Intel Employees "Very Optimistic"

[hist78]

True, I do not see AMD or Nvidia using Intel Foundry but there are plenty of whales out there where competition should not be a problem. The cloud companies for example. They do not sell their chips. Broadcom should be first in line. Lip-Bu and Hock Tan (both Malaysian) know each other quite well and that would be a strategic partnership for Intel Foundry, absolutely.

If Intel Foundry can close 6 top customers for 18AP next year that should do it. That is what my marching orders would be.

Can you elaborate a bit more on how a cloud service provider, such as Google, Amazon, or Microsoft, can become a 'whale' customer? I did some calculations and found that their volumes are limited, especially in relation to Intel Foundry's high volume production goal.

Furthermore, every chip that Intel Foundry produces for Google, Amazon, or Microsoft represents a lost opportunity for Intel’s Product division.

 

[blueone]

You pointed out one of the reasons why Samsung foundry is struggling. TSMC doesn't need to address such conflict of interest issues because it doesn't exist.

I've seen no evidence that your statement is correct. The primary reasons Samsung's foundry business is more likely to be struggling are inability to deliver leading edge process chips with reasonable yields, and inability to convince potential customers Samsung can achieve roadmap parity with TSMC. You can't be a second source foundry without comparable technology and pricing.

Do you have any evidence that potential Intel Foundry customers have competition from Intel products as their fundamental concern? Executive quotes will do.

Furthermore, every chip that Intel Foundry produces for Google, Amazon, or Microsoft represents a lost opportunity for Intel’s Product division.

The chips these three customers designed for internal use aren't good opportunities for Intel's product divisions. Merchant chip product companies, including AMD and Ampere, are too high cost for the specialized requirements of these cloud computing vendors to be competitive at all.

Positioning your personal opinions as assertions of fact without evidence isn't persuasive.

 

[XYang2023]

I've seen no evidence that your statement is correct. The primary reasons Samsung's foundry business is more likely to be struggling are inability to deliver leading edge process chips with reasonable yields, and inability to convince potential customers Samsung can achieve roadmap parity with TSMC. You can't be a second source foundry without comparable technology and pricing.

Do you have any evidence that potential Intel Foundry customers have competition from Intel products as their fundamental concern? Executive quotes will do.

The chips these three customers designed for internal use aren't good opportunities for Intel's product divisions. Merchant chip product companies, including AMD and Ampere, are too high cost for the specialized requirements of these cloud computing vendors to be competitive at all.

Positioning your personal opinions as assertions of fact without evidence isn't persuasive.

MJ mentioned that custom x86 and chiplets have generated a lot of interest. I suppose Intel could bundle that with its foundry offerings.

https://twitter.com/x/status/1930041166290137591

 

[hist78]

I've seen no evidence that your statement is correct. The primary reasons Samsung's foundry business is more likely to be struggling are inability to deliver leading edge process chips with reasonable yields, and inability to convince potential customers Samsung can achieve roadmap parity with TSMC. You can't be a second source foundry without comparable technology and pricing.

Do you have any evidence that potential Intel Foundry customers have competition from Intel products as their fundamental concern? Executive quotes will do.

The chips these three customers designed for internal use aren't good opportunities for Intel's product divisions. Merchant chip product companies, including AMD and Ampere, are too high cost for the specialized requirements of these cloud computing vendors to be competitive at all.

Positioning your personal opinions as assertions of fact without evidence isn't persuasive.

If you need further clarification on my comments, feel free to ask. There’s no need to use unusual phrasing, like in your last paragraph. Remember every blogger here is not writing a PhD paper. If you are not clear or even disagree with their opinions, please just share your thoughts, without attacking other blogger's competency. IMHO, this will promote more idea discussion and attract more people to join SemiWIki.

Samsung has had many conflicts of interest with its customers. The most well-known, and arguably the most damaging one, was the Apple-Samsung conflict. The end result was that Apple moved its processor manufacturing from Samsung to TSMC. And both Apple and TSMC went on to achieve huge success. Samsung not only lost an important customer but also an important resource provider to improve and enhance its manufacturing.

###########

The chips these three customers designed for internal use aren't good opportunities for Intel's product divisions. Merchant chip product companies, including AMD and Ampere, are too high cost for the specialized requirements of these cloud computing vendors to be competitive at all.

Positioning your personal opinions as assertions of fact without evidence isn't persuasive.

