Intel chief ready to go it alone on chip manufacturing

[Daniel Nenni]

(Reuters) - Intel Chief Executive Pat Gelsinger on Thursday said the company has no plans to spin out its contract chip manufacturing business.

Gelsinger has moved Intel's manufacturing unit into what is now called Intel Foundry Services (IFS) that operates as a business within Intel. IFS will break out financials beginning in the second quarter of next year, Gelsinger said.

Intel isn't yet prepared to break out IFS into a separate entity and list it, as it did with its Mobileye autonomous driving business and plans to with the programmable chip unit in the next two to three years.

"The idea of the internal foundry model, we think, is the right path for us in the current environment," Gelsinger said in an interview with Reuters.

In some ways, Intel already operates two separate companies - a chip design business and a factory unit - in part to give its IFS customers confidence that Intel is a "clean supplier" of manufacturing capacity, Gelsinger said.

For the moment, there are distinct advantages by operating together, in part because most of the factory capacity is used by Intel at the moment, the CEO said.
Gelsinger made the comments about Intel's manufacturing business at an event in New York that focused on PC chips with artificial intelligence features.

Running AI applications from far away data centers is too costly for the likes of Microsoft and will have to be run on local PCs, Gelsinger said.

"There isn't any possible way that they can have a billion Windows devices hitting Azure to be running these workloads in real time," Gelsinger said.
To make the economics work, the Intel CEO said Microsoft would need to achieve a 100-to-one reduction in the amount of data flowing between the cloud giant's data centers and PCs.

Intel chief ready to go it alone on chip manufacturing

Intel Chief Executive Pat Gelsinger on Thursday said the company has no plans to spin out its contract chip manufacturing business. Gelsinger has moved Intel's manufacturing unit into what is now called Intel Foundry Services (IFS) that operates as a business within Intel. IFS will break out...

www.yahoo.com

 

[Xebec]

Pat is ready for battle today. I personally really like it.

 

[blueone]

Running AI applications from far away data centers is too costly for the likes of Microsoft and will have to be run on local PCs, Gelsinger said.

"There isn't any possible way that they can have a billion Windows devices hitting Azure to be running these workloads in real time," Gelsinger said.
To make the economics work, the Intel CEO said Microsoft would need to achieve a 100-to-one reduction in the amount of data flowing between the cloud giant's data centers and PCs.

I agree with Pat about what is essentially edge or client computing for many apps, including AI, though I wonder where the 100:1 reduction in internet data flow requirement for AI apps came from? For example, if you put a database in the cloud and run the DBMS near where the data is stored, there's actually very little data (just the query and then the answer set) transmitted to and from the client or edge device. Formulating the answer set might involve processing terabytes or petabytes of data, but that mstly stays in the cloud data centers. Some aspects of AI, though not all I suspect, will be similar. ChatGPT doesn't require much in the way of client to cloud dataflow, like the cloud database example, though for something like ADAS you definitely want processing capability in the vehicles. Also, Amazon, Google, and Microsoft already deal with billions of clients simultaneously communicating with cloud data centers. I must be missing something important, because I'm not seeing the 100:1 issue.

 

[Mooredaddy]

Pat is ready for battle today. I personally really like it.

View attachment 1563

I get it but man, talk about making enemies. Going to have a hell of a time fight Apple, TSMC and its entire ecosystem, Nvidia, AMD etc. Pat is putting them on the warpath against any and everyone which is brave but I don’t know how it will turn out well

 

[Tanj]

I must be missing something important, because I'm not seeing the 100:1 issue.

Conversely there is a data flow issue with the edge use of large models - updating the models. Patching the OS on a regular basis is already a significant load, and that would be tiny compared to regular updates of a "small" 7GB model.

Only things which absolutely have to keep running when the network is unavailable, and which have relatively unchanging models, will be best sited at a mobile device. Plenty of candidates might make that cut in a vehicle, but relatively few in a PC or a phone.

We could do updates better with a broadcast model, but that is a problem not yet solved. Probably something in the middle, like storing a shared update at cell towers and then downloading the last miles on idle capacity, maybe shared by multiple recipients. A lot of implementation not yet done.

 

[MKWVentures]

So if AI is needed everywhere, does that mean Gen 12 and 13 laptops are useless? Gen 12 still outsells Gen 13.... nevermind 14. and Samsung phones still outsell Apple phones and phones outsell PCs 5 to 1. We will see what actually happens but I am not sure downloading models to the edge is the solution

Side note: I want to buy a meteor lake PC for Christmas... I can't find one.... can you send me a link with one?

