Question Intel Q3: Ouch

[jpiniero]

https://d1io3yog0oux5.cloudfront.net/_c675d259e2193ee11667e182e2fab5ca/intel/news/2020-10-22_Intel_Reports_Third_Quarter_2020_Financial_1424.pdf

Desktop volume down 16%, Notebook up 16% but ASP down 7%.

Q4 forecast is bad too, they are expecting 14% decline in revenue compared to Q4 last year.

 

[Kuiva maa]

The troubles of intel are expressed in the market in a very clear way. They are running out of people to sell Skylake to. In the datacenter the best customers have already upgraded to Skylake and some have gone from Skylake to better Skylake SKUs even. In order to (at least partially) reverse this situation they don’t even need to beat AMD. As long as they migrate their entire product stack to something clearly better than the Skylake derivatives they still rely on, they will be able to tap into both their corporate relationships and the loyal individual customers that know and trust intel. Until then, they are in big, big trouble.

 

[DrMrLordX]

LOL. If Intel designed first to TSMC then tried to port back over to Intel in-house process, it is game over. Those fabs would be useless.

I'm suggesting Intel license TSMC's designs and "pay" TSMC by essentially retooling their fabs for TSMC processes - or a slightly-modified version (think of what GF did with Samsung's 14LPP). Intel's "in-house" processes would essentially cease, or be hybrids based predominantly on TSMC tech.

 

[jpiniero]

One weak quarter doesn't mean anything, intel had similarly weak 2nd and 3rd quarters in 2019 but yearly they still raked in the mullah.

Intel's own Q4 forecast is weak. If it ends up being accurate it'd be a similar decline as Q3.

 

[chrisjames61]

Delusional thinking and or denial that TheElf seems to specialize in is what has put Intel in this huge hole. He seems unable or unwilling to comprehend the urgency Intel is facing.

 

[maddie]

Yes AMD is paying for TSMCs fab operation and R&D and margin on top of everything else while intel doesn't.

This would make sense if intel wouldn't be making 4 (to 6) bil a quarter, clean money in the pocket, from these fabs.
You are talking like intel is losing money from the fabs but intel is making a lot of money.

My lord, you don't have a clue.

Have you seen the projected cost of Fabs as we go forward?
Do you realize the chip output needed to finance R&D into further node advancement?
Why do you think the trend for companies has been to become fabless?
What do you think is Intel's desire to expand into more markets? Hint: it's a way to have enough silicon market to justify owning your own fab tech.


Many have seen the future suggests only the largest with high utilization will be able to afford the expenses needed.
Companies paying a 'fee' to utilize a fab is the best way to keep your costs low and stay at the forefront of node advancement.

You seem to think paying TSMC is a bad thing when it's really a great thing, as so many others are also helping each other have leading node tech as a lower cost.
As I said in the first line, you don't have a clue to what is a strength and what is a weakness.

 

[dmens]

I'm suggesting Intel license TSMC's designs and "pay" TSMC by essentially retooling their fabs for TSMC processes - or a slightly-modified version (think of what GF did with Samsung's 14LPP). Intel's "in-house" processes would essentially cease, or be hybrids based predominantly on TSMC tech.

I don't think TSMC will ever allow that to happen. Even if they did, they would charge such an enormous fee that it would render Intel financially unviable.

 

[AnandThenMan]

Intel should hire Hector Ruiz the man knows a thing or two about going fabless.

I don't think TSMC will ever allow that to happen. Even if they did, they would charge such an enormous fee that it would render Intel financially unviable.

I have a feeling business politics are strongly in play, meaning TSMC would put Intel at the back of the line.

 

[moinmoin]

I don't think TSMC will ever allow that to happen. Even if they did, they would charge such an enormous fee that it would render Intel financially unviable.

Yeah, the only way I can see that happen is Intel actually selling its fabs to TSMC, at the price of the tools that are of use to TSMC. Also not realistic anytime soon.

