News Intel might axe the 18A process node for foundry customers, essentially leaving TSMC with no rival — Intel reportedly to focus on 14A

[shady28]

If this is their logic:

The majority of tools required for 20A, 18A, and 14A (except for High-NA EUV equipment) are already in place at its Fab D1D in Oregon, as well as at Fabs 52 and 62 in Arizona. However, once these tools are formally put into operation, the company will have to report their depreciation as a cost.

So, they are just worried about when the accountants have to record the depreciation? And to do that, they are willing to piss off Amazon, Microsoft, and the DoD - their current 18A partners?

If this is why then Lip is nothing but another useless bean counter.

Gelsinger was at least the real deal tech head. This makes me wonder if the board is actually trying to kill off Intel so they can split the company up.

 

[bit_user]

Cutting edge nodes are all well and good, but in the Angstrom Era the benefits aren't as drastic, and a mature older process is worth more than a cutting edge less refined one.

Maybe for CPUs, you could get by with using a previous generation node. For AI, the issue seems to be that customers are comparing based on perf/$. If your perf is lower than your competitors', your selling price must be commensurately lower, hence everyone is racing to be on the latest and greatest nodes.

It's not just about clock speed, either. It's also higher density that enables more computing elements to be crammed onto a die, and that further increases the value proposition of using the latest nodes for AI.

 

[Notton]

As the saying goes: Done is better than perfect.

 

[acadia11]

So when it comes time 14A will they cut it and say let’s prioritize 9A… and so on, and so on … sometimes you need to just fight. No plan, just fight … find this more concerning not about technical merits but the inability to draw a line and start throwing some punches.

 

[Mr Majestyk]

Must say the new CEO sounds worse and worse every day. This clown is ultimately going to focus on mostly AI market, where it's probably too late since they are so far behind. AMD is way ahead and is struggling but at least battling hard especially with MI355X and later products.

18A has to be used internally as otherwise Panther Lake is DOA It's already been taped out and tested with 18A. No way they could pivot to N2 for Panther in a timely manner and also lose BSPD.

 

[bit_user]

So when it comes time 14A will they cut it and say let’s prioritize 9A… and so on, and so on …

Yeah, the precedent this sets will certainly turn away potential foundry customers of future nodes.

sometimes you need to just fight. No plan, just fight …

Uh... they do need plans, but a certain amount of faith in their business and their ability to run a competitive and profitable foundry is necessary.

In general, it's hard to simultaneously be both very successful and very cautious. Unless your business has certain "protections" or you're engaging in non-competitive practices, a degree of risk-taking is generally needed if you want to see a big pay day.

 

[fiyz]

So IFS loses billions of dollars by having all of their advanced node capacity full to the brim? I get that CapEx for new nodes is CRAZY expensive, but loans aren't going to get paid off when profits are in the negative billions. TSMC invests tens of billions per year on new nodes, but they're profitable.

I don't think so. Intel is its own customer is down on chip volume, so that's one hit right there.

I don't believe in this company at all at this point. Sell the future's concepts while literally selling off current developments that have reached fruition is not a business strategy that can continue on forever. Unless somehow can show me hard concrete evidence otherwise, it's literally living on hope.

I know banks don't run this way, so how is Intel even able to continue reneging on their promises? I wouldn't loan them one penny if I was a business, and I sure as heck aren't about to invest in them.

I hope they figure it out personally, I'm honestly considering an Intel gfx card, simply because this above MSRP nonsense is too much to support. I'm really glad Intel has products on the market that offer us consumer choices, but at the same time, they are a publicly traded company.

A lot of what a CEO does is manage the stock price.

 

[rluker5]

Intel has publicly stated that they will continue to use external foundries, as cited in the WCCFTech article.

I'm relying on Google Translate, but the Taiwan Economic Daily article says:

​

"the market heard that Intel has also joined the ranks of TSMC's first batch of 2nm customers. It will be used to produce next-generation PC processors and is currently intensively preparing for trial production at TSMC's Hsinchu plant to facilitate subsequent yield adjustments."​

​

So, it is very explicit in its statement, even if the underlying information might not be rock-solid. Not sure where you got the wording you're citing, but I know you're always very defensive of Intel so I trust you less than I trust them.

