Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/intel-needs-another-andy-grove.17376/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Intel needs another Andy Grove

Arthur Hanson

Well-known member
Having a one-on-one conversation with an Intel admin years ago, plus what I have read, Intel needs another Andy Grove or a modern version to propel Intel back into the powerhouse it used to be. A team can only be as good as its leader and vision is critical. An open mind to recognize and develop new talent is also critical, an organization is only as strong as the people in it. The ability to assemble effective teams is critical and with the semi ecosystem getting more complex every generation more talent than ever is needed at an ever increasing scale. Competition brings out the best and TSM will provide it, Intel just has to up its game.
 
The Andy Grove of today would realize that there is no future in being an leading-edge IDM with a rump foundry division. Intel needs to do what AMD did and separate manufacturing from chip design. Let the foundry work on process development, ramp and yields (and per wafer costs). Nothing else makes sense at this point.
 
The Andy Grove of today would realize that there is no future in being an leading-edge IDM with a rump foundry division. Intel needs to do what AMD did and separate manufacturing from chip design. Let the foundry work on process development, ramp and yields (and per wafer costs). Nothing else makes sense at this point.
This would literally bankrupt the both of them. AMD was on the road to chapter 11, but oil barons bought all of their debt and their foundry (which AMD was locked into using for years). AMD did not choose to go fabless they were forced to, and fortunately for them had suckers who made the transition possible. Even after going fabless AMD almost went bankrupt again.

Option A if intel was to do the same, who would buy intel's gargantuan debt and fab network that is bigger than Samsung's logic business (and in 2021 intel had 80% the capacity of TSMC's leading edge)? TSMC can't just flip a switch and magically increase their leading edge capacity by 30-40% to accommodate intel, so do you recommend that intel doesn't sell products for the next 5 years until TSMC can build out that capacity?

Option B go fablite (make less advanced tiles and trailing edge products internally while slowly exporting the leading edge bits) and probably do okay but not great. For this option the design side must crush AMD and cloud providers to be competitive as there are fewer areas to show differentiation.

Option C fix your process development and design house to have an even greater advantage than otherwise possible if only the design side was crushing it. If you add foundry then when times are good you have another way to monetize your investment in your nodes, and if design is fumbling you have a lifeline. A fixed fab also have the benefit of lower wafer cost than TSMC+margin.

The way I see it option A only leads to ruin, option B can be successful but offers the lowest reward while still requiring years to get to technological parity, and option C offers the greatest rewards but the greatest risk due to the extra capital for capex and R&D.
 
Last edited:
Well, I am not here to start WW3.
AMD market cap today is higher than intel, i would hardly called that unsuccessful. GF may not be a great entity but they have a lot of intrinsic worth (if no aspiration to compete at the top end).
All leading edge intel chips are made by tsmc. So separating design from manufacturing changes nothing and "Intel manufacturing" could always get share back if process is superior or cheaper.
At the core, intel is addicted to the large margins that come from server chips and MPUs. That's used to hide a lot of inefficiencies on the manufacturing side of things. tsmc does not have that fig leaf and in the very long run, it made a very big difference.
 
AMD market cap today is higher than intel, i would hardly called that unsuccessful. GF may not be a great entity but they have a lot of intrinsic worth (if no aspiration to compete at the top end).
Yes, but if one looks at revenue or marketshare they haven't quite earned it yet. When AMD has 50% server market share is when you can say they are a bigger company than intel. Right now they have 36% the revenue and like 10% the employees. That success also came after years of only staying afloat with multiple lifelines (console and oil barons). Who will be intel's lifeline over the 15 year journey of becoming a successful fabless firm?

All leading edge intel chips are made by tsmc. So separating design from manufacturing changes nothing and "Intel manufacturing" could always get share back if process is superior or cheaper.
Just flat out wrong. MTL is mostly TSMC but the most advanced parts are still made by intel. ADL, RPL, SPR, TGL, a good portion of the FPGA business, chipsets, and packaging/forvoros/emib dies are all 100% intel.

At the core, intel is addicted to the large margins that come from server chips and MPUs. That's used to hide a lot of inefficiencies on the manufacturing side of things. tsmc does not have that fig leaf and in the very long run, it made a very big difference.
Yes, and Pat agrees with you. Hence the internal foundry model he proposed to fix these hidden inefficiencies. Getting off of SAQP and even fixing half of these inefficiencies will probably allow for intel to be cheaper than TSMC+fat margins.
 