 

[Tanj]

Side note: I want to buy a meteor lake PC for Christmas... I can't find one.... can you send me a link with one?

Easy to find that topic today, don't know if they will ship for Christmas:

The best Core Ultra laptops that just got announced | Digital Trends

Intel's new Core Ultra chips have been announced, along with a slew of new supporting laptops. Here are the best of them.

www.digitaltrends.com

Using an Intel Core Ultra laptop for anything other than AI is a bit meh — my time with the new Asus Zenbook 14 OLED

Does your workflow not involve AI? Save yourself some money

www.tomsguide.com

 

[Xebec]

I get it but man, talk about making enemies. Going to have a hell of a time fight Apple, TSMC and its entire ecosystem, Nvidia, AMD etc. Pat is putting them on the warpath against any and everyone which is brave but I don’t know how it will turn out well

I know times are different today.. but Pat was sorta ‘raised’ either during the era, or through the leaders that took Intel from near destitution in the 70s and early 80s to THE juggernaut of the late 90s and beyond. Motorola (Operation Crush), the Japanese memory market supremacy, etc. were all significant challenges of the time that Intel overcame. The challenges are different but also with similarities today - Intel didn’t have the most advanced chip or fab tech back then.

The battle with Apple isn’t really a big deal - at this point Intel is always going to be a competitor to Apple. (Unless Intel can really leap far ahead of TSMC and somehow convince Apple to fab with Intel). Apple is more than successfully competing with Intel by using both PCs (Mac) and post-PC (iOS) type devices - so Intel needs to do something.

Agreed, Pat shouldn’t do too much to upset TSMC, but TSMC will also gladly keep taking Intel money as they also need a lot of large customers to keep funding advanced fabs. TSMC still needs Intel as a customer and partner, although not nearly as much as Intel needs access to TSMC fabs over the long haul.

The Nvidia competition is interesting; Intel has a lot to gain AND a lot to lose. Gain: GPU, AI, Server markets, Lose: Nvidia as a potential fab customer. The potential gains seem to be much higher margin for Intel than the potential loss market.

 

[Xebec]

So if AI is needed everywhere, does that mean Gen 12 and 13 laptops are useless? Gen 12 still outsells Gen 13.... nevermind 14. and Samsung phones still outsell Apple phones and phones outsell PCs 5 to 1. We will see what actually happens but I am not sure downloading models to the edge is the solution

Side note: I want to buy a meteor lake PC for Christmas... I can't find one.... can you send me a link with one?

SFG14-72-72YK - Tech Specs | Laptops | Acer United States

SFG14-72-72YK - Tech Specs | Laptops | Acer United States

www.acer.com

https://www.bestbuy.com/site/asus-zenbook-14-oled-14-3k-touch-laptop-intel-core-ultra-7-32gb-memory-1tb-ssd-ponder-blue/6569307.p?skuId=6569307

 

[blueone]

So if AI is needed everywhere, does that mean Gen 12 and 13 laptops are useless? Gen 12 still outsells Gen 13.... nevermind 14. and Samsung phones still outsell Apple phones and phones outsell PCs 5 to 1. We will see what actually happens but I am not sure downloading models to the edge is the solution

Side note: I want to buy a meteor lake PC for Christmas... I can't find one.... can you send me a link with one?

Costco has an MSI 16" Core Ultra 7 with 32GB/2TB for $1400, members only:

https://www.costco.com/msi-prestige...560-x-1600-windows-11.product.4000230800.html

 

[blueone]

So if AI is needed everywhere, does that mean Gen 12 and 13 laptops are useless?

Unlikely. Intel has positioned the NPUs in the Core Ultras as "accelerators", which means the AI functions will run but slower in CPUs without NPUs. It's the same way with AVX512 floating point acceleration; AVX512 is not in every CPU.

FWIW, I'm skeptical of client AI capabilities being indispensable for what most people do with PCs, meaning email, web browsing, gaming, streaming, video/photo editing, MS Office functions, and so on. There are some professional apps I've heard about that would probably benefit from AI acceleration in clients (like medical image processing and analysis), but I haven't read about any broad user-base apps yet. Have you?

 

[Tanj]

Gain: GPU, AI, Server markets, Lose: Nvidia as a potential fab customer. The potential gains seem to be much higher margin for Intel than the potential loss market.

Exactly why IFS needs to be its own company. Pat may love them both equally, but they cannot have the same customer relationships.

 

[nghanayem]

Exactly why IFS needs to be its own company. Pat may love them both equally, but they cannot have the same customer relationships.

