Another Perspective on Intel's Future: How Can They Generate More Revenue?

[Maximus]

There are already numerous posts in this forum discussing Intel's future business model. But let's consider another angle: How can Intel increase its revenue?

In a recent interview, Pat mentioned that Intel would be in a healthy state if revenue reached $75 billion. Intel’s 2024 revenue forecast is around $50-55 billion. Considering the time it will take to reach that healthy revenue level and factoring in inflation, a reasonable target might be $80-85 billion by 2028/2029, which is about $30 billion more than the current level.

Where could this additional $30 billion come from in 2028/2029? My estimate is $5 billion from IFS and $25 billion from products.

 

[XYang2023]

5-10b from GPU/accelerators?

 

[Barnsley]

Carbon Trading?

Its working for TESLA

 

[blueone]

5-10b from GPU/accelerators?

The Gaudi 3 revenue projections are not promising. Intel hasn't made an Xe GPU announcement since 2023, so it's difficult to get excited about Intel's GPU roadmap. I haven't seen revenue figures for Intel's IPU products they designed with Google, but I can't believe that revenue exceeds $1B/yr, and I suspect it's substantially less.

 

[blueone]

Where could this additional $30 billion come from in 2028/2029? My estimate is $5 billion from IFS and $25 billion from products.

I know everyone here will snarl when they read this, but in my opinion if the big revenue ramp doesn't come from IFS, it isn't coming. If Intel came up with a breakthrough mobile processor line it could win over Samsung and Google, but I see nothing that indicates that. Lunar Lake could steal client share back from AMD, which is probably the best product opportunity for big revenue gains, but that's not going to fill a $30B hole. Data Center CPU revenue, and especially margins, don't look good for the foreseeable future. I think Intel really is betting the company on IFS, like it or not.

 

[BruceA]

I know everyone here will snarl when they read this, but in my opinion if the big revenue ramp doesn't come from IFS, it isn't coming. If Intel came up with a breakthrough mobile processor line it could win over Samsung and Google, but I see nothing that indicates that. Lunar Lake could steal client share back from AMD, which is probably the best product opportunity for big revenue gains, but that's not going to fill a $30B hole. Data Center CPU revenue, and especially margins, don't look good for the foreseeable future. I think Intel really is betting the company on IFS, like it or not.

I have to agree with blue.

DCAI has a formidable competitor in AMD and they will never again have a one or two node process advantage so at best they trade a few % in performance and will never have cost advantage. All the big buyers will keep a two player source.

AI sadly Intel is on the outside looking in just like mobile. That ship sailed and they msissed it and are not going to get anything here.

In the client space as noted with Lunar they should finally have process parity with Apple and superior to AMD and ARM for mobile compute. This should give them hopefully a low single digit % performance / watt and maybe they can claw back some market share here, a few billion at best. Sadly this is a mature market that at best and there won’t be a repeat boom like COVID for AI IMHO, best case is going to grow low single digits.

Sadly the only way Intel gets back and get real revenue is Si manufacturing. The world wants a second source for leading edge that ain’t TSMC so they have a a sliver of possible here. So that has to be on the back of a successful 18A ramp of Panther and a few early low volume adopters and than a success delivery of 14A and a significant real couple whales.

Intel needs a perfect alignment! Perfection in execution of technology on schedule, successfully PDK deliveries, IP landing on time with no hiccups from the PDK. Let’s not forget multiple successful yield ramps too. Wow, something they haven’t done all togather once this decade, tall order.

Now with the layoff, huge losses they simultaneous need to build out Az and get Ohio ready to scale yet they don’t have the money to do all that and TD Oregon. Forget the Germany fantasy for the next decade. Maybe they can get the government and some rich Oil state to cough up the thirty billion they need the next five years to put them in place to do this.

You think the current leadership has it in them with all the distractions about money and such?

 

[hist78]

There are already numerous posts in this forum discussing Intel's future business model. But let's consider another angle: How can Intel increase its revenue?

In a recent interview, Pat mentioned that Intel would be in a healthy state if revenue reached $75 billion. Intel’s 2024 revenue forecast is around $50-55 billion. Considering the time it will take to reach that healthy revenue level and factoring in inflation, a reasonable target might be $80-85 billion by 2028/2029, which is about $30 billion more than the current level.

Where could this additional $30 billion come from in 2028/2029? My estimate is $5 billion from IFS and $25 billion from products.

Should Intel shrink to become a smaller company (whatever format) but a profitable company? For example, can Intel become a $30 billion revenue company with 20% net profit? Or should Intel maintain a $50 billion revenue with loss?

 

[osnium]

I have to agree with blue.

DCAI has a formidable competitor in AMD and they will never again have a one or two node process advantage so at best they trade a few % in performance and will never have cost advantage. All the big buyers will keep a two player source.

AI sadly Intel is on the outside looking in just like mobile. That ship sailed and they msissed it and are not going to get anything here.

