WEBINAR: Can Intel Reclaim Its Crown in the Semiconductor World?

[count]

Like any manufacturing line - you lower pricing if your utilization starts dropping, or do other things to save money. If the margins are so low that 80% utilization breaks you then you should probably raise prices or just take your extra capital and put it in VTSAX. (i.e. if you can't beat the market, why bother staying in business).

AMD limited themselves critically twice by not having enough capacity (K7, K8).

This is a very naive idea. In practice, manufacturers do not want to reduce pricing except as a last resort. Typically, a 10-15% price drop won't create enough extra sales of offset the loss of income from the price decline, and price drops have a tendency to become permanent as customers start to use that as a new baseline for negotiations. Then you become a lower margin business relative to your competition which impairs your ability to reinvest in technology and you fall further behind the competition. Price drops lead to a vicious cycle.

Your point about not staying in business if you can't beat the market is 100% accurate... Intel needs to get out of the foundry business and if they had done so 5 years ago I bet they would be doing very well today, but instead they are in crisis.

 

[Xebec]

This is a very naive idea. In practice, manufacturers do not want to reduce pricing except as a last resort. Typically, a 10-15% price drop won't create enough extra sales of offset the loss of income from the price decline, and price drops have a tendency to become permanent as customers start to use that as a new baseline for negotiations. Then you become a lower margin business relative to your competition which impairs your ability to reinvest in technology and you fall further behind the competition. Price drops lead to a vicious cycle.

Your point about not staying in business if you can't beat the market is 100% accurate... Intel needs to get out of the foundry business and if they had done so 5 years ago I bet they would be doing very well today, but instead they are in crisis.

Sorry my initial reply was a bit hasty - (I'm having a rough day)

I understand the risk of that cycle; but I would think there are a few options to get out of it -- sunset a node, raise the price annually, etc. Any TSMC competitor on leading edge right now would need to offer sweetheart deals - like Samsung did with 8nm to Nvidia. Also when customers that typically 'need' newer nodes move on you could re-advertise at a better price (NDAs) as other 'lagging' customers start to go to you as you've proven your ability to deliver.

I do wonder if Intel hadn't fumbled on 10nm, what their balance sheets would look like today. New nodes are becoming less impactful over time. Could Intel's balance sheet have remained strong enough until a "full node" is defined as a 5% improvement in 10 years or so?

 

[tooLongInEDA]

This is a very naive idea. In practice, manufacturers do not want to reduce pricing except as a last resort. Typically, a 10-15% price drop won't create enough extra sales of offset the loss of income from the price decline, and price drops have a tendency to become permanent as customers start to use that as a new baseline for negotiations. Then you become a lower margin business relative to your competition which impairs your ability to reinvest in technology and you fall further behind the competition. Price drops lead to a vicious cycle.

Your point about not staying in business if you can't beat the market is 100% accurate... Intel needs to get out of the foundry business and if they had done so 5 years ago I bet they would be doing very well today, but instead they are in crisis.

It would indeed be very interesting to see a back of the envelope (we'll never get this breakdown directly from Intel) ROI calculation on Intel's technology and fab investments since they announced the original Intel 10 (now called 7 IIRC) around 10 years ago. We could then compare that with preceding generations (up to 14 and its many + variants). And model what they might have done had they switched to TSMC 10 years ago (assuming this could have been done without major technical disruption and delays). I suspect this analysis would strongly back up your point and that we're looking at potential lost profits into the $10bns over this period.

 

[siliconbruh999]

If this is the "not TSMC" market, than whoever is "not TSMC" is guaranteed to go bankrupt. Not because TMSC never lacks capacity, but because by definition if you always have extra capacity, that means your business model requires you to run underutilized, and an underutilized fab is a money losing venture.

That is true for a pure play foundry but for IDM they have their products they should try to fill market with cheap products from their fab that are competitive it fills the fab and increases revenue.

 

[siliconbruh999]

It would indeed be very interesting to see a back of the envelope (we'll never get this breakdown directly from Intel) ROI calculation on Intel's technology and fab investments since they announced the original Intel 10 (now called 7 IIRC) around 10 years ago. We could then compare that with preceding generations (up to 14 and its many + variants). And model what they might have done had they switched to TSMC 10 years ago (assuming this could have been done without major technical disruption and delays). I suspect this analysis would strongly back up your point and that we're looking at potential lost profits into the $10bns over this period.

That is the Management's fault for not running the fab and design efficiently during the Otleni Kranzich era they used to tape out frequently run hot lots and what not.

 

[count]

That is true for a pure play foundry but for IDM they have their products they should try to fill market with cheap products from their fab that are competitive it fills the fab and increases revenue.