Using TSMC for some tile(s?) =/= using TSMC for compute tiles as per your claim and the claim of WCCF.

But my two translations of the source shows speculation at best as to whether Intel will use TSMC for the compute tile for Nova Lake

Maybe you should trust the source more than WCCF's or my interpretation? Your and WCCF's evidence that Intel is sourcing TSMC specifically for Nova Lake's compute tiles appears imaginary unless you have something better. I don't know, I'm just reading the words and looking for where the source says what you claim and I can't find it.


And if the source is unnamed "two people familiar with the matter told Reuters" maybe you shouldn't even trust that as anonymous sources having familiarity as their only credential are notoriously unreliable.

This is the basis of this article: two anonymous sources familiar with things told Reuters Intel is giving up on the foundry business except for some large customers. Are these sources MLID and his dog? Who knows, they are anonymous.

I'm not saying that Intel not offering 18A to outside customers other than Amazon, Microsoft and the DOD is impossible, just that this Reuters article is not credible with all of the similar accusations that have occurred over the last few years, combined with the complete lack of supporting evidence.

 

[thestryker]

Using TSMC for some tile(s?) =/= using TSMC for compute tiles as per your claim and the claim of WCCF.

I don't know where I saved the transcript, but on the earnings call in January Holthaus mentioned NVL using both Intel and TSMC. This was contrasting with PTL being Intel, but there's basically no chance PTL is entirely Intel so the implication (she may have even said Compute Tiles, but I don't want to say for sure without seeing the transcript) is that some NVL Compute Tiles will be on TSMC and N2 is the only logical choice for that.

 

[George³]

but in the Angstrom Era

Yes, the era of the angstrom, which is still measured in tens of nanometers of actual transistor element sizes.

 

[jlake3]

So when it comes time 14A will they cut it and say let’s prioritize 9A… and so on, and so on … sometimes you need to just fight. No plan, just fight … find this more concerning not about technical merits but the inability to draw a line and start throwing some punches.

If they don't have clients (beyond the three mentioned in the article) lined up for the node that's in risk production now and launching high-volume manufacturing this year, they may have already lost the fight on this one and just be trying to save face. Broadcom and Nvidia both supposedly sampled 20A (if I recall correctly) and passed and now are passing on 18A as well, I don't recall Qualcomm seriously investigating either, AMD and Apple are both extremely close partners with TSMC, and Intel themselves is using TSMC for some things now... but Intel is insisting that the foundry plans are going fine, 20A and 18A were de-prioritized by choice, and 14A is totally gonna be different!

 

[JRStern]

So this foundry talk isn't flying, but exactly why isn't clear to me. Process seems good enough for Intel to continue to use internally, or is that whistling past the graveyard? If 18a fails Intel is history - but I can't tell from this if 18a is failing or not.

 

[acadia11]

If they don't have clients (beyond the three mentioned in the article) lined up for the node that's in risk production now and launching high-volume manufacturing this year, they may have already lost the fight on this one and just be trying to save face. Broadcom and Nvidia both supposedly sampled 20A (if I recall correctly) and passed and now are passing on 18A as well, I don't recall Qualcomm seriously investigating either, AMD and Apple are both extremely close partners with TSMC, and Intel themselves is using TSMC for some things now... but Intel is insisting that the foundry plans are going fine, 20A and 18A were de-prioritized by choice, and 14A is totally gonna be different!

Totally! I think politics may save them but at this point it might be a question of who will acquire intel. I can’t see Washington allowing the only advanced leading edge manufacturers in the world especially with the importance of leading edge nodes to the burgeoning AI market and the power consumption needs to be Samsung and TSMC, and who ever emerges from the Chinese think tank and onslaught. tSMC building factories in Arizona doesn’t mean isht inventing and having command of the advanced processes for building leading edge does mean something. I can’t see this admin just seeding that ground … if they do … then not sure they understand rhe problem. So leads me to wonder who will buy Intel if this continues or who will Washington negotiate to buy Intel.