In the client space as noted with Lunar they should finally have process parity with Apple and superior to AMD and ARM for mobile compute. This should give them hopefully a low single digit % performance / watt and maybe they can claw back some market share here, a few billion at best. Sadly this is a mature market that at best and there won’t be a repeat boom like COVID for AI IMHO, best case is going to grow low single digits.

Sadly the only way Intel gets back and get real revenue is Si manufacturing. The world wants a second source for leading edge that ain’t TSMC so they have a a sliver of possible here. So that has to be on the back of a successful 18A ramp of Panther and a few early low volume adopters and than a success delivery of 14A and a significant real couple whales.

Intel needs a perfect alignment! Perfection in execution of technology on schedule, successfully PDK deliveries, IP landing on time with no hiccups from the PDK. Let’s not forget multiple successful yield ramps too. Wow, something they haven’t done all togather once this decade, tall order.

Now with the layoff, huge losses they simultaneous need to build out Az and get Ohio ready to scale yet they don’t have the money to do all that and TD Oregon. Forget the Germany fantasy for the next decade. Maybe they can get the government and some rich Oil state to cough up the thirty billion they need the next five years to put them in place to do this.

You think the current leadership has it in them with all the distractions about money and such?

Given all these challenges, do you think the core engineering staff working on Intel 18A and 14A will choose to stay? If you were in their position, would you stay? On one hand, if you are an engineer working on these nodes and if the node ships on time, it’s a good achievement to be associated with. On the other hand, it’s very risky and if the org goes bust, the flood of laid off engineers will fill all the fab openings for years to come.

 

[XYang2023]

The Gaudi 3 revenue projections are not promising. Intel hasn't made an Xe GPU announcement since 2023, so it's difficult to get excited about Intel's GPU roadmap. I haven't seen revenue figures for Intel's IPU products they designed with Google, but I can't believe that revenue exceeds $1B/yr, and I suspect it's substantially less.

I don't think 5-10b is difficult to achieve:
1. It is near the end of the decade
2. Intel has software assets to bundle with the accelerators
3. It complements with Xeon
4. The overall market is quite large
5. Falcon Shores line has a yearly cadence. It is a XPU platform, as I can understand it.

 

[XYang2023]

There are already numerous posts in this forum discussing Intel's future business model. But let's consider another angle: How can Intel increase its revenue?

In a recent interview, Pat mentioned that Intel would be in a healthy state if revenue reached $75 billion. Intel’s 2024 revenue forecast is around $50-55 billion. Considering the time it will take to reach that healthy revenue level and factoring in inflation, a reasonable target might be $80-85 billion by 2028/2029, which is about $30 billion more than the current level.

Where could this additional $30 billion come from in 2028/2029? My estimate is $5 billion from IFS and $25 billion from products.

If the fab only brings in 5b external foundry businesses, it does not make sense to split out IFS. Another function of IFS is margin stacking.

Mobileye and Altera could generate that amount of revenue at that time frame.

What I wrote previously:
1. Split out the foundry business, e.g. merge with GF with the backing from Saudi
2. Use the proceeds to payoff debts
3. Cancel the splits of Mobileye and Altera and form a group to focus on automotive and edge AI
4. Expedite GPU/accelerator development
5. Provide software solutions, for example, something similar to Nvidia's AI fundary, digital twins, etc
6. Develop competitive client and data centre CPUs
7. Improve the efficiency of the organisation

Maybe can change 1. to TSMC instead of GF.

 

[DaRAGingLunatic]

I think the Saudi's are glad to be out of Global Foundries.

I don't see them dishing out for foundries that are losing billions of dollars quarterly.

 

[Jozo035]

Saudis are broke anyway

 

[DaRAGingLunatic]

I think Intel should think about training costs for A.I.

So for instance Cerberus makes those large wafer scale chips.
You use those wafer sized chips to train A.I on.
A lot of companies are paying huge amounts of money for hardware or just to train on available hardware. This training can take MONTHS! I've heard some of the large costs such as $500 million, $1 billion, soon they estimate $10 billion.
Cerberus is currently the only company that I know of using wafer scale chips, and using them to scale down training time.

Lets say you use one of Cerberus's machines to train a 24 trillion parameter A.I.
Imagine it takes 1 whole year.
Well Cerberus can scale to 2048 machines.
1 machine = 365 days
2 machines = 182.5 days
4 machines = 91.25 days
8 machines = 45.625 days
16 machines = 22.8125 days
32 machines = 11.40625 days
64 machines = 5.703125 days
128 machines = 2.8515625 days
256 machines = 1.42578125 days
512 machines = 0.712890625 days
1024 machines = 0.3564453125 days
2048 machines = 0.17822265625 days = 134.6600622452111 minutes LOL look at those decimals hahahaha

All i'm saying is Intel has fabs, has engineering teams. Could build a wafer scale machine like Cerberus. Scale it out ASAP, Become an AI training service.

Instead of people buying tons of AI hardware for training purposes, they would end up using your service to train their AI because not only is it fast as all hell, but will be cheaper too.

You charge electricity costs + whatever amount of profit to cover the build costs, R&D, and profit margin, etc.

People will come to you for training as a service.

