Exclusive: Intel's new CEO explores big shift in chip manufacturing business (Write-off 18A and move focus to 14A)
Posted by DubiousLLM@reddit | hardware | View on Reddit | 184 comments
SignalButterscotch73@reddit
That would be a stupid move.
18A is struggling to get customers because Intel hasn't rebuilt confidence that I can launch it on time. Stick to the roadmap, prove that manufacturing troubles are a thing of the past, improve the customer design experience much as they can with good design tools.
Customers will only come with trust, trust is built by successfully doing what you say you will do when you say you'll do it and for the price you say (if you disclose a price before launch)
TSMC's biggest advantage from everything I've read is from the design experience, that should be Intel's main target not pushing to the next node early, overambition is what caused the massive stall in progress with 10nm/Intel7.
Eat the elephant one bite at a time.
Exist50@reddit
That ship has already sailed. 18A is years behind their initial schedule. So now they have to hold 14A schedule, and if and only if that remains on track through volume production, then some customers will be interested in the next node (10A?).
But in broad strokes, you're right, Intel needs to make the customer experience as seamless as possible. 3rd parties won't tolerate the hoops Intel's internal teams have had to jump through.
hardware2win@reddit
You still lying about 18A timeline?
Exist50@reddit
Lmao, you someone's alt account? Still insisting 18A is on tract?
hardware2win@reddit
You literally wrote that 18A is years behind when it was originally announced as 2025 node and only years later for some peroid of time they believed they could make it 2024 H2, but later went back to 25, so how it can be years behind?
Exist50@reddit
H2'24 -> H2'25 is one year. -10% perf is another year's worth.
hardware2win@reddit
Now youre using your own, not expert logic about feature/perf to time translation?
Wow.
grumble11@reddit
It was in Intel messaging basically supposed to be launched right now with products on the shelf. They aren't 'years' late but they are a year late with products likely not on the shelf at scale until at least Q1 2026, and likely Q2. That's pretty bad. TSMC has been reliable, partly because of their partnership with apple - the set in stone cadence of apple's device releases means that their manufacturing partner MUST hit their schedule, it doesn't matter what happens, TSMC would work people do death to do it.
hardware2win@reddit
So even with such approach my point stands
SignalButterscotch73@reddit
Kinda depends on what roadmap you look at. In 2021 they were talking about Q3 2025 but in 2023 it was Q1 2024, in 2024 it was H2 2024. Right now it's planned for H2 2025 again.
That's one of the biggest problems Intel had under Pat Gelsinger, the plan was never consistent or executed.
Announcing a plan and succeeding on execution is all the new guy needs to do for his first step, making it harder by pushing ahead early isn't likely to be the right move. If he gets 18A out this year then he can point at the 2021 roadmap and claim success. Focus any spare effort on that rather than pushing ahead early.
ElementII5@reddit
The problem with all of that is that Intel massively relaxed 18A specs over time that it is closer to 20A than to the original 18A. The original 18A specs are closer to 18A-PT. That is the kind of revisionism we are talking about.
ShubhamDeshmukh@reddit
What he wanted to say is, you keep your shop open even if you have no actual customers and your goods are not getting sold. There are people passing by and people will notice it if you are persistently available ( as an option). No matter how "last ditch" option it may be.
By withdrawing 18A from external marketing, they are basically closing the shutters which gives a pretty doomed optic to external parties including the public. So even if Intel knows very well it is a mature node & they will release their own 18A products on time, do you think people will be eagerly waiting to buy 14A? This incident will be forever etched in their minds. They will remember the shop was closed last time and won't even bother checking on 14A.
SlamedCards@reddit
18A isn't years behind schedule
They wanted risk production 2H 24. They announced risk production April 1st of 2025. 3 month slip. Panther Lake was announced as a 2H 25 product ages ago. 3 month slip forced it to be meteor lake launch
Exist50@reddit
That was supposed to be the volume production timeline, not risk.
And on top of pure schedule, they announced a 10% backoff in performance. That's like a year's worth by itself.
SlamedCards@reddit
No. Intel's manufacturing ready was always risk production
Theres some cadence article about it.
https://community.cadence.com/cadence_blogs_8/b/breakfast-bytes/posts/iedm-keynote-ann-kelleher-on-future-technology
Exist50@reddit
That may be a retcon or marketing CYA, but was not the original intention/claim. You can see that with Intel 4 schedule when they announced that delay, and also evidence in the ARL 20A timeline. Or look at Intel 3. "Ready in second half of [2023]" was the start of volume, more or less, because that actually hit the (revised) schedule.
Less formally, they also compared themselves to TSMC volume nodes at given times. "Unquestioned leadership in H2'24, etc etc".
SlamedCards@reddit
I mean article is Dec 2022 but ok
Intel 3 HVM didn't start in late 2023. They announced HVM like middle of last year
Exist50@reddit
The plan, at least, was SRF HVM by end of '23. Looks like that slipped a little, but should have been well past ES by summer. Intel doesn't typically do launch events until after that milestone has passed. Also heard something about a late stepping (C-something?) along the way.
PrestigiousBeach555@reddit
if inel is, considering doing it I guarantee they aren't taking thedecision lightly
auradragon1@reddit
It's not a stupid move. Clearly no one wants 18A except for a few small test chips from Amazon and Microsoft. It's time to admit defeat on the marketability of 18A and focus on the next node. Stop sunk cost. It hurts but it's pragmatic.
yabn5@reddit
No, the ship has already sailed for 2nm, TSMC won decisively. Intel’s best bet to win more than a token amount of external fab customers is 14A. Hopefully the PDK will be in a much better state much earlier.
rubiconlexicon@reddit
Am I just uninformed or does it seem like Intel has been on a roll of "well this process didn't really work out, let's focus on the next one" for 10 years now? When will we see high volume fabrication of a desktop CPU or GPU on an Intel node?
Exist50@reddit
Lower end NVL, probably.
Helpdesk_Guy@reddit
IF it doesn't again gets Arrow-Lake'd and happens to be shifted to TSMC just mere weeks before release.