 

[Qwertilot]

Well, if Intel's fabs do face falling behind enough that they're not market relevant then something pretty dramatic would have to happen.

How many (if any) new fabs would TSMC have to construct to cover the volume? I presume there'd be a serious time lag involved there too.

 

[SarahKerrigan]

Nothing personal, but I am really sick of hearing "Intel has this on their roadmap" and "in xxx timeframe they will have yyy product".

The only thing I have seen as new is a limited quantity of Rocket lake laptop chips that have mixed reviews. They are losing badly in server, HEDT, and in 2 weeks, desktop. And their laptops are fighting with the new Renior chips.

What Rocket Lake laptop chips are you referring to?

 

[Doug S]

I'm suggesting Intel license TSMC's designs and "pay" TSMC by essentially retooling their fabs for TSMC processes - or a slightly-modified version (think of what GF did with Samsung's 14LPP). Intel's "in-house" processes would essentially cease, or be hybrids based predominantly on TSMC tech.

I'm not sure why you think TSMC would be interested in such a deal. Imagine if the tables were turned, like they were only a few years ago. Does anyone seriously believe Intel would have been willing to license their process to anyone else?

 

[Zucker2k]

My memory is hazy on this but didn't Intel help TSMC with getting Finfets (or some other chip technology) right sometime around '07, '08 timeframe?

 

[DrMrLordX]

I don't think TSMC will ever allow that to happen. Even if they did, they would charge such an enormous fee that it would render Intel financially unviable.

Only way it happens is if -

Yeah, the only way I can see that happen is Intel actually selling its fabs to TSMC, at the price of the tools that are of use to TSMC. Also not realistic anytime soon.

Right. That's what I was suggesting. Honestly I think it's more realistic than TSMC building full-scale fabs in the US (as opposed to the mini-fab in AZ) or Intel just letting their foundry drag the entire company to the grave.

I'm not sure why you think TSMC would be interested in such a deal. Imagine if the tables were turned, like they were only a few years ago. Does anyone seriously believe Intel would have been willing to license their process to anyone else?

If Intel would have gotten control of fab time for an extended period of time, or forever, then yes. Same deal today with TSMC. Intel has a lot of fabs. If Intel retools those fabs for TSMC and lets TSMC take a significant portion of all the wafers that go through those fabs for whatever customer TSMC wants, then it's a major boon for TSMC. Free capacity! All for extending the privelege to Intel of selling a certain wafer allocation to Intel. Would you not take a deal like that?

It's the only way Intel gets advanced node tech for their designs in the (eventual) future. Since they aren't doing it themselves.

 

[VirtualLarry]

I'm suggesting Intel license TSMC's designs and "pay" TSMC by essentially retooling their fabs for TSMC processes - or a slightly-modified version (think of what GF did with Samsung's 14LPP). Intel's "in-house" processes would essentially cease, or be hybrids based predominantly on TSMC tech.

Wouldn't that make Intel a direct competitor for TSMC then? Couldn't Intel try to woo over TSMC's fab customers to Intel's "TSMC clone" fabs then, should they decide to whore themselves out to fabless customers?

 

[coercitiv]

Wouldn't that make Intel a direct competitor for TSMC then? Couldn't Intel try to woo over TSMC's fab customers to Intel's "TSMC clone" fabs then, should they decide to whore themselves out to fabless customers?

Not that I'm really entertaining this TSMC know-how transfer idea, but there are ways to do it while enforcing a no-compete clause from Intel.

 

[moinmoin]

The only way I can see any of that happening is full out sales of the foundries to TSMC, with Intel letting go all of its rights to it, possibly except a first take on booking the production capacity there as a client. No, hard to imagine current Intel being able to make such a decision anytime soon. It will have to go through the current process of self-discovery for another decade to get to such a state. At which point the foundries may well be not worth much anymore.