Can you please read my posting again:

"Intel’s server and AI division generated only $4.1 billion, just 32.28% of its $12.7 billion total revenue in Q1 2025. In this segment, Intel is also up against formidable competitors, including AMD, Nvidia, Google, Amazon, Microsoft, Meta (Facebook), Broadcom, Qualcomm, and MediaTek."

Let me explain it further:

My comment is about Intel lacking competitive products that the market needs. The Intel Data Center and AI (DCAI) group is facing an onslaught from every direction. AMD, Nvidia, Google, Amazon, Microsoft, Meta (Facebook), Broadcom, Qualcomm, and MediaTek all have, or will have, server and/or AI products that overlap with what Intel is doing or may pursue in the future. Except for MediaTek, every company I mentioned has a larger market capitalization than Intel.

Among these companies, some products may fall under what you might categorize as "niche," which Intel's product division may choose not to pursue. However, many of their offerings are far from niche. AMD’s EPYC and MI300/MI400, Amazon’s Graviton, Google’s Axion, Microsoft’s Cobalt, and especially Nvidia’s Hopper/Blackwell platforms cannot be simply dismissed as niche products.

Additionally, if Google were to ask Intel Foundry to manufacture an Google designed AI processor, should Intel be concerned that this represents a lost opportunity for its own Gaudi AI processor? This is just one example of the kind of conflict of interest that Intel's IDM model is inherently vulnerable to.

 

[XYang2023]

Additionally, if Google were to ask Intel Foundry to manufacture an Google designed AI processor, should Intel be concerned that this represents a lost opportunity for its own Gaudi AI processor? This is just one example of the kind of conflict of interest that Intel's IDM model is inherently vulnerable to.

Different companies within Intel group, it is their duties to maximise profits. The correct way of seeing it is that Intel needs to have competing products in AI space instead of worrying Gaudi sales.

https://twitter.com/x/status/1930316466202783901

https://twitter.com/x/status/1930323528139845920

 

[hist78]

Different companies within Intel group, it is their duties to maximise profits. The correct way of seeing it is that Intel needs to have competing products in AI space instead of worrying Gaudi sales.

https://twitter.com/x/status/1930316466202783901

https://twitter.com/x/status/1930323528139845920

Intel has only one publicly traded stock and only one meaningful accounting book. Can you explain a little bit how maximizing each Intel division's profit (such as product division's and Intel Foundry's) can lead to Intel's success?

 

[XYang2023]

Intel has only one publicly traded stock and only one meaningful accounting book. Can you explain a little bit how maximizing each Intel division's profit (such as product division's and Intel Foundry's) can lead to Intel's success?

Intel's valuation is dragged down by the foundry. IFS needs to break even then moves towards profitability. Intel's products needs to be competitive.

 

[Daniel Nenni]

Can you elaborate a bit more on how a cloud service provider, such as Google, Amazon, or Microsoft, can become a 'whale' customer? I did some calculations and found that their volumes are limited, especially in relation to Intel Foundry's high volume production goal.

Furthermore, every chip that Intel Foundry produces for Google, Amazon, or Microsoft represents a lost opportunity for Intel’s Product division.

Microsoft, Google, and Amazon make more than cloud chips and they write some big checks inside the ecosystem. They are big name companies and pull a lot of weight as a named customer. If you are an EDA or IP company and have Google, Microsoft, and Amazon as customers TSMC will listen closely to you, absolutely.

Taking NOT TSMC business from Samsung does not take away business from Intel.

 

[blueone]

MJ mentioned that custom x86 and chiplets have generated a lot of interest. I suppose Intel could bundle that with its foundry offerings.

https://twitter.com/x/status/1930041166290137591

I have no doubt that's the case, for those server CPUs that need to be x86 compatible. Compared to the custom Arm-based CPUs from any of the three cloud companies doing custom CPUs for their own applications, I'm still a deep skeptic, based on power efficiency and cost.

 

[XYang2023]

I have no doubt that's the case, for those server CPUs that need to be x86 compatible. Compared to the custom Arm-based CPUs from any of the three cloud companies doing custom CPUs for their own applications, I'm still a deep skeptic, based on power efficiency and cost.

I think she meant those custom solutions are also targeted for csp's internal workloads. Intel has efficient cores to address low power.

 

[Daniel Nenni]

I have no doubt that's the case, for those server CPUs that need to be x86 compatible. Compared to the custom Arm-based CPUs from any of the three cloud companies doing custom CPUs for their own applications, I'm still a deep skeptic, based on power efficiency and cost.

Commercial chiplets are still a work in progress. At the Design Automation Conference this month there is Chiplet Pavillion section with a dozen different companies which I expect will be very high traffic. Quite a few chiplet technical sessions as well which is a nice break from all of the AI nonsense. The ecosystem is just not there yet for chiplets unless you are making your own like AMD, Intel, and Nvidia.