SlamedCards@reddit
There's exist 50 skepticism which I understand. Then we got conspiracy theorist help desk. Panther lake is 5-6 months from launch on 18A. Maybe hold off a bit
Former-Plant-3834@reddit
Exist50 is negative on Intel.. because reality is negative on Intel. He's not a hater, just stating the facts.
joomla00@reddit
that's partly how they got into this mess in the first place. Skipping steps adds risk. Do it enough and the risk factor compounds into the market kickin yo ass
Vushivushi@reddit
Celestial dGPUs should be on 18A. I don't know if they are, but they need to be.
Intel's CPU market share continues to hemorrhage risking fab underutilization and they have no significant foundry customers besides themselves. They need a new product to manufacture and fill the fabs.
They're also producing CPU compute tiles at TSMC to keep up with competition so that's also bad.
Intel GPUs use way more area on the same node so that's why Battlemage supply sucks, they are first to lose if they ramp and Nvidia/AMD engage in a price war.
They pretty much have to go IDM mode to compete in the dGPU market.
At least gaming GPUs are stagnating and demand is pent up from AMD and Nvidia drying up inventories and laddering up prices. Should be a decent opportunity for Intel if they don't miss the 2026 window.
They need to be on 18A when Nvidia/AMD are still on N4.
vandreulv@reddit
I can't wait for all of the 14A++++++++++++ refreshes.
Shadow647@reddit
Intel 7 was very high volume
d00mt0mb@reddit
You mean 10mm
Tasty_Toast_Son@reddit
We really deadnaming process nodes now huh
Geddagod@reddit
I get why people dislike the newer nodes of the Intel renaming, but I don't think the Intel 10nm SF to Intel 7 renaming was all that controversial.
SmashStrider@reddit
Pretty much. Most of the last remaining hope that some are clinging onto for Intel is that 'this product was a failure even if a step in the right direction, hopefully the next product is better", something that has been wearing away over time, and now confidence in the company is at an all-time low as a result.
Exist50@reddit
Objectively, 18A is in an awkward spot. You have an N3 competitor arriving, from a foundry customer's perspective, a solid 2-3 years later. So you need a customer who cares about PPA enough to bite the bullet vs the very mature N4, but not enough to actively pursue the leading edge (N2, A16, A14), and they can't be strongly dependent on ecosystem IP, and they must be risk tolerant enough to go with Intel. There's not many who're going to fit that bill.
The bigger question is if 14A actually changes that status quo. If you just shift the window +1 node for each in 2 years, the problem remains. So what does 14A need to deliver to actually get customer buy-in?
At least Tan might be better at selling the node to potential customers. Gelsinger supposedly had some rather significant fuck-ups in that area.
Responsible-War-2576@reddit
The wild thing is now Intel is trying to be revisionist in claiming 18A was never meant to be a big external node.
No, we just scaled back that capacity when we realized we had no one willing to jump on board.
PastaPandaSimon@reddit
They can find plenty of customers looking for value while looking for a "close-enough-to-leading" node. They just don't want to trade their issue for the pickle Samsung is in as a "not as good discount fab".
imaginary_num6er@reddit
At this rate, TSMC will beat Intel on actual back-side power delivery. It was supposed to happen as far back as 20A
Exist50@reddit
Seems to be basically a tie, though you can rightfully point out that Intel's implementation is more advanced.
Doesn't really matter at the end of the day though. Customers don't buy based on bullet points of node features; they buy based on PPAC. There's no inherent reason a BSPD node must be better than one without, for example. Or even GAAFET vs FinFET.
If anything, this argument reminds me of 10nm. Intel was first to some cool bullet point features like Cobalt wires that were being hailed as the future. Didn't really pan out that way.
Helpdesk_Guy@reddit
Have we seen actual hard proof of even existing of their BSPD aka PowerVia, aside from nice PowerPoint-slides?
Exist50@reddit
No more than anything else 18A, but I don't think its existence is a reasonable thing to doubt. You've seen my comments. You know I don't trust Intel implicitly. But that's not the kind of thing they lie about.
QuestionableYield@reddit
Maybe PPACD would be a better acronym where D is dependability (dates, yield/volume). Nobody wants to jeopardize years of design work and the opportunity cost of missing your product launch window or promised customer specs by going with such an unproven partner. How much better does the PPAC side of things have to be to make up for the existential product threat of low dependability?
grumble11@reddit
Not only that, but you also have a sizable IP stack - when you use a new manufacturer, they come with quirks - their own way of doing things, their own software, their own support networks and so on. Intel is 1) different in this respect, and 2) no doubt materially behind.
-protonsandneutrons-@reddit
18A was the "bet the company" node. If nobody trusts 18A enough for a major contract…
Stratecherry's analysis rings true: today, no market-based reason exists for [an external] Intel Foundry. Older nodes are at plenty of foundries + TSMC + Samsung. Leading edge have TSMC's offerings and, in the worst case, Samsung's offerings.
More competition is good, but you got to be competitive enough (and then some, as you state, it will be a risk). Surely, 18A should be (or should have been?) good enough for some mid-range mobile SoCs or GPUs or smartwatches or some smaller datacenter CPUs, if not any start-up's AI accelerator.
Perhaps we'll know for sure in a year when external 18A products should be on the market, but I agree: it just feels like kicking the can down the road, with just as little transparency as before.
Exist50@reddit
Startups are probably the last ones to make such a move. In addition to all the problems around licensed IP, they have too much at stake. If Intel drops the ball and it kills your first and only product, that might be enough to doom your entire company. And even if the node works well, your team will absolutely take longer to develop on it by inexperience alone. Startups generally want to get their first silicon out as fast as possible, because every day before that is burning through money without anything to show investors or would-be customers.
The ones that care about cost optimizing their node choice are generally big companies shipping a lot of volume that can afford to take a risk. Like Nvidia with their gaming GPUs. They'd be very unhappy, but if the 6000 series were delayed a few months, for example, Nvidia would still be just fine as a whole.
Healthy-Doughnut4939@reddit
Intel should practically give away 18A for free, if only to build up trust and long term foundry customers in the industry.
Earning customer trust and ensuring they get familiar with porting their designs to your process node is crucial for long term foundry viability
Creative-Expert8086@reddit
Intel's 0 margin might still be higher than the price TSMC is going to charge.
