News Intel 3Q21 Results

[Hitman928]

Intel EPS beats by $0.60, misses on nnon-gaap revenue

Intel (INTC): Q3 Non-GAAP EPS of $1.71 beats by $0.60; GAAP EPS of $1.67 beats by $0.61.Non-GAAP Revenue of $18.1B (-1.1% Y/Y) misses by $170M.Shares -3.6%.Press Release Achieved...

seekingalpha.com

• EPS beat by $0.60
• Revenue very slight miss by $170M
• Guidance: Raising full-year 2021 EPS and gross margin guidance. Now expecting GAAP EPS of $4.50 and non-GAAP EPS of $5.28 from prior guidance of $4.80 vs. $4.11 consensus. and GAAP gross margin of 55% and non-GAAP gross margin of 57%.
• Q4 Guidance: Revenue of $18.3B vs. $18.26B consensus, EPS of $0.90 vs. $0.94 consensus.
• Q4 gross margin guidance weak at 51.4% compared to 56% Q3.

Edit:
Revenues

	3Q21​	Relative 2Q21​	Relative 3Q20
CCG​	$9.7B	96%	98%
DCG​	$6.5B	100%	110%
IOTG​	$1.0B	101.6%	154%
Mobileye​	$326M	99.7%	139%​

 

[prtskg]

They've increased R&D, that's good. In 3-4 days we will be able to do comparison with AMD. Then we will know what AMD has gained and what was the actual effect of supply constraints.

 

[jpiniero]

3) The chip sizes are far, far smaller now going with tiles instead of a giant monolith.

I assume the CPU tile is smaller than that, but Cannonlake was pretty small - 70 mm2. And yields then were so bad they couldn't even get a minimally viable product with some portion of the IGP working despite being tiny.

Also the spending bill is in some trouble. If Intel doesn't get the money, there's going to be real pressure on the exec team to actually deliver the nodes in a reasonable timeframe.

 

[gdansk]

Capital investments in manufacturing process is the correct choice. And I think it is likely to pay off. Some customers will want to diversify their fabrication facilities after this latest fiasco.

But I guess the market thinks they won't pay off or it'll be too late?

 

[moinmoin]

Also the spending bill is in some trouble. If Intel doesn't get the money, there's going to be real pressure on the exec team to actually deliver the nodes in a reasonable timeframe.

But I guess the market thinks they won't pay off or it'll be too late?

The danger for Intel is that it will be too late. Intel basked in record income by milking the 14nm node dry. That would have been the correct time to do the capital investments in manufacturing process. Instead they did... stock buybacks. Now that 14nm is a thing of the past and investment into the future is actually being done of course the real pressure is on them to actually execute for a change.

 

[Saylick]

Yeah, the unfortunate part of the industry is that cutting edge nodes are becoming more and more expensive to develop, and the best way to recoup that investment is to be first to market. I am not sure what fraction of the cutting edge node market is required to break even on R&D costs, but it seems like that % increases with every generation. To use some arbitrary numbers, let's say you only needed to capture 20% of the cutting edge node market to recoup the cost of developing 7nm, but that number may be like 40% to recoup the cost of developing 5nm. The real numbers may be even higher. In theory, if the revenue/demand for cutting edge nodes increases at the same rate as the cost to develop the node, this % would stay flat generation over generation, but I doubt that's the case since not every chip needs to be on the cutting edge node. More and more effort is now spent on only allocating the cutting edge node to the parts of the chip that need it the most. The point I guess is that there may not be enough revenue to support more than one or two players in the cutting edge node market. Global Foundries dropped out for this reason, as did many fabs before them. Intel and Samsung maybe chasing a reality that grows dimmer and dimmer the slower they are to reaching parity with TSMC.

 

[TheELF]

Intel and Samsung maybe chasing a reality that grows dimmer and dimmer the slower they are to reaching parity with TSMC.

Intel and TSMC are selling to completely different markets* , they are not direct competitors so it doesn't matter which one is ahead or not.
*except for the percentage of CPUs they make for AMD

Intel doesn't have any competition except for AMD and we have seen for the last few years that even while "loosing" to AMD, even with every server company switching to ARM, even with losing apple as a client, they are still only making more and more money.
In fact I'm very surprised that they are releasing rocket lake since their revenue is so stupid high right now selling previous gens.

 

[Saylick]

Intel and TSMC are selling to completely different markets* , they are not direct competitors so it doesn't matter which one is ahead or not.
*except for the percentage of CPUs they make for AMD

Intel doesn't have any competition except for AMD and we have seen for the last few years that even while "loosing" to AMD, even with every server company switching to ARM, even with losing apple as a client, they are still only making more and more money.
In fact I'm very surprised that they are releasing rocket lake since their revenue is so stupid high right now selling previous gens.

Sorry, but if it didn't matter which one is ahead, Intel would not be in it's current predicament. Honestly, if the node didn't matter and Intel could miraculously make more and more money sticking to their super deprecated 14nm node, they would do so. I mean, why blow money trying to move to a more advanced node if it didn't matter, right? But alas it doesn't work that way because the rest of the market moves forward irrespective of what Intel does, and if Intel decides to stay idle, they bleed revenue.