 

[DrMrLordX]

Not that I'm really entertaining this TSMC know-how transfer idea, but there are ways to do it while enforcing a no-compete clause from Intel.

Pretty much. GF and Samsung already paved the way for something like this to happen with the cross-licensing deal they cut years ago.

 

[clemsyn]

With all the issues with Intel and its foundries, what are the chances of NVIDIA buying Intel at a low PE? This will make them a direct competitor to AMD in most areas.

 

[Hitman928]

With all the issues with Intel and its foundries, what are the chances of NVIDIA buying Intel at a low PE? This will make them a direct competitor to AMD in most areas.

Intel's market cap is still ~$200B, so it's highly doubtful. Intel would essentially have to agree to a merger rather than a buyout but I just don't see that happening.

 

[ThatBuzzkiller]

Intel's market cap is still ~$200B, so it's highly doubtful. Intel would essentially have to agree to a merger rather than a buyout but I just don't see that happening.

I highly doubt much of Intel stock holders would currently even be interested in taking a stake in Nvidia since Intel has a ~5x higher net income and over a`~20x higher dividend yield compared to Nvidia. Many Intel investors are probably at an older age compared to Nvidia's investors so a high dividend yield is preferable when they're getting nearer to retirement age ...

Nvidia's valuation is currently positioned as a "growth stock" but how sustainable that growth is very uncertain since we have less than 5 years left to exploit advancements in process technology. We will literally see the tech bubble bursting before our very own eyes since many tech corporations probably won't survive the post-silicon world where transistor technology have stopped shrinking or seeing much improvement over very long periods of time ...

I'd argue that Intel's current valuation is just right while Nvida's valuation is very likely to be overvalued ...

 

[Kenmitch]

With all the issues with Intel and its foundries, what are the chances of NVIDIA buying Intel at a low PE? This will make them a direct competitor to AMD in most areas.

Blue Gorilla mating with the Green Goblin?

 

[dmens]

With all the issues with Intel and its foundries, what are the chances of NVIDIA buying Intel at a low PE? This will make them a direct competitor to AMD in most areas.

Zero. Jensen is too smart to spend money on dead weight.

 

[Doug S]

that growth is very uncertain since we have less than 5 years left to exploit advancements in process technology

People have been saying Moore's Law has only 5-10 years to run for most of my life. They are nowhere near atomic limits (headline process size e.g. "5nm" decoupled from actual physical widths long ago) so when Moore's Law does end, it will end for economic reasons not physical limits. Which means it will end for Intel before it ends for TSMC, just like it ended for AMD over a decade ago.

 

[LightningZ71]

Moores law dies in a slow whimper. As it gets more and more expensive to advance node processes, fewer and fewer companies can afford the investment required to do so. Eventually, it devolves to one company and any governments willing to foot the R&D bill via captive companies or agencies. Its getting harder and harder to make progress each generation.

 

[ThatBuzzkiller]

People have been saying Moore's Law has only 5-10 years to run for most of my life. They are nowhere near atomic limits (headline process size e.g. "5nm" decoupled from actual physical widths long ago) so when Moore's Law does end, it will end for economic reasons not physical limits. Which means it will end for Intel before it ends for TSMC, just like it ended for AMD over a decade ago.

The bolded doesn't make any sense. Physical limits are the absolutes we can reach in our universe and no amount of economic resources even if infinite in theory will push us past those limits ...

The so called 'limit' for CMOS transistors is actually bigger than just a single atom according to recent literature and even then we have to move off of silicon and use different materials to reach gate lengths as little as 1nm ...

This time Moore's Law really is coming to an end and the tech bubble is going to burst sooner than you think ... (there's no turning back and we're quickly running out of options with silicon)

Intel can't be saved unless the American government places sanctions on TSMC and Nvidia's days of high growth are coming to an end since they rely the most process technology advancements ... (GPUs are easier to scale with more transistors so once the future stops offering them a free lunch of process technology advancements they'll start starving the most)