The Intel Execution in [2024]

Are FPGAs profitable these days? AMD's Xilinix has been collapsing for a while, and so is Intel’s Altera, FPGAS are nowhere to be seen in the current AI boom, it does seem that GPUs have sidestepped them.
???? The market isn't experiencing explosive growth but the TAM is expected to double by 2030, nice and consistent growth.
FPGAs are also nicely diversified across a number of industries; datacenter, consumer electronics, industrial, automotive, defense/aerospace, and telecom.

Things have definitely slowed down the past year or so due to boom-to-bust cycle of the "chip shortage" and the general market/inventory correction that follows but they are still extremely profitable and have decent growth potential, mainly with automotive and industrial.

Edit- Also FPGAs are everywhere in the AI boom, just behind the scenes and overshadowed by AI-centric chips. It also doesn't really show since the numbers are down in the other segments.
 
FPGAs are also used by chip designers routinely for simulating/testing parts of their chips.

They will never really go away unless something truly drastic changes: FPGAs will always have their place as middle-of-the-road between general purpose CPUs and ASICs. So in essence they will always be "profitable", just revenues or volumes might change
 
Why Samsung? Not like Samsung foundry has been doing the best of late but they have their own fabs and leading edge GAA tech (Not sure how it compares to Intel's technically aside from the obvious missing backside power supply), and existing customers including the likes of Nvidia, Qualcomm, etc, and presumably making a profit (Including memory). Why would they want to take on a negative margin foundry when they can barely make their own foundry business work? And the less said about Rapidus the better, it's a vapourware project and I don't see it ever going anywhere.

Yep, that's the key issue. When you have to commit to fab processes years in advance with products worth Billions of dollars at stake, who would you rather bet on? A plan/potential, or a company which has been executing better than anyone else for the past decade or more? Most potential customers seem to have chosen TSMC.

Will they be able to make enough money to fund everything through 2026-2027 by the time that happens though? That's what seems unlikely to me as well.

Apparently Broadcom is not happy with 18A - https://www.reuters.com/technology/...broadcom-tests-disappoint-sources-2024-09-04/

And Trendforce reporting that TSMC/Samsung unlikely to be buyers for Intel foundry - https://www.trendforce.com/news/202...ers-for-intels-rumored-foundry-business-sale/
 

One of the benefits of our early success on Intel 18A is that it enables us to shift engineering resources from Intel 20A earlier than expected as we near completion of our five-nodes-in-four-years plan. With this decision, the Arrow Lake processor family will be built primarily using external partners and packaged by Intel Foundry.

Interesting development.
 
Nothing new, it's similar to Meteor Lake which had the CPU tile on Intel 4 but the Graphics, SOC and IO tiles manufactured by TSMC. Likely to be similar for Arrow Lake, with the CPU tile on Intel 20A instead.
I think the strong implication is that won’t be the case for a significant percentage of products, i.e. they’ll be using N3B for the CPU tile of at least some SKUs.
 
I think the strong implication is that won’t be the case for a significant percentage of products, i.e. they’ll be using N3B for the CPU tile of at least some SKUs.
True that is quite possible, I think we should find out soon given the rumoured October launch.
 



Interesting development.
Arrowlake has long been rumored to be N3B for compute except for 6+8 die -S desktop chips.

Also there's nothing fantastic about 18A. They claimed 20A is 15% perf over Intel 3 and 18A is 10% on top of that. Latest slides show 18A is just 15% on top of Intel 3. I think what happened is they couldn't reach perf goals(maybe even density, considering it's a disappointing 30% over Intel 3, which is just 10% over Intel 4, if you chose to use HD cells) for 18A, so they added extra libraries to make 20A to work with more configs, and renamed it "18A".

Another news is saying Qualcomm is looking to buy Intel. Intel is well on it's way to complete destruction.
 



Interesting development.

So a number of outlets are actually reporting this as an outright cancellation of 20A - https://www.tomshardware.com/pc-com...-goes-with-external-nodes-instead-likely-tsmc

So either this is inaccurate since if any Arrow Lake tiles were planned on 20A, they would have entered mass production months back, or the entire lineup is on TSMC N3B with no 20A CPU tiles.
 