Last edited:
From the Gelsinger interview going on now, it sounds like Intel has become a welfare case for the US government. Putting tax payer money into Intel and TSM are two totally different propositions. One has a long term record of success and very sadly Intel is living in the past. Intel should not be paying a dividend when they are not executing financially or technology wise. Gelsinger keeps talking about government grants to support a model that so far is failing. Any thoughts or comments on this are appreciated. Does anyone see any chance of Intel catching up without massive government support. I would rather see my tax money going to TSM facilities in this country. It looks like Biden and Gelsinger are not acting in the best interest of the people they represent at this point in time. It looks like supporting Intel could be a spoiler for the well run companies in the sector and the street has totally lost faith in Intel, especially with their road map totally up in the air right now.


















Gelsinger keeps talking about government grants to support a model that so far is failing. Any thoughts or comments on this are appreciated.
 
Last edited:
It's no longer possible to be a logic IDM (at the finer geometries). TSMC is able to manufacturing at economical scale (i.e. in very large fabs), by combining multiple customers. When one drops, another one picks up. This lessens the risk. If AMD's sever is having a bad year, they can make it up with NVidia. That model is the only sustainable model in the logic world (analog like Ti is different).

The long term net is that intel will build fabs in the USA (in high cost areas) without a solid underlying business plan. Ongoing subsidies may be needed.

BTW: the same is likely to be true for the TSMC fabs but to a lesser extent (they are inhenrently more cost efficient). I do not see NVidia or Apple paying 20-30% more for their chips. But hey, we will see.
 
Way too early to say if the model is failing. That is like saying the zen architecture is failing back when excavator was being released. My book Pat gets until 2025 to show he can execute and 2028 to see if his vision is any good. i4/3 was being worked on back in the BK days and with much of the share of getting it HVM ready probably happening during BS's days. Same deal with SPR, ADL, MTL, ect. 20/18A, ARL, ect were probably defined in the BS days, but much of the execution was done under Pat's watch/during intel's engineering culture transformation. Things after 2025 can probably be completely contributed to him just like how Zen4 and beyond is completely the result of the new AMD team (rather than Jim and Read's teams who can take most of the responsibility for Zen1 and 2/building the new AMD culture).
 
One final thought: DoD needs a stable supply of cutting edge chips (albeit in low volume) and they seek to DENY that same supply to foes. Money does not come into it from their point of view. Taiwan is not a foe but it could be hobbled in by blockade or hot war. We must keep this in mind when it comes to subsidies and incentives. We may see incentives to private entities for a very long time indeed.

Splitting Intel would still meet the objectives of DOD, provided both entities stay in business of course.

@nghanayem Wall Street will not wait until 2025 or 2028. see NVidia = $500b and Intel $115b. I would argue that the value of the intel chip portfolio minus the manufacturing assets is LARGER than the $115b.
 
This would literally bankrupt the both of them. AMD was on the road to chapter 11, but oil barons bought all of their debt and their foundry (which AMD was locked into using for years). AMD did not choose to go fabless they were forced to, and fortunately for them had suckers who made the transition possible. Even after going fabless AMD almost went bankrupt again.

Option A if intel was to do the same, who would buy intel's gargantuan debt and fab network that is bigger than Samsung's logic business (and in 2021 intel had 80% the capacity of TSMC's leading edge)? TSMC can't just flip a switch and magically increase their leading edge capacity by 30-40% to accommodate intel, so do you recommend that intel doesn't sell products for the next 5 years until TSMC can build out that capacity?

Option B go fablite (make less advanced tiles and trailing edge products internally while slowly exporting the leading edge bits) and probably do okay but not great. For this option the design side must crush AMD and cloud providers to be competitive as there are fewer areas to show differentiation.

Option C fix your process development and design house to have an even greater advantage than otherwise possible if only the design side was crushing it. If you add foundry then when times are good you have another way to monetize your investment in your nodes, and if design is fumbling you have a lifeline. A fixed fab also have the benefit of lower wafer cost than TSMC+margin.

The way I see it option A only leads to ruin, option B can be successful but offers the lowest reward while still requiring years to get to technological parity, and option C offers the greatest rewards but the greatest risk due to the extra capital for capex and R&D.
I’ll add a little bit to this — there are a lot of parallels in other markets.

I know not exactly comparable, but the full vertical integration is what’s allowing Tesla to do things other car manufacturers are struggling to do. OTOH - it makes sense why Ford went from fully vertically integrated to largely outsourced sometime in the 1940s-1960s. They did so as others innovated on individual pieces faster than they could internally, and they weren’t putting the right processes in place to stay ahead with their initial advantages. However, there were a lot more than 2-3 companies that could build those parts.

While there are some benefits to ‘giving up the fab businesses’ to reduce production risk, doing so also reduces the max possible (financial) reward. Pat was schooled and colored by Andy Grove — and I also think Andy would drive the fab business forward rather than out.
 