Yes, technically speaking Intel and TSMC are not direct competitors because Intel designs and sells chips while TSMC just makes chips, but the moment TSMC makes chips for one of Intel's competitors, it matters where Intel stands node-wise to TSMC. If anything, Intel is quite lucky that TSMC can't make more chips for AMD to sell.

 

[DrMrLordX]

TSMC's advancements (and Samsung's advancements) also make it harder for Intel to pursue Foundry 2.0 . As it stands, it's not realistic to think that Intel will be able to offer significant volume to customers on any of its processes more-advanced than 10ESF/Intel 7.

 

[TheELF]

Sorry, but if it didn't matter which one is ahead, Intel would not be in it's current predicament. Honestly, if the node didn't matter and Intel could miraculously make more and more money sticking to their super deprecated 14nm node, they would do so. I mean, why blow money trying to move to a more advanced node if it didn't matter, right? But alas it doesn't work that way because the rest of the market moves forward irrespective of what Intel does, and if Intel decides to stay idle, they bleed revenue.

Yes, technically speaking Intel and TSMC are not direct competitors because Intel designs and sells chips while TSMC just makes chips, but the moment TSMC makes chips for one of Intel's competitors, it matters where Intel stands node-wise to TSMC. If anything, Intel is quite lucky that TSMC can't make more chips for AMD to sell.

The predicament of making 77bil revenue? When a few years ago they only made half of that? I wish I had that sort of predicament.

The only rest of the market is AMD, these are the only two that make x86 (and via but they make almost nothing) , and as you yourself said "Intel is quite lucky that TSMC can't make more chips for AMD to sell." but also AMD doesn't have enough money to increase that number. AMD went up to 20% market share in consumer without changing anything for intel.

 

[NTMBK]

The predicament of making 77bil revenue? When a few years ago they only made half of that? I wish I had that sort of predicament.

The only rest of the market is AMD, these are the only two that make x86 (and via but they make almost nothing) , and as you yourself said "Intel is quite lucky that TSMC can't make more chips for AMD to sell." but also AMD doesn't have enough money to increase that number. AMD went up to 20% market share in consumer without changing anything for intel.

Intel is moving into foundry- they are explicitly trying to compete with TSMC and Samsung.

They need the increased Total Addressable Market from being able to manufacture phone SoCs, ARM chips for servers, GPUs, etc. As pointed out above- every single node gets more expensive to develop. x86 market isn't going to be sufficient to pay for future development at the leading edge.

 

[BorisTheBlade82]

Bringing Geek Back: Q&A with Intel CEO Pat Gelsinger

www.anandtech.com

Charlie Demerjian, SemiAccurate: One of the key elements in your messaging has been a commitment to being more open and more transparent with the industry, with hardware, software, and all the extra details. This is good, as Intel has pretty much stopped talking about the technology these past few years, sliding in the wrong direction in my view - things like transistor densities or die sizes on products that are launching the next day haven’t been shared. I’ve been requesting Intel to be more open, for years, only to not get any real satisfactory explanation. If you are committed to being more transparent, could you look into providing these numbers [and numbers like these] in the future?

PG:

Thanks Charlie - we're anxious to see you in person as well, sometime soon! So as we're swinging the doors open wide, we are getting much more engaged with the technology, and with technologists [like yourself]. I think we're going to be doing a lot more in this regard.

These particular comments, I'll say I haven't heard them before this. Feel free to send me an email with your top five that you'd like to see from us, and we'll have robust conversations. I won't promise anything today, because I don't know why people have been hesitant in those regards. But we do want to be engaged with technologists, with developers, and with our tech analyst community. Send me some thoughts directly, and I'll happily follow up on them.

Did hell just freeze? When have Charlie and a current Intel CEO ever exchanged words in such a nice and kind manner?
But on a serious note: When Pat was announced Charlie already showed his deepest respect for him. I guess he is also convinced that if anyone can turn Intel around it is Gelsinger.

 

[naukkis]

Anobody has any thoughts about that from same interview?

"Secondly, also with our IDM 2.0 strategy, we said we're going to consistently use external foundries. With that, maybe a quarter of my capacity is from my external foundry partners."

So Gelsinger exepts one quarter of their products coming from external foundies. That's huge, isn't it?

 

[TheELF]

Intel is moving into foundry- they are explicitly trying to compete with TSMC and Samsung.

TSMC and Samsung cannot offer x86 to anybody, that's intel only that can do that.

x86 market isn't going to be sufficient to pay for future development at the leading edge.

If intel drags out each node for 5-6 years like they did with 14nm it will pay for new node development just fine.

 

[DrMrLordX]

TSMC and Samsung cannot offer x86 to anybody, that's intel only that can do that.

Intel isn't offering x86 to anyone (per se). That's not what a foundry service does.

 

[BorisTheBlade82]

Intel isn't offering x86 to anyone (per se). That's not what a foundry service does.

Maybe you should read the interview first.