Exist50@reddit
Surely they are. But in many cases, even "free" might not be good enough. When Intel's had this problem in the past, they've been able to brute force the issue with subsidies ("contra-revenue"), but they don't have the funds for that these days.
-protonsandneutrons-@reddit
That is a great point; I was thinking "startups make risky bets to save cash, what's one more risky bet", but, how many would bet their start-up when safe alternatives are widely available, especially on something as unforgiving as hardware.
Tgrove88@reddit
Can't forget 20A
SlamedCards@reddit
I mean people are still doing 16nm,7nm,5nm, etc. Of which TSMC will only have 1 4nm and one 3nm fab in US.
So they need to overtime catch those customers who slowly fall down the funnel who can be enticed into wanting a US product
I agree that first wave customers the ship already sailed on those that wanted 2nm for 26/27 products. But 18A should get 2nd, 3rd wave etc over the years
Exist50@reddit
Legacy nodes are carries by a combination of cost, IP availability, and design reuse. Maybe 18A is cost competitive enough to go against a mature N3 (with a hit to margins), but that isn't a given. The other two factors are more challenging. Intel has not had time to build up anything like the kind of design ecosystem TSMC has. Companies on the leading edge often do a lot more in house, so less of an issue for them, but it's a bigger problem for selling a legacy node. I assume Intel's covering most of the basics somehow, but it only takes one missing part to be a deal-breaker.
The other problem is a lot of companies have a "no one ever got fired for using TSMC" attitude. Not everyone has a multi-hundred person design team, and schedule certainty matters. When in doubt, a lot are going to pick the suppliers they and their team are familiar with, even if the technical merits suggest otherwise. Another tough nut to crack.
I'm not convinced "US-made" is really a selling point. Maybe for some industries like defense, but they're not who Intel needs to get on board.
Healthy-Doughnut4939@reddit
I think Intel 16 is a good start in finally moving the legacy 193i fabs away from making Intel's chips on the horrible and expensive Intel 7 node
They do need to get more external customers for Intel 16 though as AFAIK only one small customer in Europe is making chips using that node.
Intel-UMC 12 should greatly help with this shift away from Intel 7 as it could allow Intel to make chips for UMC's customers if Intel offers a better deal or if UMC is experiencing capacity issues.
Exist50@reddit
IIRC, Mediatek has been rumored, but also for something small. And yeah, unclear how they're marketing it with the 12nm. Don't think we've heard much about that in a while?
SlamedCards@reddit
TSMC had delayed 2nd fab and stopped construction in 2024. Now they are jacking up prices for us nodes (reportedly) and have announced 3 more fabs trying to accelerate timeline
US product demand is certainly increased if their doing that. Wether its tariff threat idk. But it's definitely a sea change since last year
On IP side a big thing for Intel are players moving down from 7nm and 5nm in next wave. Big thing is IP from cadence and synopsis. If Intel has that IP they should be sorta ok. Since IP from EDA guys is becoming really common now
Something like 28nm or above is much more cost conscious and probably custom IP
Exist50@reddit
Tbh, I'd wait and see what happens right now. There're a lot of companies announcing "plans" with big numbers attached to curry favor with the current administration. How real those intentions are remains to be seen.
scytheavatar@reddit
Yeah if what this article wrote is true then I question what was the point of replacing CEOs. This would be the kind of out of touch with reality move that makes me think Tan is just counting down the days before he gets fired too. The reality is that Intel is now the AMD of the GPU market and TSMC is the Nvidia, customers are not going to pick Intel over TSMC just because of better performance. Intel needs a win streak to build up their reputation to a level where customers can trust them. It is sad that Intel has been number one for so long that they have no idea how to be number two (except they are really number three in the foundry wars).
Exist50@reddit
I mean, Pat's firing was simple. He failed at everything he said he'd do, and burned through a massive amount of money in the process, halving the share price along the way. That gets you fired. Now Tan has a mandate from the board to cut costs, and that's what he's doing. The board won't fire him as long as he's doing what they want him to do. And Intel probably can't survive another halving of share price and remain independent.
scytheavatar@reddit
Board wants Tan to make money, and there's more to doing that than just firing people. If Tan is just doing everything Pat was doing then it is not clear to me why he will be able to get customers the way Pat can't.
Exist50@reddit
The board wants Tan to make money, but that's long term. For now, they're just desperate to stem the bleeding.
Tan reportedly left the board because of his dispute with Pat over how much Intel should be cutting. Also, there's some indication that Pat himself was an obstacle to acquiring foundry customers. If you believe rumors, he announced that Qualcomm was a foundry customer without asking Qualcomm for permission first, and before any formal commitment was made. This pissed them off to an extreme degree, and drove them away as much as the technical failures did.
Healthy-Doughnut4939@reddit
I think that Lip Bhu Tan's experience as the CEO of a EDA design tool company will be invaluable to Intel as he will have first hand experience dealing with external customers and what they want out of a particular node.
He's the most qualified replacement for Pat Gelsinger and probably the only one willing to take up the job.
I think Pat was too tunnel visioned with the foundry and he made the decision to neglect Intel's product division with horrible consequences for the company.
Chronic under-investment in the product division allowed AMD to gain significant serve, client and gaming market share. it also allowed AMD to make significant inroads with OEM's
This underinvestment was also a huge reason on why they missed out on the AI boom.
If they invested significant amounts of money into Habana Labs then they could've been able to take part in that gold rush.
Instead Intel Is being left behind while Nvidia and AMD rake in boatloads of money from this huge cash cow.
I think that's what really pissed off the board and why they fired gin.
SlamedCards@reddit
For people that don't read the piece. They aren't writing off 18A nor scrapping internal products
Rumour related to external customer marketing
Reactor-Licker@reddit
“To put aside external sales of 18A and its variant 18A-P, manufacturing processes that have cost Intel billions of dollars to develop, the company would have to take a write-off, one of the people familiar with the matter said. Industry analysts contacted by Reuters said such a charge could amount to a loss of hundreds of millions, if not billions, of dollars.”