Intel does not have enough volume on it's own to fill a leading edge fab. That's the fundamental problem. That's been the problem since mobile, because leading edge volume moved to mobile chips instead of PC CPUs, and Intel didn't have competitive mobile products.

So what did Intel do?

-First they tried to build their own mobile chips, but were not competitive on cost or power efficiency.
-Next they tried their Foundry 1.0 strategy to try to bring in external volumes, but even though at the time they still had leading edge manufacturing, they were expensive and did not have a mature PDK.
-Because they couldn't attract external volumes, they went on a shopping spree, buying all kinds of chip companies like Altera, Nervana, MobileEye... but Intel's culture didn't translate to those companies and many of them resisted porting over to Intel's manufacturing.
-Then they started talking about Foundry 2.0, but by this point Intel had fallen behind on Manufacturing, and still did not have a competitive PDK
-Pat G came on board and doubled down on Foundry 2.0 going on a mad spending spree, tens of billions of dollars, to try to catch up on Manufacturing, and I think some good progress was made here, but at the cost of Intel's balance sheet. And just because you build it doesn't mean they will come.
-Now Intel has an impaired balance sheet, a bunch of half finished fabs, and a process that looks good on paper but hasn't stood up to the test of high volume manufacturing... and not much else to show for it. At best they become a marginal second source to TSMC.

I think the foundry 2.0 model is busted, and this is a path to bankruptcy. Intel could be a very profitable fabless company but egos and hubris about regaining "leadership" are pushing the company off a cliff.

 

[siliconbruh999]

I will be blunt there is nothing leadership about their design as of rn the fab has carried the design for decades and the moment Intel lost it everything became Haywire.
They need improvements to design as well it's not about ego there is no need to spun out one of your core asset. In AMDs case it made sense cause the fabs were never that good but their designs were better.

They have very good software though which AMD lacks.

Also on enough volume yeah they don't otherwise they won't be capacity constrained on Intel 7 .

 

[Xebec]

I will be blunt there is nothing leadership about their design as of rn the fab has carried the design for decades and the moment Intel lost it everything became Haywire.

100% agree here.

The death spiral of Intel isn't about fabs vs. design. It's about overall engineering execution.

Intel could have done IDM for longer if they kept executing well, even with the mobile failure. Now there's a chance that neither will be successful in the long term.

 

[count]

I will be blunt there is nothing leadership about their design as of rn the fab has carried the design for decades and the moment Intel lost it everything became Haywire.
They need improvements to design as well it's not about ego there is no need to spun out one of your core asset. In AMDs case it made sense cause the fabs were never that good but their designs were better.

They have very good software though which AMD lacks.

Also on enough volume yeah they don't otherwise they won't be capacity constrained on Intel 7 .

I'd argue that Intel lost it's manufacturing lead as a result of mobile, because that gave TSMC the volumes needed to invest and surpass Intel. TSMC wouldn't be where it is today if it hadn't been for the mobile market.

I'd also argue that constraints on Intel 7 is because they wanted to shift capacity to Intel 4 and Intel 3, but those nodes are too high cost... so customers wanted to go back to Intel 7. Cost of Intel 4 and Intel 3 would have been lower if Intel had comparable economies of scale but they don't.

 

[siliconbruh999]

I'd argue that Intel lost it's manufacturing lead as a result of mobile, because that gave TSMC the volumes needed to invest and surpass Intel. TSMC wouldn't be where it is today if it hadn't been for the mobile market.

I think it more to do with messing up 10nm was the reason not volume at the time

I'd also argue that constraints on Intel 7 is because they wanted to shift capacity to Intel 4 and Intel 3, but those nodes are too high cost... so customers wanted to go back to Intel 7. Cost of Intel 4 and Intel 3 would have been lower if Intel had comparable economies of scale but they don't.

You can consider this as a bad planning on part of swan he didn't create enough internal products In roadmap even now only all Intel 3/4 product is GNR/SRF

 

[Xebec]

I'd argue that Intel lost it's manufacturing lead as a result of mobile, because that gave TSMC the volumes needed to invest and surpass Intel. TSMC wouldn't be where it is today if it hadn't been for the mobile market.

I'd also argue that constraints on Intel 7 is because they wanted to shift capacity to Intel 4 and Intel 3, but those nodes are too high cost... so customers wanted to go back to Intel 7. Cost of Intel 4 and Intel 3 would have been lower if Intel had comparable economies of scale but they don't.

Intel chart from 2010/2011 showing revenue vs. fab costs, right before Mobile really took off for TSMC:
(Intel thought they had "last man standing" in the bag...)

 

[count]

Intel chart from 2010/2011 showing revenue vs. fab costs, right before Mobile really took off for TSMC:
(Intel thought they had "last man standing" in the bag...)