 

[DrMrLordX]

Maybe you should read the interview first.

x86 brings with it licensing costs. They will fab non-x86 for anyone, including people who might be considering Samsung and TSMC. Intel isn't going to offer provisional x86 licenses "for free".

 

[BorisTheBlade82]

No one said they would do it for free. One can argue that it is basically the same as having a semi custom design like the console SoCs from AMD. But from what I get it is as if a company can take x86 cores and do what they want with them. That is a new vector indeed and something truly different to TSMC and Samsung.

 

[DrMrLordX]

No one said they would do it for free. One can argue that it is basically the same as having a semi custom design like the console SoCs from AMD. But from what I get it is as if a company can take x86 cores and do what they want with them. That is a new vector indeed and something truly different to TSMC and Samsung.

They can "do what they want with them" so long as:

a). Intel agrees to the license/doesn't feel threatened by the design and
b). Intel gets their cut

TSMC and Samsung can just as easily fab x86 CPUs for anyone that has a license through other means. Yes, that is limited to AMD and VIA/Xhaoxin but you get the point. It isn't really clear who would want to design their own x86 chips and be locked in to one foundry that also requires your product to be non-competitive with theirs. If Intel expects to move any wafers they're mostly going to be dealing with ARM, RISC-V, or bespoke ISAs.

 

[BorisTheBlade82]

Pat stated that there will be no lag between Intel employing new x86 IP and their contractors being able to use it. So for me this sounds that they are much more open now than ever before. And I can imagine the hyper scalers to come up with big ML/AI designs that combine custom logic with x86 for basic workloads.

 

[NTMBK]

TSMC and Samsung cannot offer x86 to anybody, that's intel only that can do that.

If intel drags out each node for 5-6 years like they did with 14nm it will pay for new node development just fine.

x86 may not be a "big deal" for much longer. Just look at how well Apple can run x86 on ARM under emulation. When Qualcomm bring out their laptop SoCs using the custom CPUs they got from Nuvia, I think we're going to start seeing the same on Windows too. ARM chips that can run x86 binaries almost as fast as an Intel chip, or can outperform them in native code.

Dragging out each node for 5-6 years is tantamount to giving up at cutting edge tech. TSMC and Samsung will not stand still. And Intel have announced that they aren't doing that- they have an aggressive roadmap, and aim to be back in the lead in about 5 years' time.

 

[DrMrLordX]

And I can imagine the hyper scalers to come up with big ML/AI designs that combine custom logic with x86 for basic workloads.

. . . why?

Intel already produces custom Xeons for big customers. They've been doing it for years. Nothing has changed there. But that is done with the understanding that the customer wants a Xeon first and foremost, and that they need it to have a few extra instructions to fit their particular workload.

If you're looking for a 100% custom piece of silicon, why would you choose x86 at all?

Otherwise, if someone comes up with an x86 desktop CPU they can fab on 10ESF/Intel 7 that runs circles around Alder Lake or Sapphire Rapids, you can rest assured that Intel will not allow a foundry customer to undercut their core business.

x86 may not be a "big deal" for much longer.

x86 will have adherents so long as the leading-edge designs keep it relevant.

 

[BorisTheBlade82]

. . . why?

I chose the word "imagine" for a reason. I can give you no specific answer because I am in no way an expert in that field. But in such a big and complex landscape I can imagine that there are tasks were one might choose custom x86 over custom ARM.
Ironically this increased ISA competition might help nVidia to get the ARM deal through. Maybe Pat and Jensen struck a deal

 

[DrMrLordX]

The only thing that x86 really has going for it right now is momentum. It's still big in the cloud, enterprise, various workstations, etc. Some very good implementations keep it in its current position, along with a large library of software. The times are-a-changin, though, and ARM is showing that it is even more of a Swiss army knife ISA than x86. ARM can be used in anything from cell phones to servers; in fact, the same core design (with tweaks made to branch prediction) can be used in a cell phone or a server! Look at all the places A76 cores find themselves thesedays.

As an ISA, I don't see a whole lot of reason for people to specifically choose x86 unless they're trying to cater to an existing installed base. Maybe someone else here can give a technical reason as to why a small firm looking for custom/semi-custom would pick a clean sheet x86 design over ARM?

 

[NTMBK]

Maybe someone else here can give a technical reason as to why a small firm looking for custom/semi-custom would pick a clean sheet x86 design over ARM?

Potentially per-chip costs? An AMD x86 semicustom chip doesn't need to pay any per-device license revenues for the ISA, or for the CPU IP block. That could help improve costs when margins are tight (e.g. for games consoles).

 

[DrMrLordX]

Potentially per-chip costs? An AMD x86 semicustom chip doesn't need to pay any per-device license revenues for the ISA, or for the CPU IP block. That could help improve costs when margins are tight (e.g. for games consoles).

Maybe. That might be why AMD finds itself in consoles, while the only major ARM install is of old-chips from a vendor that has to pay the ARM licensing fees itself (instead of Nintendo footing the bill).

One wonders if Intel will try to charge foundry customers a per-chip licensing fee for x86 or just charge a flat rate for the license?