SlamedCards@reddit
External sales
Process is literally not getting written off. All internal 18A products. 70% of Panther Lake, north of 70% of Nova Lake and 100% of diamond rapids wouldn't exist.
If it was write off, write off would be in tens of billions and products would be cancelled
ElementII5@reddit
If it is not viable for external customers it won't be viable as an internal node.
We'll probably see Intel lean more on TSMC, more than they already do. It's probably a yield/cost issue.
SlamedCards@reddit
Every Intel node that has ever existed has not viable for external clients.
Didn't stop the node from producing some great products
nanonan@reddit
They've had a few small ones, but Gelsingers betting on landing a major external customer certainly hasn't panned out.
ElementII5@reddit
That has been true for 14Finfet and before. Intels foundry is hemorrhaging money and they are losing market share and its about to accelerate. SFR won't be a match for Zen6 EPYC. They just won't have the volume for their own fabs.
Geddagod@reddit
The situation in client (the much larger and more important segment) can only get better. The situation with ARL is so bad that getting better is pretty much the only scenario lol.
PTL looks like it will be outright better than the Strix Point refresh for 2026 too.
Probably, but DMR core counts will prob be high enough where it will compete with Venice Dense. Could even match it tbh, I think it's rumored to be 256 cores? Bionic thinks it's >192 at least.
Of course I doubt even if it does match core counts, it would match in efficiency or perf/watt, but the thing is though that the competitive situation vs now, where Intel realistically has no dense part competitor at all, is an improvement.
They have plenty of 18A volume. Hence WLC being a thing, IO dies being built on 18A, and why Intel claims they can dramatically expand 18A volume too, if need be, for external customers.
ghenriks@reddit
Not true
When you have been a business struggling as long as Intel has been external customers are reluctant to trust that you can deliver
So one way for Intel to (hopefully) build trust with external customers would be to shift focus to selling the next node while demonstrating competency in manufacturing your own products on the current node
ElementII5@reddit
That is assuming that the reason customers are not signing on is because of trust and not because of technical issues. 18A is not an economical node though. The proof will be in the pudding. If the 18A products Intel will release are small chiplets with low performance it's all we need to know.
Geddagod@reddit
According to who?
Intel claims it will be economical, and economics of the node is the least likely reason why external customers won't use it.
For economics?
scannerJoe@reddit
A write off, in the context of corporate financials, is not (necessarily) that something gets cancelled, but rather that the value of an asset is reassessed, so that the loss in value can be used to reduce tax payments. So if Intel invested $10B into 18A, but now says it's only worth $5B because no external customer is using it and they will therefore not make their investment back, they can declare the "missing" $5B as a loss and calculate that against the income taxes they have to pay.
Professional-Tear996@reddit
"People familiar with the matter" - yeah these people don't know what write-off means. Garbage as usual from Reuters.
auradragon1@reddit
I think it's you who doesn't know what "write-off" means.
It can mean a reduction in value to the the estimated market value. It doesn't just mean worth $0.
In this case, this article is basically saying that 18A is worth much less now because it will stop trying to market to external customers (because no one wants it). Instead of Intel continuing to pour money to market 18A which no one wants, it will simply lower the value of 18A assets due to loss of potential customer income.
Professional-Tear996@reddit
Reduction in value to something non-zero is a write-down. A write-off is always a reduction in value to 0. Go read your definitions. And maybe ask Reuters to do the same.
Why should 18A be worth much less due to not having external customers? Did they cancel PTL, NVL, CWF and all undisclosed products that are being made on 18A?
auradragon1@reddit
Reuters used the phrase correctly here. Intel is writing off the investments used to develop the 18A external side
Why wouldn't it? Fewer customers using the fab means lower utilization rate and shorter life span of the equipment and R&D spent.
Professional-Tear996@reddit
You don't write-off "investments" which have been fully realized LOL - we SAW working Panther Lake at Computex in May.
Shitstain of an article and why are you justifying this garbage when you have lost money on $INTC even after tracking it for 25 years?
auradragon1@reddit
How many times do we and reuters need to repeat the word "external" for you to understand that it's the external part of 18A that is getting written off?
Professional-Tear996@reddit
Define "external part of 18A".
auradragon1@reddit
The cost that went into developing 18A for external customers including R&D, software, node changes, as well as expected lifetime equipment revenue, etc.
Professional-Tear996@reddit
This is absolutely not how things work. You are intentionally muddling accounting terms to push your BS agenda.
The way Intel intends to use 18A is going to be similar to the way external customers would have used 18A for most of the overlapping product segments they all operate in.
There is no "special sauce" RnD costs needed for 18A just for external customers if Intel didn't have uses for them as well.
auradragon1@reddit
Do you have a source?
Professional-Tear996@reddit
Reuters has a 100% track record being wrong about Intel for the past 1-2 years.
auradragon1@reddit
Here's a year's worth of Reuters Intel articles. https://www.google.com/search?q=reuters+intel&sca_esv=ebea5675d131f87f&source=lnt&tbs=cdr%3A1%2Ccd_min%3A4%2F1%2F2023%2Ccd_max%3A4%2F1%2F2024&tbm=
Go ahead and prove to us that they're wrong on each one.
Professional-Tear996@reddit
Yeah filter the articles that are specifically rumors based on "exclusive" reports based on "sources" and "people familiar with the matter" and then we'll talk.
You don't even understand the difference between a claim based on a report by Reuters with information they supposedly have exclusive access to and regular articles based on what Intel is doing as a business.
auradragon1@reddit
I'm waiting for you to do that. I'm not the one claiming they're 100% wrong on Intel.
I sense that you just don't like all reports that Reuters puts out on Intel due to some internal bias; maybe x86 gamer? Intel stock holder that doesn't think what Reuters is reporting is good for Intel stock? Who knows.
Clearly saying that Reuters is making up stuff about Intel because they want to manipulate the Intel stock is a preposterous claim.
Professional-Tear996@reddit
Sure, here we go:
And in most of the cases we had significant stock price movements because all these reports were put out during market hours or after the preceding day's close.
