anger at the reputational damage caused by this incident…

It is best not to underestimate someone with real talent, no matter how small the scale, as you will inevitably be caught off guard later on.

 

[count]

Well, everyone is underestimating China...
No, not just China, there is a tendency to underestimate newcomers with momentum.
The same can be said for Rapidus in Japan,
When they were 10 years late and when they created the 2nm process, those who underestimated them seemed to be quite surprised.
For the reasons mentioned above, when the TSMC N2 confidential information leak incident occurred, groundless and malicious rumors and damage to Rapidus' reputation spread in Taiwan. This may be because they looked down on Rapidus. They may have assumed that "they can't even make a 2nm process," which is why such rumors spread...
Those involved in the Japanese semiconductor industry felt tremendous anger at the reputational damage caused by this incident…

It is best not to underestimate someone with real talent, no matter how small the scale, as you will inevitably be caught off guard later on.

Yeah I have worked in China and Japan and Japan also has a lot talent and are also very hard working.

Taiwan learned how to do manufacturing from Japan, and then China learned how to do manufacturing from Taiwan. There are a lot of similarities in how they do things.

However my experience with Japan is the very rigid organizational structures and need to get approval from everyone for doing any little task really slows things down and stifles innovation. That said, when they do execute they do a very good job with high quality. I am not saying every Japanese company is like this but the ones I have worked with are. In China they prioritize speed a lot more.

 

[MKWVentures]

IDM is not a good idea unless you cannot get support for the manufacturing required for your field. You need scale and the flexibility of multiple customers. Both Pat and Swan 100% agreed that internal IDM cannot work (Its the math). Swan wanted outsource, Pat wanted IDM2.0 (or as I call it IBM2.0).

The Plan now is to get major customers just to allow 14A to ramp. Intel is using a 5 year old technology (Intel 7/10) to manufacture the majority of its wafers today and it has 70% CPU market share (including all wafers used for all products).

if no customers, 14A doesnt ramp (licensing agreement and codevelopment agreement with someone will follow... like IBM)

 

[tomatoma]

IDM is not a good idea unless you cannot get support for the manufacturing required for your field. You need scale and the flexibility of multiple customers. Both Pat and Swan 100% agreed that internal IDM cannot work (Its the math). Swan wanted outsource, Pat wanted IDM2.0 (or as I call it IBM2.0).

The Plan now is to get major customers just to allow 14A to ramp. Intel is using a 5 year old technology (Intel 7/10) to manufacture the majority of its wafers today and it has 70% CPU market share (including all wafers used for all products).

if no customers, 14A doesnt ramp (licensing agreement and codevelopment agreement with someone will follow... like IBM)

If you say IBM 2.0, there is no future LOL

Absolutely

 

[tomatoma]

IDM is not a good idea unless you cannot get support for the manufacturing required for your field. You need scale and the flexibility of multiple customers. Both Pat and Swan 100% agreed that internal IDM cannot work (Its the math). Swan wanted outsource, Pat wanted IDM2.0 (or as I call it IBM2.0).

The Plan now is to get major customers just to allow 14A to ramp. Intel is using a 5 year old technology (Intel 7/10) to manufacture the majority of its wafers today and it has 70% CPU market share (including all wafers used for all products).

if no customers, 14A doesnt ramp (licensing agreement and codevelopment agreement with someone will follow... like IBM)

We aimed to create an IDM in a new era with the flexibility of being able to create chips like Foundry, like Foundry.

While maintaining the conventional system of IDM... Flexible by external customers manufacturing and packaging chips in their own factories

 

[delong.height]

I think you just ended with the best argument why customers would not go to INTEL3.0. Why move to INTEL3.0, when they are much more flexible and cost-effective working with TSMC, without the crazy governance responsibility of running/operating the Foundry, as you propose for INTEL 3.0.

Scale and Trust are the key determining factors why TSMC has 10 times more wafers processed than INTEL2.0

Your proposal seems dead on arrival to me, in a world that flourishes in OPEN INNOVATION

I do believe what you said is partially correct. But there are points that you have likely missed out on.

I think that IDM has its benefits and drawbacks. The biggest drawback is it is uneconomic to invest in leading edge and cannot scale out as much as foundry does.

On the other hand, being IDM has some benefits, like vertically integration and tight supply chain control and ability to win on supply.

Provide IDM-like experiences can potentially attract more customers, which mean Intel provide rebate to fabless, expose all end-to-end capabilities to fabless, hyperscalers, etc.

I don't disagree on OPEN INNOVATION. But in addition to OPEN INNOVATION, there needs to be proprietary action, like TSMC is doing to Apple and Nvidia through custom node. Provide IDM-like experience to most important company, can help that company win in a supply game assuming its product are competitive but competitor are catching up quick. Like AMD vs Nvidia.

Software moat can be destroyed or weaken easily over time assuming enough assets being poured into it. But supply chain cannot be easily replicated because they are expensive, especially nowadays, you need 30 billion to build the fab. For competitors like AMD, they won't have that much of assets to be put into wafer commitment.

 

[tomatoma]

Yeah I have worked in China and Japan and Japan also has a lot talent and are also very hard working.

Taiwan learned how to do manufacturing from Japan, and then China learned how to do manufacturing from Taiwan. There are a lot of similarities in how they do things.