 

[blueone]

If you need further clarification on my comments, feel free to ask.

I don't need clarification. I understand your meaning completely.

There’s no need to use unusual phrasing, like in your last paragraph. Remember every blogger here is not writing a PhD paper. If you are not clear or even disagree with their opinions, please just share your thoughts, without attacking other blogger's competency. IMHO, this will promote more idea discussion and attract more people to join SemiWIki.

Promoting discussion based on assertions which are not based on knowledge or facts is a waste of time.

Samsung has had many conflicts of interest with its customers. The most well-known, and arguably the most damaging one, was the Apple-Samsung conflict. The end result was that Apple moved its processor manufacturing from Samsung to TSMC. And both Apple and TSMC went on to achieve huge success. Samsung not only lost an important customer but also an important resource provider to improve and enhance its manufacturing.

While one can conjecture that Apple was worried about using a foundry that was a division of a direct retail-level competitor to their products, TSMC's superior technology and fabrication capabilities would still be the most significant factor, perceived conflicts of interest or not.

Can you please read my posting again:

"Intel’s server and AI division generated only $4.1 billion, just 32.28% of its $12.7 billion total revenue in Q1 2025. In this segment, Intel is also up against formidable competitors, including AMD, Nvidia, Google, Amazon, Microsoft, Meta (Facebook), Broadcom, Qualcomm, and MediaTek."

I understood what you said the first time I read it. The problem is that you don't understand the server CPU market or the AI processor market, so you are pronouncing market equivalency where there isn't any.

Let me explain it further:

My comment is about Intel lacking competitive products that the market needs. The Intel Data Center and AI (DCAI) group is facing an onslaught from every direction. AMD, Nvidia, Google, Amazon, Microsoft, Meta (Facebook), Broadcom, Qualcomm, and MediaTek all have, or will have, server and/or AI products that overlap with what Intel is doing or may pursue in the future. Except for MediaTek, every company I mentioned has a larger market capitalization than Intel.

Nvidia and AMD are competitors to Intel's DCAI group, because they are merchant chip vendors. That means they are trying to sell their products to multiple customers, so their feature sets have to be broad enough to appeal to the near sum of the market requirements. Google, Amazon, Microsoft, and Facebook do not sell their chips to system builders, like Intel, Nvidia, and AMD do, so thinking that the cloud vendors are competitors to Intel is incorrect. The resulting chips from merchant versus in-house designers will be different.

Among these companies, some products may fall under what you might categorize as "niche," which Intel's product division may choose not to pursue. However, many of their offerings are far from niche. AMD’s EPYC and MI300/MI400, Amazon’s Graviton, Google’s Axion, Microsoft’s Cobalt, and especially Nvidia’s Hopper/Blackwell platforms cannot be simply dismissed as niche products.

You're mischaracterizing the markets. There are two categories here, merchant chips and proprietary designs for in-house use. I don't consider Amazon's Gravitons or Google TPUs to be niche products. The are custom designs for specific applications and software stacks.

Additionally, if Google were to ask Intel Foundry to manufacture an Google designed AI processor, should Intel be concerned that this represents a lost opportunity for its own Gaudi AI processor? This is just one example of the kind of conflict of interest that Intel's IDM model is inherently vulnerable to.

No. TPUs and Gaudi processors have completely different design targets. For example, Google designed its TPUs to work with its extraordinary fully optical and proprietary interconnect called Jupiter. Gaudi has integrated Ethernet links which support RoCEv2 protocol. There is zero chance Gaudi could replace a TPU on Google's fabric.

 

[blueone]

Commercial chiplets are still a work in progress. At the Design Automation Conference this month there is Chiplet Pavillion section with a dozen different companies which I expect will be very high traffic. Quite a few chiplet technical sessions as well which is a nice break from all of the AI nonsense. The ecosystem is just not there yet for chiplets unless you are making your own like AMD, Intel, and Nvidia.

I agree. I do think, however, that for some markets, network processing, security, or very specialized accelerators, interoperable merchant chiplets have a very compelling value proposition.

 

[blueone]

I think she meant those custom solutions are also targeted for csp's internal workloads. Intel has efficient cores to address low power.

Which would you rather have for the best combination of power efficiency and capability? An Arm v9 core or an x86 Intel efficiency core?

 

[Daniel Nenni]

I agree. I do think, however, that for some markets, network processing, security, or very specially accelerators, interoperable merchant chiplets have a very compelling value proposition.