3rd March 2025 - down by 4%
13th March 2025 - up by 14%
5th April 2025 - down by 11%
Textbooks definition of stock price manipulation through unsubstantiated rumors disguised as news.
auradragon1@reddit
It takes months/years to pull off a sale like this. Companies going bankrupt will also have job postings.
This was reported by The Information. Not Reuters.
This was wrong.
So was this wrong?
How is this wrong?
Here's an exclusive report by Reuters that was in fact correct: https://www.reuters.com/business/intels-top-strategy-officer-depart-this-month-2025-06-28/
So not 100% wrong as you claimed already.
Professional-Tear996@reddit
I have a simple standard for judging claims as right or wrong. If it turns out to be true, it is right, else it is wrong. Either accept this simple calculus or don't pretend that you know any better.
And in all those aforementioned instances, Reuters was demonstrably wrong.
You should be the last one to have a say on stock prices because you have investments that are probably down by 50% given your claims that you have been tracking INTC for 20 years and go around asking people if they have positions as well.
auradragon1@reddit
Which is wrong. Reuters was right here, for example: https://www.reuters.com/business/intels-top-strategy-officer-depart-this-month-2025-06-28/
Reports of Broadcom and Nvidia testing on Intel nodes were never proven wrong. Reports of Intel looking to sell NEX has not been proven wrong. Reports of TSMC/Broadcom/Qualcomm looking into buying parts of Intel were not wrong.
Just because the end goal did not materialize does not mean the report was wrong.
Professional-Tear996@reddit
Reuters has been more wrong about these claims about Intel than a certain country in the ME having the USA as its lapdog has been wrong about its enemy acquiring nuclear weapons.
Your barking is not going to change the reality of Reuters and their reporting on Intel.
auradragon1@reddit
But they're not wrong in many cases.
Professional-Tear996@reddit
I have given you 5 instances in the last 4 months alone where they have been wrong.
auradragon1@reddit
Except that they weren't wrong in at least 3 of those instances.
aminorityofone@reddit
dont feed the troll
Professional-Tear996@reddit
If things claimed don't reflect the actual reality, they are wrong. Period. No amount of semantic gymnastics from your part is going to change that.
Illustrious_Bank2005@reddit
What is the flavor of TSMC's acquisition of Intel?
Illustrious_Bank2005@reddit
At TSMC What are the benefits?
imaginary_num6er@reddit
Must be Tom from MLID
Professional-Tear996@reddit
These are fluff pieces aimed at stock price manipulation. To write-off something literally means to set the value of an asset to 0.
Reuters would like you to think that "people familiar with the matter" are telling them that Intel's Chandler, AZ fab is going to be worthless.
tempacc_nit@reddit
Well 18a will lose them a huge sum of money so it is actually worse than zero.
Silent-Selection8161@reddit
I mean, they basically already wrote off 18A for customers and are now touting 18AP.
For those wondering why, I've seen industry professionals claim the PDK, the thing you use to actually design/tapeout chips, is too hard to work with and has resulted in too many hardware bugs for third parties to be confident enough to go with Intel. The actual process itself would be competitive, if they could get good chips out of it.
So I guess the decision is between trying to fix it by 18AP or concentrating on 14A
Dangerman1337@reddit
Yeah makes sense, shove much PTL, Xe3P etc onto it while curtailing expansion of 18A beyond 18AP-T.
Illustrious_Bank2005@reddit
Intel is a group of people with intellectual disabilities, so they are liars
NeroClaudius199907@reddit
Doesnt this mean its over? Pat did bet the entire company on 18a
Creative-Expert8086@reddit
Not only that, in his IDM2.0 framework, intel will setup firewall, and also the design team can choose the most cost effective fab instead of only in house. Intel has outsourced a lot of fabrication to TSMC while shipped 0 wafer to non intel client.
xternocleidomastoide@reddit
No. It means most people commenting haven't read the article.
heylistenman@reddit
He was obviously being hyperbolic. It’s not as black and white as that, it rarely is.
PrestigiousBeach555@reddit
intel is losing on all fronts right now, have you seen some, of the data center stuff? it's downright embarrassing, as in intel losing by 100% and the r reviewer asking why the product even exists. Intel proposed 52 core triple is a fucking mess nobody wants. GPU not doing well, behind in gaming no heft at all.
thr only thing intel had going was spending all their money on 18a, I've had a feeling for thr last 5 months it is a failure, but yes this might be thr end of intel if it doesent work
tsm may not have thr capacity, and even if they did it's too expensive for intel right now and they don't have their discount
Geddagod@reddit
Honestly saying that Intel is now betting the entire company on 14A might just be true though.
Intel claims they won't build out 14A if they don't get external customers, so Intel's future as a leading edge foundry seems to be contingent on that node.
ElementII5@reddit
If he would have split up foundry and design the moment he became CEO both of those entities would be in a much better place today.
As it is the old Intel won't come back. Foundry will never work out. They will close the foundry side down as node development cost rises exponentially over time and you can't just support that with the business that just intel brings in. They will remain a design company like AMD and Qualcomm but in a much weaker state than if they had split up 5 years ago.
So as such yes, he bet the company and destroyed it.
jorel43@reddit
Yep and it couldn't have happened to a nicer company. Karma really is a b$&#h
NeroClaudius199907@reddit
In the meanwhile tsmc gets to entrench current customers more and more, and their r&D ever more increases. How would the fab financing even work? Can intel afford to keep 18a not fully used? They're already bleeding money on their fabs
PotentialAstronaut39@reddit
Wasn't 18A THE node that was touted for years to bring Intel "back in the game"?
Astigi@reddit
Intel never ceases to disappoint
SteakandChickenMan@reddit
They did deliver on 5N4Y - they iterated through all of the processes and shipped products on all of them within the timeframes they outlined back in 2021. I’m not going to go into the semantic game of whether what Intel called process readiness = products shipped or risk production because we’ve had that debate multiple times before
18A is in the N3 ballpark. 4 years ago, TSMC was shipping N5 and there was nothing on the Intel side remotely close. That’s a huge difference.
You straw manned my last point so I’m not going to further engage in this discussion in earnest good faith.
Geddagod@reddit
20A? Literally just 20A is enough to refute this argument.