View attachment 3124

This perfectly illustrates the point I've been trying to make.

 

[XtorTourist]

I think it more to do with messing up 10nm was the reason not volume at the time

You can consider this as a bad planning on part of swan he didn't create enough internal products In roadmap even now only all Intel 3/4 product is GNR/SRF

Meteor Lake is Intel 3/4.

Swan gets credit for setting up the TSMC N3 backup plan which is now making Arrow Lake and Lunar Lake compute tiles. Without it, Intel would be in even worse shape than it is now. Note that every product on Intel 3/4 has had or is having troubles with getting to volume compared to Intel expectations.

 

[Daniel Nenni]

Meteor Lake is Intel 3/4.

Swan gets credit for setting up the TSMC N3 backup plan which is now making Arrow Lake and Lunar Lake compute tiles. Without it, Intel would be in even worse shape than it is now. Note that every product on Intel 3/4 has had or is having troubles with getting to volume compared to Intel expectations.

Swan did a very big and lucrative deal for TSMC. It really was a milestone for the semiconductor industry. Intel is now designing to TSMC N2 and will continue to work with TSMC in my opinion. That does not mean, however, that Intel should be fabless. I really have faith in Lip-Bu and feel Intel will be a viable semiconductor manufacturer again in the not to distant future.

 

[siliconbruh999]

Meteor Lake is Intel 3/4.

Swan gets credit for setting up the TSMC N3 backup plan which is now making Arrow Lake and Lunar Lake compute tiles. Without it, Intel would be in even worse shape than it is now. Note that every product on Intel 3/4 has had or is having troubles with getting to volume compared to Intel expectations.

He along with Kranzich was the reason Intel are in such a situation how can you call this a good thing if you shot yourself in the foot and go to your competitor for help not to mention the node they prepayed for and didn't even utilized to it's full capacity.

They don't have anything to sell on Intel 3 cause they never designed anything but few products that is changing though.
They are going to use Intel 4/3 in lots of support tiles going forward in Panther Lake/Diamond Rapids/Clearwater forest/Granite Rapid D including alongside GNR/SRF design to products at Intel take 3-4 years so if they planned new design in 2022 it will take 2025/26 for them to show up.

 

[fansink]

Swan did a very big and lucrative deal for TSMC. It really was a milestone for the semiconductor industry. Intel is now designing to TSMC N2 and will continue to work with TSMC in my opinion. That does not mean, however, that Intel should be fabless. I really have faith in Lip-Bu and feel Intel will be a viable semiconductor manufacturer again in the not to distant future.

In 2011, Intel had 4 times the revenue of TSMC.

In 2025, TSMC will have more than 2 times the revenue of Intel.

In 2025, Intel will have less revenue than they had in 2011.

In 2025, TSMC will have 8 times their 2011 revenue.

 

[Daniel Nenni]

He along with Kranzich was the reason Intel are in such a situation how can you call this a good thing if you shot yourself in the foot and go to your competitor for help not to mention the node they prepayed for and didn't even utilized to it's full capacity.

I agree with you on BK, he was the all time worst Intel CEO. Swan should not have been CEO, I blame the Intel Board for that one. Swan was between a rock (Intel Design) and a hard place (Intel manufacturing) and no match for either one. I think the TSMC deal was the only way for him to cross that chasm and it really was a brute force move. Giving Intel Design a choice of fabs really motivated Intel Manufacturing. Looking back and looking forward I think it was the right move, absolutely.

 

[Xebec]

I agree with you on BK, he was the all time worst Intel CEO. Swan should not have been CEO, I blame the Intel Board for that one. Swan was between a rock (Intel Design) and a hard place (Intel manufacturing) and no match for either one. I think the TSMC deal was the only way for him to cross that chasm and it really was a brute force move. Giving Intel Design a choice of fabs really motivated Intel Manufacturing. Looking back and looking forward I think it was the right move, absolutely.

I think Otellini is a formidable challenger to BK's title for worst Intel CEO. His was fairly active in damaging Intel's future:
- Had Mobility in his hands, gave it away
- Cancelled any products that could have been used for AI later (Larrabee, other Graphics efforts)
- Continued the Intel arrogance
- Kept money flowing into Itanium for half his tenure
- Did mass Engineering / talent layoffs in 2007

His big win seems to have been the Apple Mac business.. for a few years.

BK by comparison just showed up to work and didn't seem to do very much. Though he did tweet about jogging/running at times, so that's good I guess. (BK certainly lied many times about process and products).

Swan did put Intel in a really hard spot - by reducing risk on fabs and buying even more TSMC capacity, he pulled away enough products (revenue) from Intel's future fabs to *force* them to either go Foundry as a Service or fabless.