@Xebec There are parallels but there are also differences. The car business is mature in the sense that the optimal factory size is by now well known and that car prices increase each year or at least stay the same. The IC business has a growing factory size and a dropping ASP. If a factory did 1m transistors per day, it needs to do 2m next year and so on.

So in your analogy, a Ford factory would do 100,000 cars one year but another factory would be under construction that would do 200,000 the next year and so on and so forth.

TSMC says: factory sizes (and their corollary technology that makes that possible) and says "I will make cars for ALL comers." That will minimize my risk and drive my efficiency.

Tesla is making cars because foundries do not exist in that business for the most part.

Hon Hai is trying in Lorsdtown OH to become a car foundry. Jury still out (but deliberating).
 
@mozartct Time will tell the results...
I don't disagree with you on the gloomy financials. GNR, MTL, and i4/3 need to bring intel back to or near parity to slow the bleeding persicly because investors won't wait. My comments on 2025 and 2028 were more so me saying that it is too early to say if Pat's plan is sound, or if he is the next Andy Grove.
 
OTOH - it makes sense why Ford went from fully vertically integrated to largely outsourced sometime in the 1940s-1960s. They did so as others innovated on individual pieces faster than they could internally, and they weren’t putting the right processes in place to stay ahead with their initial advantages. However, there were a lot more than 2-3 companies that could build those parts.
Actually, GM and Ford didn't spinout their parts divisions until the late 1990s. Fiat-Chrysler didn't spinout its parts divisions until about 2018.
 
One major reason that helped the pure play foundry's rise was the IDM and IDM+foundry business model didn't work. It's been like this for the past 30+ years.

Why the old business model that hasn't worked for so many years will suddenly start working under Pat Gelsinger's new slogan or labels (IDM 2.0, IFS, Intel Internal Foundry, etc.)?
 
Actually, GM and Ford didn't spinout their parts divisions until the late 1990s. Fiat-Chrysler didn't spinout its parts divisions until about 2018.

Actually, GM and Ford didn't spinout their parts divisions until the late 1990s. Fiat-Chrysler didn't spinout its parts divisions until about 2018.
My timing is a little off - but I’m referring to this:

At one point you could literally dump in the raw materials on one end of the building (iron, rubber, etc) and come out with a completed car at the other end. This peaked in the 1930s-1950s, but “after the 1960s Ford began to decentralize”.

That decentralization and outsourcing of parts continued afterwards.
 
@Xebec There are parallels but there are also differences. The car business is mature in the sense that the optimal factory size is by now well known and that car prices increase each year or at least stay the same. The IC business has a growing factory size and a dropping ASP. If a factory did 1m transistors per day, it needs to do 2m next year and so on.

So in your analogy, a Ford factory would do 100,000 cars one year but another factory would be under construction that would do 200,000 the next year and so on and so forth.

TSMC says: factory sizes (and their corollary technology that makes that possible) and says "I will make cars for ALL comers." That will minimize my risk and drive my efficiency.

Tesla is making cars because foundries do not exist in that business for the most part.

Hon Hai is trying in Lorsdtown OH to become a car foundry. Jury still out (but deliberating).
I’m actually referring to when Ford used to design *and* build every single aspect of the car, rather than letting someone else take care of designing and supplying headlights, or those sorts of things.

Intel is both designing and manufacturing, like Ford (and some other makers) used to do long ago. That went out of vogue, for reasons - but now the tighter integration seems to be helping some of these newer companies (like Tesla).

.. Way off topic - but I feel Lordstown Motors, Rivian, Nikola, and Lucid are all companies designed just to put a couple of products out there and be bought.. not to stand on their own :). The Lordstown Motors story isn’t sounding too great but I hope all of these companies succeed at their goals.
 
@Xebec There are parallels but there are also differences. The car business is mature in the sense that the optimal factory size is by now well known and that car prices increase each year or at least stay the same. The IC business has a growing factory size and a dropping ASP. If a factory did 1m transistors per day, it needs to do 2m next year and so on.

So in your analogy, a Ford factory would do 100,000 cars one year but another factory would be under construction that would do 200,000 the next year and so on and so forth.

TSMC says: factory sizes (and their corollary technology that makes that possible) and says "I will make cars for ALL comers." That will minimize my risk and drive my efficiency.

Tesla is making cars because foundries do not exist in that business for the most part.

Hon Hai is trying in Lorsdtown OH to become a car foundry. Jury still out (but deliberating).