You just have to scale it out each new fab node ASAP to remain competitive, which is something you can do.

 

[Barnsley]

Its the UAE that are majority shareholder in GF.

Why folk keep saying Saudi , unless they imply Saudi own UAE

 

[XYang2023]

I think Intel should think about training costs for A.I.

So for instance Cerberus makes those large wafer scale chips.
You use those wafer sized chips to train A.I on.
A lot of companies are paying huge amounts of money for hardware or just to train on available hardware. This training can take MONTHS! I've heard some of the large costs such as $500 million, $1 billion, soon they estimate $10 billion.
Cerberus is currently the only company that I know of using wafer scale chips, and using them to scale down training time.

Lets say you use one of Cerberus's machines to train a 24 trillion parameter A.I.
Imagine it takes 1 whole year.
Well Cerberus can scale to 2048 machines.
1 machine = 365 days
2 machines = 182.5 days
4 machines = 91.25 days
8 machines = 45.625 days
16 machines = 22.8125 days
32 machines = 11.40625 days
64 machines = 5.703125 days
128 machines = 2.8515625 days
256 machines = 1.42578125 days
512 machines = 0.712890625 days
1024 machines = 0.3564453125 days
2048 machines = 0.17822265625 days = 134.6600622452111 minutes LOL look at those decimals hahahaha

All i'm saying is Intel has fabs, has engineering teams. Could build a wafer scale machine like Cerberus. Scale it out ASAP, Become an AI training service.

Instead of people buying tons of AI hardware for training purposes, they would end up using your service to train their AI because not only is it fast as all hell, but will be cheaper too.

You charge electricity costs + whatever amount of profit to cover the build costs, R&D, and profit margin, etc.

People will come to you for training as a service.

You just have to scale it out each new fab node ASAP to remain competitive, which is something you can do.

Or via partnership:

Intel and IBM Collaborate to Provide Better Cost Performance for AI Innovation

IBM and Intel have announced a collaboration to deploy Intel® Gaudi® 3 AI accelerators as a service on IBM Cloud.

newsroom.ibm.com

 

[DaRAGingLunatic]

Yeah well a partnership could be good as well.

 

[blueone]

You think the current leadership has it in them with all the distractions about money and such?

I don't think Gelsinger does, and distractions aren't the problem.

 

[hist78]

I think Intel should think about training costs for A.I.

So for instance Cerberus makes those large wafer scale chips.
You use those wafer sized chips to train A.I on.
A lot of companies are paying huge amounts of money for hardware or just to train on available hardware. This training can take MONTHS! I've heard some of the large costs such as $500 million, $1 billion, soon they estimate $10 billion.
Cerberus is currently the only company that I know of using wafer scale chips, and using them to scale down training time.

Lets say you use one of Cerberus's machines to train a 24 trillion parameter A.I.
Imagine it takes 1 whole year.
Well Cerberus can scale to 2048 machines.
1 machine = 365 days
2 machines = 182.5 days
4 machines = 91.25 days
8 machines = 45.625 days
16 machines = 22.8125 days
32 machines = 11.40625 days
64 machines = 5.703125 days
128 machines = 2.8515625 days
256 machines = 1.42578125 days
512 machines = 0.712890625 days
1024 machines = 0.3564453125 days
2048 machines = 0.17822265625 days = 134.6600622452111 minutes LOL look at those decimals hahahaha

All i'm saying is Intel has fabs, has engineering teams. Could build a wafer scale machine like Cerberus. Scale it out ASAP, Become an AI training service.

Instead of people buying tons of AI hardware for training purposes, they would end up using your service to train their AI because not only is it fast as all hell, but will be cheaper too.

You charge electricity costs + whatever amount of profit to cover the build costs, R&D, and profit margin, etc.

People will come to you for training as a service.

You just have to scale it out each new fab node ASAP to remain competitive, which is something you can do.

Intel can definitely start offering AI compute services, but what business model does Intel really want? It needs to make up its mind. If Intel begins offering compute services, it will likely end up competing against many of its own cloud service customers, such as Google, Amazon, and Microsoft. The reaction and outcome could be very ugly.

 

[delong.height]

Intel definitely can start offering AI Compute service. But what the business model Intel really wants? Intel needs to make up its mind. If Intel starts offering compute services, it will probably start competing against most Intel cloud service customers, such as Google, Amazon, Microsoft, etc. The reaction and outcome can be very ugly.

Nvidia is competing with its customers, Google, Amazon, Microsoft are all competing with their supplier with their in-house silicon. So why can't Intel compete with its offerings?

 

[count]

Should Intel shrink to become a smaller company (whatever format) but a profitable company? For example, can Intel become a $30 billion revenue company with 20% net profit? Or should Intel maintain a $50 billion revenue with loss?

This is what I think should happen.

Intel could spin off or sell everything that isn't profitable, except for GPU/AI.

It is still early days in AI, and if Intel got rid of it's other capital sinks, and kept just the x86 cash cow along with the AI/GPU business, they could invest $10b a year into GPU/AI accelerator development and associated software ecosystem.