SteakandChickenMan@reddit
Technically sure you’re correct - only 4N4Y (though they had that internal node they disclosed that they developed as well). Regardless, we all know the point of 5N4Y was to get their process competitiveness to a certain point, which they did.
Reactor-Licker@reddit
The goalposts are practically being towed by a truck at full speed at this point. 20A was supposed to be Intel’s first mass production node for external customers, then no one wanted it and it got canned in favor of 18A. And now, the supposed “TSMC killer” is also on the chopping block.
The whole “5 nodes in 4 years” looks to have pretty much failed.
Intel 7 - Mostly successful, roughly equivalent to TSMC N7, but can’t be used for external customers.
Intel 4 - Pretty much a one off for Meteor Lake. Nothing else uses it, within Intel or externally.
Intel 3 - Appears to be another one off, this time for Granite Rapids and Sierra Forest. Doesn’t look like anything else will use it either.
Intel 20A - Cancelled. Rumors say very poor to weak demand externally.
Intel 18A - Remains to be seen, but doesn’t look good. Heck, even Intel themselves supposedly chose the “inferior” TSMC 2nm node for Nova Lake and reserved the use of 18A for low power and low clock chips (Panther Lake and Clearwater Forest).
If and when this announcement goes official, I think the idea of Intel beating or even matching TSMC will be dead.
experiencednowhack@reddit
5 nodes in 4 years was a fraud on day one. If you have execution struggles because each node is exponentially harder...how does your CEO declaring it will be a thing make a difference. The problems are still there and they were just taking bigger dice roll bets.
ResponsibleJudge3172@reddit
Intel 7, Intel 4 and Intel 3 are fruits of that effort
Reactor-Licker@reddit
Intel 7 is just rebranded “10nm Enhanced SuperFin” and was effectively the node where Intel finally eliminated all their issues with 10nm/7nm. It was already completed well before the 5 nodes in 4 years thing.
Geddagod@reddit
I think 10SF wasn't bad either. With the 10nm version (ICL) before that, they fixed the yields, but I think it still had clear Fmax and even perf/watt issues, which was improved in 10SF. TGL was hitting 5GHz atp.
I do wonder, at which point in this sub node development process, the node started to become really expensive.
SteakandChickenMan@reddit
This is very revisionist. When 5N4Y was first announced, people literally didn’t think Intel could ship or develop EUV processes. 10nm parts were barely out. Go back to forums and articles from that first event and you’ll see it. Now, the conversation is whether their process is on or at parity with TSMC’s leading edge. That’s a night and day shift from what people were saying 4 years ago. By that measure it’s been an incredible success. No part of 5N4Y was that it was all for external consumption.
Exist50@reddit
So basically when you set the bar all the way through the floor, it looks good. But 5N4Y is a very explicit roadmap, and they've objectively failed to deliver on it.
There isn't any legitimate conversation on that. Everyone even remotely informed knows it's behind. Remember, people were also claiming Intel 3 would be better than N3.
"Incredible success" means a multi-billion dollar money pit, no major external customers, and cancelling a bunch of fabs for lack of demand?
Exist50@reddit
I think it was a particularly bad sign when Intel lied to the public about why they cancelled 20A. The reality was it wasn't anywhere close to ready, and if they launched a "leadership" 20A ARL part in H1'25 that got its shit kicked in by N3B ARL, it would be a PR disaster. But that lie only works for the public. Their customers knew the reality.
They are using Intel 3 in a bunch more stuff (NEX custom, ARL-U, PTL-U iGfx, IO tiles?), but it seems to be almost entirely Intel internal.
It'd be more accurate to say they're using 18A where they can't afford to use TSMC. They know damn well N2 is just straight up better, and not by a small amount.
Somehow this idea never seems to die. Go back a year or so and you can find people insisting Intel 3 was a better node than N3. The goalposts always seem to move to the N+1 node.
jaaval@reddit
I think using N2 is probably a mistake. Intel’s bigger issue is margins, not getting the last 5% of performance. And using external looks bad, even if it’s just for the relatively low volume top end stuff.
Exist50@reddit
And if it's not 5%, but 10%? 15%? That's about what AMD charges another $100 for with X3D.
jaaval@reddit
I very much doubt that. Top performance difference can be mitigated by using a bit more power.
Obviously if the architecture sucks then they are trouble.
Exist50@reddit
Either node will be pushed as hard as possible for the top SKUs. Is it that unbelievable that N2 is essentially a full node of perf vs 18A? Certainly it explains Intel's choice better than 5% would.
jaaval@reddit
If Intel’s claims about speed improvement from intel3 are true I doubt that N2 can do that much more at maximum. Iirc TSMC claimed 15% faster than N3.
Though of course they report speeds at 1.1V or something like that. Who knows what they do at 1.4V
Exist50@reddit
You should take those numbers with a grain of salt. If you remember the headline about 18A being higher perf than N2, that conclusion was reached by naively assuming Intel 7 == N7, and multiplying all the marketed generational gains since. Clearly, something doesn't add up there. Perhaps, for example, their cherry picked sub block of the IP benefits disproportionately from the node's changes. And that's ignoring more malicious ways to fudge the numbers, which can and have happened. Not that you should trust TSMC's either (e.g. N3P is funny when you break it down), but Intel seems to be the most prone to exaggeration right now.
So yeah, ask yourself what number makes sense to justify Intel going to all the trouble and expense to source compute dies from TSMC. Is 5% really enough?
Healthy-Doughnut4939@reddit
I think that a big problem with 18A is that it's still a node designed for Intel's internal use. It was intercepted early on in development and converted to an external customer node
Intel is claiming to rectify this with 14A.
At the low end of the voltage curve 18A is 18% better than Intel 3
At the high end of the voltage curve it's 25% better than Intel 3.
This is a big problem as I suspect a majority of the world's chips are designed for products with low power targets like Cell Phones, tablets, laptops etc.
We don't exactly know how competitive 18A with N3 and N2, Intel's product choices seem to suggest that N2 is better although how much better remains to be seen.
Exist50@reddit
That wasn't the story at the time though. 18A was supposed to be the node where Intel pivoted towards being more low power and ease of use focused.