There is a car "foundry" already. Magna Steyr has been making cars for other companies for the past 20+ years. Currently Magna Steyr is making the following cars:

Mercedes-Benz G-Class (2018–present)
BMW 5 Series (2017–present)
Jaguar E-Pace (2017–present)
Jaguar I-Pace (2018–present)
BMW Z4 (2018–present)
Toyota Supra (2019–present)
W Motors Fenyr SuperSport (2019–present)
Fisker Ocean (2022–present)

Source: https://en.wikipedia.org/wiki/Magna_Steyr
 
My timing is a little off - but I’m referring to this:

At one point you could literally dump in the raw materials on one end of the building (iron, rubber, etc) and come out with a completed car at the other end. This peaked in the 1930s-1950s, but “after the 1960s Ford began to decentralize”.

That decentralization and outsourcing of parts continued afterwards.
Got it. But Rouge River was an industry one-off. Even Intel's fabs are nothing like iron ore going in and Model As come out. The major auto manufacturers used to do their own transmissions, differentials, suspension parts, audio systems, etc. They were much more vertical than Intel. IBM's mainframe days (360, 370) is probably the best computer industry example. They fab'd their own chips, stuffed their own boards, power supplies, bent metal, wrote their own OS and applications.

Tesla seems unique in doing its own ASICs, but they buy a lot of parts too. I agree, however, that EVs are easier to go vertical with. The lack of complex mechanical fabrication and assembly is very attractive to automakers. They need far less union labor per unit, which is also why unions despise EVs.
 
One major reason that helped the pure play foundry's rise was the IDM and IDM+foundry business model didn't work. It's been like this for the past 30+ years.

Why the old business model that hasn't worked for so many years will suddenly start working under Pat Gelsinger's new slogan or labels (IDM 2.0, IFS, Intel Internal Foundry, etc.)?
I have a different view on the matter. It seems you regard Samsung logic as a failure, and if the metric is being a better foundry than TSMC, then your right. To me the ordering of your words matter (IDM+foundry). By this metric I think they are actually successful but have been in a rough patch for the past 2-3 years.

Now before you get your pitch forks out let me elaborate. Samsung electronics is first and foremost an appliance/consumer electronics manufacturer. With or without foundry Samsung would have a memory business and a logic business because it enables them to have pricing power over their other suppliers, lower manufacturing prices for their core products, and consistent guaranteed supply. If Samsung is going to have these fabs anyways (because totally relying on TSMC or Micron/SK would put them at a supply and pricing disadvantage) then it makes sense to monetize all of these investments that they were going to make anyways by selling off wafers to whomever will buy them at whatever price they will pay. After all, a full fab amortizes capex and R&D expenditures. Ignoring the totally unrealistic statements of "becoming the number 1 foundry", Samsung foundry and Samsung memory accomplish their goals of generating revenue on investments that Samsung electronics made to improve their core systems business. In some ways you could even argue that Samsung foundry is more successful now then it was in the late 2000s, because even though their technology is worse, they have more customers and volume than just Apple.

Under these same guiding principles what is to stop IFS from becoming another Samsung foundry? IF intel can not only achieve PPW leadership but also hold onto it, what is to stop IFS from turning into what we all thought Samsung might become when they blew the fabless ecosystem's socks off with 14LPP? If they have PPW leadership what is to stop them from at least having technology and cost parity with AMD in their core businesses? To be clear I personally don't see intel ever going back to 99% client and server CPU share, or intel displacing TSMC as the number 1 foundry. At least the way I see it, that isn't the point of IDM+foundry. The point is to provide the extra scale and profits to the core businesses.

I guess an interesting counter argument might be that intel’s core business is actually as a semiconductor manufacturer rather than a cpu vendor. Noyce Moore and Grove are all manufacturing and device guys, and the cpus only became the main product many years later. While FPGAs and data center accelerators may be considered as complementary products to sell more Xeons. I suppose you could turn the whole thing on its head and say intel was kind of a foundry before such a thing even existed so they grew to become an IDM that plays in all product categories (with a focus on x86 cpus because that was how you get the highest ASP per wafer). It is almost a mirror to how TSMC grew from embedded to mobile and then finally to fabing chips for all.

Taken to the Nth degree I wonder how Foundry+IDM would differ from IDM+foundry. My assumption is badly because if foundry and being a semiconductor manufacturer is your primary goal then pure play is the way to go. However if selling products or systems is your goal then I think IDM still has a place (obviously assuming said IDM has the scale to justify the investments they make).
 
Last edited:
When both PC and non-PC demand are down, having the foundry is not an advantage. The foundry also does not have enough process bandwidth (16/3/18A), yet there's more demand at mature nodes. Outsourcing some tiles is another bad sign.
 
Back
Top