This is often stated, but I don't know if I believe that. Why can't there be two sandboxes? Competitors collaborating when there is a common interest is not exactly unusual. They're companies not rival ideologies. If your argument is that IFS could never overtake TSMC as a foundry because you can get the trust close but not all the way; sure I could get behind that. But I just don't buy that nobody would ever work with IFS even if the company is split in half like it is now. I guess there is no way to prove things one way or the other. But for funziez let's say tomorrow IFS-TD-MSO was now Moore-Noyce Semiconductor with intel owning 0% stake in this new company. At the last earnings they said they signed 3 18A customers and had line of sight on a fourth by EOY. Does anyone think that Moore-Noyce would be wildly more successful than IFS? I don't. I think the priorities go something like this 50% foundry service, 40% PPAC of the tech, and 10% pureplay vs IDM (assuming there is sufficient fire-walling for customers to trust their partner). If we see intel get to like I don't know maybe 30% of the external N and N+1 market, then I think I could see not being a pureplay foundry starting to be an impediment to growing share (assuming process and service are comparable). The question is if intel got to that point would it be to intel's benefit to divest themselves of a major growth engine that is also accretive to company wide margins? Again the answer is no, because why would an rational investor ever want a company to do that? Maybe a mobileye situation where intel gets to siphon away the profits, but at that point how is having Pat lead the BOD any different than having Pat has CEO of Moore-Noyce?

 

[Tanj]

This is often stated, but I don't know if I believe that. Why can't there be two sandboxes?

It is not just two sandboxes, it is two kinds of industry, and they are different from one end to the other, from customers to use of capital. Why would investors attracted to putting capital into solid, predictable manufacturing want to see their money diluted in high risk volatile chip designs? Conglomerates like GE at least tried to have a general similarity to their portfolio.

how is having Pat lead the BOD any different than having Pat has CEO of Moore-Noyce?

I agree there is no immediate split, they are deeply connected and together need to make the journey together for the next 2 or 3 years. Beyond that, each has to stand on its own feet. A three-legged pair will not win when everyone else can run freely.

 

[blueone]

It is not just two sandboxes, it is two kinds of industry, and they are different from one end to the other, from customers to use of capital. Why would investors attracted to putting capital into solid, predictable manufacturing want to see their money diluted in high risk volatile chip designs? Conglomerates like GE at least tried to have a general similarity to their portfolio.

The evidence is that investors are bought into Intel's strategy, since the stock has doubled from the 52 week low to the 52 week high as of this posting. Even more curious, INTC's PE ratio divides by a negative number, while TSMC's PE ratio is an embarrassing 18.5, though probably that's due almost entirely to geopolitical risk. If investors are leery of Intel and Gelsinger's IDM strategy, there's little sign of that in the equity market.

 

[MKWVentures]

The evidence is that investors are bought into Intel's strategy, since the stock has doubled from the 52 week low to the 52 week high as of this posting. Even more curious, INTC's PE ratio divides by a negative number, while TSMC's PE ratio is an embarrassing 18.5, though probably that's due almost entirely to geopolitical risk. If investors are leery of Intel and Gelsinger's IDM strategy, there's little sign of that in the equity market.

When you see the finances for IFS, Capex, development costs. It may become clear (IMO) why you cannot split the two companies up YET. It will take 5+ years to get the IFS revenue to the point of being self supporting. There are lot of reasons why this will be slower than Intel thinks

 

[blueone]

When you see the finances for IFS, Capex, development costs. It may become clear (IMO) why you cannot split the two companies up YET. It will take 5+ years to get the IFS revenue to the point of being self supporting. There are lot of reasons why this will be slower than Intel thinks

What does Intel think?

 

[cliff]

The engineer at Intel probably thinks there are a lot of short sighted namby pambies in this forum.

The whole is worth more than the sum of the parts. Trim the fat and play to win.

 

[MKWVentures]

What does Intel think?

Intel originally said they want to be second largest foundry (external customers) by the end of decade. this would mean ~10B in revenue per year external. Intel revised this to say they will be the second largest foundry next year. this is because Intel will count internal sales as foundry. How they value internal is highly debatable. Intel has a plan for external but it is not shared with public. I published a model for intel external sales through 2028. I can review details of our model and Intel expectations and how the accounting will work but it is a long discussion.

 

[cliff]

The world is changing rapidly, especially in our sector. IMO, revenue projections are pure BS.

Intel geographically is in the sweet spot. The USG won't let them fail. Build it. Keep it. Put your company in a position to capitalize when changes occur. Constrict and expand... yes. Divest... nah. Don't listen to shareholders!