What about x86 chiplets?

 

[blueone]

What about x86 chiplets?

That is a tough question, I think only answerable by talking to potential customers. Intel x86 processors drag along a lot of baggage with them. Memory controllers and management, I/O chiplets, management circuits... if that's all still true, the result will just be a customer-designed Intel x86 CPU. I don't know if there's a market for that.

 

[hist78]

Microsoft, Google, and Amazon make more than cloud chips and they write some big checks inside the ecosystem. They are big name companies and pull a lot of weight as a named customer. If you are an EDA or IP company and have Google, Microsoft, and Amazon as customers TSMC will listen closely to you, absolutely.

Taking NOT TSMC business from Samsung does not take away business from Intel.

Let's do some calculations:

Assumptions:

1. Amazon needs Intel Foundry to manufacture an Amazon designed processor for server.
2. Amazon needs one millions units each year for three years.
3. Assume a 300mm wafer can be cut into 300 units of such processors and assume 90% yield.

Yearly wafers needed: 3,704 = 1,000,000 / ( 300 * 0.9 )
Monthly wafers needed: 309 = 3,704 /12

So, to fulfill Amazon’s order, Intel Foundry would need to produce approximately 3,704 wafers per year, or 309 wafers per month.

For a modern fab with a monthly production capacity of 30,000 to 50,000 300mm wafers, should we consider Amazon’s 309-wafer monthly demand a significant number? Does a customer requiring 309 wafers per month qualify as a "whale"?

Feel free to substitute your own numbers to see how the comparison changes.

 

[XYang2023]

Let's do some calculations:

Assumptions:

1. Amazon needs Intel Foundry to manufacture an Amazon designed processor for server.
2. Amazon needs one millions units each year for three years.
3. Assume a 300mm wafer can be cut into 300 units of such processors and assume 90% yield.

Yearly wafers needed: 3,704 = 1,000,000 / ( 300 * 0.9 )
Monthly wafers needed: 309 = 3,704 /12

So, to fulfill Amazon’s order, Intel Foundry would need to produce approximately 3,704 wafers per year, or 309 wafers per month.

For a modern fab with a monthly production capacity of 30,000 to 50,000 300mm wafers, should we consider Amazon’s 309-wafer monthly demand a significant number? Does a customer requiring 309 wafers per month qualify as a "whale"?

Feel free to substitute your own numbers to see how the comparison changes.

For 18A, it is not an issue. To break even, only a small amount of external orders is required.

 

[user nl]

Let's do some calculations:

Assumptions:

1. Amazon needs Intel Foundry to manufacture an Amazon designed processor for server.
2. Amazon needs one millions units each year for three years.
3. Assume a 300mm wafer can be cut into 300 units of such processors and assume 90% yield.

Yearly wafers needed: 3,704 = 1,000,000 / ( 300 * 0.9 )
Monthly wafers needed: 309 = 3,704 /12

So, to fulfill Amazon’s order, Intel Foundry would need to produce approximately 3,704 wafers per year, or 309 wafers per month.

For a modern fab with a monthly production capacity of 30,000 to 50,000 300mm wafers, should we consider Amazon’s 309-wafer monthly demand a significant number? Does a customer requiring 309 wafers per month qualify as a "whale"?

Feel free to substitute your own numbers to see how the omparison changes.

Very interesting discussion. I wonder what all the Intel Employess are thinking going over all these discussion threads and arguments in SemiWiki? (Artificer60, Dan: any feedback from your Intel contacts?)

Suppose the wafer price would be say 45 k$, this order would bring in 3704*45 = 167 M$ / year, right?

Not sure if this changes the needle sufficiently at Intel Foundry even if they produce this at 40% Foundry gross margin?

I have no idea about wafer prices, but here some "rumors":
https://www.tomshardware.com/tech-i...ent-increase-in-pricing-over-prior-gen-wafers

 

[benb]

My base case is that Intel is locked into a death spiral of downsizing, cost cutting, de-investment (50% GM required!) and has a few quarters left. I’m not really committed to that position, but need to see the following to re-evaluate it:
—Intel can mass produce an EUV product with a large die size with good yields and turn a profit on it (ie performance is improved enough to justify the added cost of the newer node).
—Intel can find a new pool of capital and make a deal that doesn’t result in shutting down R&D or saves design while de-facto shutting down Foundry.

The talk about whales and such is just that, talk. Intel needs to make a deal to save itself, but not with customers, with financiers. Only once the financial foundation is in place, can you start to convince customers.