I'd also take those numbers with a grain of salt. They're not even a core to core comparison, but rather cherry picking some sub block of a core. It maybe be that some part happens to benefit disproportionately from PowerVia or something, but the curve is different as a whole.
But that's also Intel's largely own market. Laptops and servers (also low-mid voltage) make most of their CPU market. They should be prioritizing that operating range. Also, GPUs if that ever took off.
Professional-Tear996@reddit
Intel themselves said why they cancelled 20A. It had nothing to do with external demand.
Healthy-Doughnut4939@reddit
No but you see we've made so much progress with 18A, that 20A is already obsolete! /s
thismeowmo@reddit
Intel fab is dead. After intel 7 the products are slowly moved to tsmc. There is currently no high volume product on the market that utilizes intel 4 or 3. When the new tsmc fabs comes online intel will surely goes fabless and spin off these fabs.
Exist50@reddit
GNR/SRF? PTL and to an extent degree NVL also bring volume back in house.
SlamedCards@reddit
Intel actually has higher planned wafers starts on Intel 3/4 vs 18A in there HVM fab in Az
Indoamericanus@reddit
Let me elbow my way past the large crowds of external customers who’ve signed up for 18A. They will all be disappointed.
iwannasilencedpistol@reddit
Seeing the guy for the first time, lip-bu tan looks rather similar to pat on first glance huh
Professional-Tear996@reddit
Yet more Reuters BS.
They ARE moving focus to 14A - most of the ecosystem partners at the Foundry Connect event were talking about how they are working on 14A.
18A was to get external customers notice that Intel can build it for them - they developed a PDK that external customers may actually use, they talked about DTCO options available to customers at VLSI 2025, and they are building third party chips for a start with Microsoft.
Just because 18A won't have major 3rd party customers in volume doesn't mean it is going to be written-off.
-protonsandneutrons-@reddit
I think Reuters has been more correct than not on Intel. It's more "show, don't tell" at this point for Intel.
In 2021, Qualcomm signed up for Intel 20A. Not one product shipped. Then abruptly in 2024, Intel cancelled 20A to focus on "bet the company" 18A.
Thus per Intel, Intel 18A should produce major external foundry wins to "save" the company. Intel cancelled an entire node—with a public customer—for 18A. The 18A Microsoft & Amazon contracts Reuters mentioned seem to be minor.
Surely, if Intel's hypothesis was correct, major 18A contracts should exist for mobile SoCs or AI chip or GPUs or CPUs.
//
For example, Arm's 2025 uArches were certified by three foundries: TSMC, Samsung, and Intel. Intel stated Intel 18A was to be used:
But where are those Intel 18A Arm SoCs? Samsung Foundry (expectedly) won Exynos 2500; TSMC won MediaTek & Xiaomi; Intel Foundry won… anything? The 2025 SoCs will be Gen n-1 in a few weeks (or months).
Professional-Tear996@reddit
The cancellation of 20A had nothing to do with demand from external customers. Process nodes are co-developed having different stages of development for different products on the roadmap.
Like just before the cancellation of 20A, the work on it may have been 70% done, the work on 18A may have been 50% done, and that on 14A may have been 30% done. These aren't actual numbers but reasonable guesses at abstracting away the details in order to understand progress in layman's terms.
And at the stage at which 20A was cancelled, the D0 defect rate for 18A was less than 0.4 defects per sq cm.
The industry consensus for starting limited production trials for any new process node is when you have a D0 of 0.5.
That is why 20A was cancelled and what "betting the company on 18A" means.
auradragon1@reddit
I don't believe that. Intel is not going to cancel a node if it had commitment from enough external customers to make it worth it. Intel is desperate for any customer win.
Professional-Tear996@reddit
Did you read what Intel themselves said when they announced the cancellation? I literally laid it out what they said and yet people like you insist that it had to do with "lack of customers".
auradragon1@reddit
This? https://newsroom.intel.com/opinion/continued-momentum-for-intel-18a
Read between the lines when it comes to corporate communication.
Translation from the announcement: No customers for 20A. Time to focus on 18A. Accept sunk cost.
Professional-Tear996@reddit
Literally the reason for shifting to 18A.
Do you actually read for yourself the articles you purport others to read?
auradragon1@reddit
That's how you interpret that?
The way I interpret the press release was:
Professional-Tear996@reddit
The way I interpreted it is the ONLY way to interpret it because Arrow Lake was supposed to be on 20A according to roadmaps at around the same time as the announcement was made (August 2024).
It has absolutely nothing to do with lack of customers. If it had they would have announced Qualcomm as a customer and not have Pat say how "excited he was with the opportunity" to partner with Qualcomm in 2021.
auradragon1@reddit
It's not the only way. Clearly throwing in 18A in the press release is just to say "Intel 20A is a failure, no one wants it, but don't worry, 18A is coming along".
Professional-Tear996@reddit
Only the outrageously biased would ignore the fact Arrow Lake launched a month after that announcement and that low-end Nova Lake prototypes were being shipped around 3 months after that.
auradragon1@reddit
Arrow Lake being on TSMC just tells me that Intel knew internally that Intel 20A was dead long before they announced it to the public.
Professional-Tear996@reddit
Yeah as if the Arrow Lake 6+8 tile wasn't strongly rumored to be on 20A for a long time before that.
auradragon1@reddit
You lost me. I don't understand at all how Arrow Lake is related to Intel cancelling 20A solely because 18A had a 18A had a defect density of D0 <0.40 as you claimed.
Professional-Tear996@reddit
Because they consciously decided to move it to TSMC for the same reason they didn't want to ramp production of 20A for Arrow Lake in H2 2024 when they already achieved a D0 of 0.4 during the same time?
How hard can it be to understand - why ramp production for a node that will have a single product when the next one that will have multiple products is coming along extremely well?
auradragon1@reddit
Makes zero sense.
Your claim:
Intel cancelled 20A not because no one wanted to use it but because 18A had D0 of 0.4.
Intel's Arrow Lake was moved to TSMC because of the above
That makes no sense at all. Arrow Lake was moved to TSMC because of the failure/uneconomical nature of 20A. It's not because of 18A had D0 of 0.4. You're stretching things way too far.