However my experience with Japan is the very rigid organizational structures and need to get approval from everyone for doing any little task really slows things down and stifles innovation. That said, when they do execute they do a very good job with high quality. I am not saying every Japanese company is like this but the ones I have worked with are. In China they prioritize speed a lot more.

Yes, I agree
But I didn't mean to say that
I wanted to emphasize that no matter how small the scale, we should not underestimate newcomers with excellent skills. I would be happy if you would take the example of Japan that I mentioned as just an example. In the industry There is a common tendency to look down on newcomers.

 

[tomatoma]

I do believe what you said is partially correct. But there are points that you have likely missed out on.

I think that IDM has its benefits and drawbacks. The biggest drawback is it is uneconomic to invest in leading edge and cannot scale out as much as foundry does.

On the other hand, being IDM has some benefits, like vertically integration and tight supply chain control and ability to win on supply.

Provide IDM-like experiences can potentially attract more customers, which mean Intel provide rebate to fabless, expose all end-to-end capabilities to fabless, hyperscalers, etc.

I don't disagree on OPEN INNOVATION. But in addition to OPEN INNOVATION, there needs to be proprietary action, like TSMC is doing to Apple and Nvidia through custom node. Provide IDM-like experience to most important company, can help that company win in a supply game assuming its product are competitive but competitor are catching up quick. Like AMD vs Nvidia.

Software moat can be destroyed or weaken easily over time assuming enough assets being poured into it. But supply chain cannot be easily replicated because they are expensive, especially nowadays, you need 30 billion to build the fab. For competitors like AMD, they won't have that much of assets to be put into wafer commitment.

I accept
And It's not that IDM can't do open innovation.
Personally, I think it would be better if it just evolved into a new form of IDM.
You don't have to throw away the IdM, It's to just evolve

 

[Jozo035]

Everyone is making same chips: DSP array, SRAM, DDR/HBM phy and some fast IO. Sometimes it looks like just different SKU's of same architectural family. This is ideal situation for foundries, which has most of reusable IP already qualified.

To make IDM work, we would need to "overwhelm" TSMC with unreasonable requirements. Like order of magnitude more of DSP at ultra-low power. 10-20GHz pipeline frequencies. In memory computing. Maybe quantum computing? And so on.

Basically, the only way to make IDM work is to make process development uneconomical, so IDM would benefit from vertical integration.

 

[delong.height]

Everyone is making same chips: DSP array, SRAM, DDR/HBM phy and some fast IO. Sometimes it looks like just different SKU's of same architectural family. This is ideal situation for foundries, which has most of reusable IP already qualified.

To make IDM work, we would need to "overwhelm" TSMC with unreasonable requirements. Like order of magnitude more of DSP at ultra-low power. 10-20GHz pipeline frequencies. In memory computing. Maybe quantum computing? And so on.

Basically, the only way to make IDM work is to make process development uneconomical, so IDM would benefit from vertical integration.

Exactly, go on the route that make TSMC difficult to compete in by having a deeper specialization into things.

And that is true when most chips are becoming a commodity, the ability for fabless companies to be more vertically integrated is the most important way.

 

[hist78]

I accept
And It's not that IDM can't do open innovation.
Personally, I think it would be better if it just evolved into a new form of IDM.
You don't have to throw away the IdM, It's to just evolve

Why would the entire semiconductor industry, including supply chains, customers, and even Intel’s competitors, wait for Intel to figure out IDM 2.0, IDM 3.0, IDM X, or whatever “IDM” version comes next? The train left a long time ago, and most advanced logic players are already on board, except Intel and Samsung.

This “IDM xyz” approach worked in the past, when IDM players were the dominant force. Advanced logic IDMs like Intel and IBM used to be at the center of the universe. But those days are long gone.

 

[Daniel Nenni]

Everyone who says TSMC needs competition is ignoring all the competition coming from China... today it is only on training edge nodes >7nm... but they will catch up.
My understanding is that margins on legacy nodes are facing a lot of pressure from Chinese foundries.
Trust me there is going to be lots of competition, so much competition coming from China that people in the semiconductor world are going to call it unfair competition in 5-10 years.

China will never catch TSMC. Big diverse customers and the ecosystem drives TSMC and China does not have that. If you say 20-30 years? Sure, anything could happen. We already know what will happen 5-10 years out and it is all TSMC.

If China takes Taiwan that is another story but that is even more incentive to get Intel back on the leading edge.

Mature nodes are anyone's for the taking. TSMC has the best yield and economies of scale so if they want the mature node business it would be theirs.

Foundry Gross Margins for 2024:

TSMC 56.1% (Mature node gross margins are actually higher / Leading edge nodes are lower)
SMIC 18%
GF 24.5%
UMC 32.6%

But yes, the China Government has told China companies that they must buy China wafers if possible and still stay competitive. But the rest of the world will buy on wafer price.

 

[tomatoma]

Why would the entire semiconductor industry, including supply chains, customers, and even Intel’s competitors, wait for Intel to figure out IDM 2.0, IDM 3.0, IDM X, or whatever “IDM” version comes next? The train left a long time ago, and most advanced logic players are already on board, except Intel and Samsung.

This “IDM xyz” approach worked in the past, when IDM players were the dominant force. Advanced logic IDMs like Intel and IBM used to be at the center of the universe. But those days are long gone.

Intel is a stupid outdated company
However, IDM itself is not a problem.
Intel that couldn't change while maintaining IdM is bad