It's lucky that all this drama is occuring at the time of Lunar Lake's release, which looks to be the first exciting Intel product in a long time IMO (hopefully independent reviews will prove it's as good as Intel claims). It counteracts (at least slightly) the narrative that Intel has nothing of value left; but it does seem to play into the hands of Intel Foundry being the weak side of the business... (in reality though, Intel's design side has been very weak in the last few years as well, and it is 100% responsible for its failure to penetrate AI as Gaudi/Nervana have only ever been manufactured at TSMC even for packaging afaik)

I've been reading the Semi Wiki forum a bit where Daniel Nenni claims that the problem is the PDK and vendors who evaluated 18A have nearly unanimously chosen TSMC's N2 (Daniel is the site's owner and very well informed, although he's been a bit of a TSMC fanboy for a long time, but I guess for logical reasons and history has proven him right - still, he does have a bit of a bias in my opinion): https://semiwiki.com/forum/index.ph...oadcom-18a-tests-disappoint.20913/#post-73823

Looking through other Intel disclosures people mentioned on that forum, this Intel slide showing 18A at parity (and Intel 3 being behind on everything including wafer cost) with 14A now being the next miracle (just one more node... always one more node...) is not very promising about 18A being competitive, also see Pat's statements that the bulk of their wafers in 2025 are Intel 7 and Intel 10, with Intel 18A only being a small amount: https://www.tomshardware.com/tech-i...a-based-cpus-will-ramp-to-high-volume-in-2026

And even much more worrying, the timeframes for significant 3rd party revenue for 18A foundry customers is further out than I thought according to their CFO with it being little in 2026 and "meaningful" in 2027 (which could mean not much at all in the grand scheme of things): https://semiwiki.com/forum/index.ph...ee-meaningful-revenue-in-2027-cfo-says.20915/

Putting it all together, I am pretty sure Intel was betting on becoming competitive(-ish) with TSMC's wafer costs by achieving similar economies of scale for 1 or 2 megafabs. They simply do not have either the internal volume or fully committed 3rd party customers to achieve those economies of scale right now with their current internal & external wafer volume forecasts. That means that even *if* 18A is competitive in principle, which is far from certain (it looks likely to be better than N3P but significantly worse than N2P in my opinion - which would be fine, *if* their time to market was what they were hoping for a few years ago), it is likely to be uncompetitive in practice in terms of the pricing they can offer to customers without bankrupting themselves.

---

I'm honestly not sure what I'd do in Pat's/the BoD's shoes. Fundamentally I don't think there is space in the industry for 3 foundries at the leading edge unless the revenue distribution was a lot more equal than it is today (which it realistically will never be unless TSMC screws up massively). You just can't be competitive with a 1.4nm-class process at 5% of TSMC's leading-edge wafer volumes (unless your process was massively better and had huge price power which Intel clearly does not have); it just doesn't work that way.

Splitting off the foundry business doesn't fix any of the foundry's issues whatsoever; any investor looking at the business plan for foundry right now would likely refuse to commit the 10s of billions required without some firmer commitments from 3rd party customers. You can't split off a business that would fail on its own; you can only shut it down, and needless to say that's an really hard position to justify both internally and geopolitically. So it feels to me like Intel is stuck.

The only 2 options that really solve Intel's problems in my opinion are:
1) Some kind of deal with Samsung, e.g. a Joint Venture merging their foundry business as a new independent company. Unlikely due to internal politics/pride on both sides and the amount of financing required, but not impossible in theory.
2) Splitting off the foundry as an independent company that is still majority owned by Intel (for now) but with large equity investments (and/or wafer purchase guarantees) from multiple interested customers, e.g. Broadcom/Qualcomm/MediaTek/Microsoft/etc.
3) Being very lucky with both TSMC and Samsung underdelivering at the same time as Intel actually delivers. Pat's religious tweets sometimes make me think that his entire strategy is literally just praying to God for this scenario to happen...