Professional-Tear996@reddit
This is a heap of meaningless word salad when you seem unaware that Intel has made the transition to have their CPU designs be node-agnostic with Arrow/Lunar Lake and have given presentations about it from top people in their design teams.
auradragon1@reddit
CPU designs can't be node agnostic. You need people to design them based on node specs, verified, and tested.
Professional-Tear996@reddit
https://www.youtube.com/watch?v=EJGr-HWzGFs
From 4:17 onward -
Ori Lempel, Senior Principal Engineer, Intel Core Design says
Yet another instance where you have demonstrated your cluelessness.
auradragon1@reddit
And if you keep watching, he basically explained that they still need to be optimized for a specific node.
Intel is doing what others outside of Intel are already doing - designs that are often node agnostic. Intel previously only designed for Intel nodes.
Professional-Tear996@reddit
Here optimization means fine-tuning. The 1% he left out.
And he literally says that the choice of node is based on time-to-market vs performance/power considerations.
Arrow Lake is on N3B purely due to the combination of TTM and the choice to forgo 20A.
-protonsandneutrons-@reddit
But if Intel had a public customer commited on 20A, why would Intel cancel it? Qualcomm would be a massive win for any Foundry not-named-TSMC. Even a few actually-shipped Intel Foundry SoCs would be a huge shift in the narrative for Intel.
It would've been an unprecedented victory for Gelsinger's vision and, hell, probably its stock price, not to mention the major morale boost after years of difficult times:
But after this announcement, Qualcomm made no further mention of Intel Foundry and then before anything ever ships, the whole node gets cancelled.
If that is how customers were treated on Intel 20A, 18A has a lot of ground to make up, no?
Professional-Tear996@reddit
https://youtu.be/neu4GEX6DXk?si=ZMYqL2r1A32e-pkJ
Go to 32:50 and hear what Pat actually says - "excited about the opportunity to partner with Qualcomm" is a very different statement than "AWS is our first customer to use our packaging solutions"
In other words, Qualcomm was never formally announced as a customer.
-protonsandneutrons-@reddit
It seems to be semantics, no? Qualcomm's CEO stated it was "excited" for the breakthrough technologies in Intel 20A and Intel was a "foundry partner"
Emphasis mine. Thank you for the timestamp. Intel's CEO does cite the opportunity Qualcomm as a "partner" for Intel 20A + "development of mobile compute platforms."
Emphasis mine. This seems about as direct as one would expect from a foundry announcement.
SteakandChickenMan@reddit
Qualcomm never signed on. Articles from that time refute this.
Professional-Tear996@reddit
Except that the language used clearly says that Qualcomm is pleased to have a foundry service that is based in the USA; and they are not explicitly talking about using IFS for their products but that they are excited about RibbonFET and PowerVia in a general sense.
tempacc_nit@reddit
18a, the process the whole company was bet on. Is a total failure. Now they are already pimping 14a. You already know how this story ends.
Exist50@reddit
Don't worry. People on this sub have repeatedly "informed" me that it's as good as or better than N2. Surely they have customers lining up around the block for it /s.
a5ehren@reddit
Yeah anyone who said 18A is an N2 competitor is delusional. It’s basically N3P.
brand_momentum@reddit
They already said that most future customers were aiming 18A-P back at Foundry Connect...
This isn't news.
Illustrious_Bank2005@reddit
Intel has ended... But it's sad that the x86 is gone... I don't want to migrate to ARM that insulted Open Source (RISC-V)... what should i do
CalmSpinach2140@reddit
x86 is closed source as well lol. RISC-V isn’t all roses, you need to pay licensing fees as well for good RISC-V cores.
Strazdas1@reddit
To be fair, its understandable not wanting to migrate from what consisders one bad system to another bad one. One would preferably migrate to a good one.
Illustrious_Bank2005@reddit
Yes, that's right Then, move to a new frontier RISC-V, and learn from our predecessor ISAs. From the x86 series, learn the essence of high performance processors and compatibility… Learn embedded systems and low power consumption technologies from ARM… Learn to avoid repeating bad points and make RISC-V good. Being able to learn from our predecessors is also the privilege of young people. Well, I really like x86, so I want to believe in the possibility of x86...
Illustrious_Bank2005@reddit
However, I am not making any silly remarks about OpenSource. I know that the RISC-V also has a license. In the first place, I have a subtle feeling about ARM, which has ridiculed open source in the past...
Dangerman1337@reddit
Wonder if this means RTX 60 is Samsung SF2Z than 18A?
ElementII5@reddit
I'm not going to say I told you so, but I told you so...
"No, of course they will have external customers for 18A. It would be securities fraud! Blah, blah blah....."
Healthy-Doughnut4939@reddit
According to this article, interest in 18A has been declining with external customers
If this is true then de-emphasizing 18A by refusing to market it to external customers, producing chips for existing customers and shifting all of the fabs budget into developing 14A is a sensible move.
Unfortunately due to the CHIPS Act deal with Uncle Sam struck by Pat Gelsiger divesting from the fabs is neither easy nor straightforward.
Intel should put all of their resources into developing 14A if it's clear that customers are not showing any interest in 18A
travelin_man_yeah@reddit
They're supposed to use 18a for a number of internal products, but it still remains to be seen if 18a is as healthy as they claim. It's not supposed to hit HVM phase until later this year and Clearwater Forest has already been delayed once.
Also, it takes years to onboard foundry customers so shifting customers to 14a means even more foundry revenue delays while they spend billions more to bring that process up. And if 18a falters, it won't bite well for customer confidence.
Strazdas1@reddit
"We want Intel price to go down so we will say something scary" - price manipulators at reuters says.
Jellym9s@reddit
The takeaway is that nothing for this year to early part of next year has changed for Intel. It's only that the company is coming to grips with the reality that if they want to market their foundry to external customers, they have to design a node with customers in mind. 18A was not that. 18A-P is probably the first node that they could, but it seems they may have to start with a "foundry from the ground up" node, which would be the 14A node.
ithinkitslupis@reddit
Didn't they write off 20a and go all in on 18a...and are supposed to be entering high volume production right now?