They all seem unlikely in the short-term. Intel's shares are in an interesting position right now with huge upside potential but also the risk they will become effectively worthless and acquired for scraps. I'm not confident enough to say which way it will go yet.
 
Considering that we're at a point where future silicon scaling is dubious and focus is moving onto advanced packaging and the demand for silicon production likely won't go down any time soon Intel's foundry business is in a weird spot where on the one hand it's a massive money sink at the moment which seem to be incapable of making even Intel's own chips to the required specs; but on the other hand all production lines will likely even out at the physical scaling limit over the next 10-20 years anyway and those which are struggling right now will probably get up there eventually too so there's not a lot of reasons to just cut that off at this point.

Maybe the idea to just use TSMC for performance/power critical products for now can in fact be a correct one? Would allow them to sell good products while their foundry would just catch up slowly arriving to a point where it would be just as good as any other option eventually. They will still need to cut off some fat and downsize quite a bit to survive though.
 
Considering that we're at a point where future silicon scaling is dubious and focus is moving onto advanced packaging and the demand for silicon production likely won't go down any time soon Intel's foundry business is in a weird spot where on the one hand it's a massive money sink at the moment which seem to be incapable of making even Intel's own chips to the required specs; but on the other hand all production lines will likely even out at the physical scaling limit over the next 10-20 years anyway and those which are struggling right now will probably get up there eventually too so there's not a lot of reasons to just cut that off at this point.

Maybe the idea to just use TSMC for performance/power critical products for now can in fact be a correct one? Would allow them to sell good products while their foundry would just catch up slowly arriving to a point where it would be just as good as any other option eventually. They will still need to cut off some fat and downsize quite a bit to survive though.
It’ll just get even more expensive to catch up even if the benefit of being at the leading edge is smaller, so they’d be guaranteed to just fall further behind in terms of the number of years, and the relative positioning would be constant at best (smaller benefit of the leading edge but further behind it). Intel doesn’t have 10-20 years, and if everyone ends up eventually all getting to the same point, then China will beat them on price anyway.

They do have the Intel 3 process for base dies of Foveros/advanced packaging and the ability to mix that with TSMC advanced processes for the compute die, i.e. similar to the TSMC N6 dies that AMD is using in the MI300X but denser (I suspect there’s not much point going below that for the base die anytime soon so it might have a very good long amortisation period). They could probably decide to use it for all SoC/IO/last level cache functionality with leading-edge TSMC for the CPU/GPU/XPU only, and try to market that to 3rd party customers as an alternative to TSMC for packaging+base die. But that’s really « giving up on the leading-edge » with all the internal and geopolitical implications it has in my opinion.

The other problem is 14A uses High-NA EUV at least for critical layers so it’s not clear to me how much of existing lines they could reuse. There’s an argument based on that & likely 3rd party volumes that Intel needs to significantly reduce the amount of fabs/wafer volume they build for 18A even if it means it doesn’t have the economies of scale required to be competitive on cost at all, and bet it all (again) on 14A once they have « proven themselves » on 18A and have a more mature EDA/PDK/etc ecosystem. That’s a pretty big ask for shareholders and the board to give Intel & Pat Gelsinger especially even more time and money to do this though given their failures in the last several years.

I suppose a lot depends on whether Pat survives this and buys more time, or if he’s replaced and his successor is forced to significantly change the strategy (whether it makes sense or not; they can’t replace him then do business as usual). My expectation is that he will survive this, but all bets are off if he doesn’t.
 
If Qualcomm (who's stock is around $160) bought Intel (who's stock is around $20) what happens to the valuation of the stock for Intel stockholders?

Sorry, I've been obsessed with Intel's stock since it dropped and I have no clue what would happen to it if Qualcomm bought a big chunk of Intel.
In theory, if Intel sells a chunk of their business for a fair price the value shouldn't change. They had an asset worth $x, now they have $x. Wall Street sentiment as to whether it's a good strategy and they're putting the cash to good use will of course play a part, but that's true regardless of whether they sell parts of their business or not.

The actual relative share prices are irrelevant, what matters is the market cap, i.e. share price*number of shares
 
Back
Top