 

[abufrejoval]

Cutting edge nodes are all well and good, but in the Angstrom Era the benefits aren't as drastic, and a mature older process is worth more than a cutting edge less refined one.

That statement is so broad, it could be considered always true and always false.

To my understanding Broadcom is helping companies like Google and Microsoft to design machine learning chips, proprietary alternatives to Nvidia and AMD.

These are massive single function designs with tons of logic, which shrinks well, but also tons of SRAM, which doesn't.

In any case going with the most advanced node translates into both perhaps a performance advantage vs. what everybody else can do (buying Nvidia), but more significantly less energy consumption for running their inference workloads at scale.

And there a few Watts saved on every chip translates in perhaps one less nuclear plant every so often at the scale of their deployment.

Now all of that is without bubbles bursting, of course.

 

[TerryLaze]

If this is their logic:



So, they are just worried about when the accountants have to record the depreciation? And to do that, they are willing to piss off Amazon, Microsoft, and the DoD - their current 18A partners?

If this is why then Lip is nothing but another useless bean counter.

Gelsinger was at least the real deal tech head. This makes me wonder if the board is actually trying to kill off Intel so they can split the company up.

How do you got the piss off part?!
Those clients already are on 18A ,they have contracts that will be fulfilled.
The article says that intel will stop promoting 18A to (additional) customers and not that they are scrapping it, nvidia and qualcomm are already promoted to so if they decide to use intel they will probably produce for them as well.
They got 3 FABs fitted and if the money they will get from the 3 customers plus what intel themselves will make from their own line is enough to cover those 3 FABs then what's the issue?

Why fit more FABs with 18A tech if you can fit them with 14A instead?

18A done, let's move on.

 

[atmapuri]

Will Intel even manage to have excess capacity for 18-P to be offered to others?

Intel never was a foundry. If they don't enter the foundry buisness, their fabs might be running 100% for their own hardware anyhow.

To me this entire "Intel as a foundry" story is looking more like having something more to counter TMSC as a rival and to weaken rival positions (Apple, AMD, NVidia), because their own products are no longer addressing the bulk of the market.

But weaking TMSC is vey easy, two three dozen cruise missiles from China or a bigger earthquake and TSMC is finished.

Betting on TMSC is a very fragile bet.

 

[bit_user]

tSMC building factories in Arizona doesn’t mean isht inventing and having command of the advanced processes for building leading edge does mean something. I can’t see this admin just seeding that ground …

I've seen no evidence they care about that. The main issue of concern seems to be where the factories are geographically located. They don't seem to show a lot of interest in IP, in general.

 

[bit_user]

Those clients already are on 18A ,they have contracts that will be fulfilled.

Do we know that, for sure? Contracts of that scale always have exit clauses. It would cost Intel some money to cancel them, but it's not impossible.

 

[TerryLaze]

Contracts of that scale always have exit clauses. It would cost Intel some money to cancel them, but it's not impossible.

I've seen no evidence they care about that.

The only thing the article says is that intel stopped promoting 18A to customers...
Which makes sense because they already have customers and already promoted to those they didn't get.
Who are they supposed to still promote it to?!

 

[bit_user]

Will Intel even manage to have excess capacity for 18-P to be offered to others?

Intel never was a foundry. If they don't enter the foundry buisness, their fabs might be running 100% for their own hardware anyhow.

You're missing the point, which is that process R&D costs have ballooned to the point where they were becoming unsustainable on the old Intel's production volume. The new Intel's production volume is even less. So, unless the business somehow magically starts selling an order of magnitude new chips, they can't continue to fund development of new nodes. Becoming a foundry is the most obvious way of increasing production volume, at this point.

But weaking TMSC is vey easy, two three dozen cruise missiles from China or a bigger earthquake and TSMC is finished.

First, TSMC is the goose that lays golden eggs. It's the biggest prize of capturing Taiwan. There's no way China would damage it. Second, they have fabs outside Taiwan, so even if an earthquake causes an extended outage to their Taiwanese fabs, it wouldn't completely shut them down, nor erase their IP advantage.

Betting on TMSC is a very fragile bet.

I agree. Personally, I don't understand why Intel's fabs aren't seen as being more strategically important.

 

[thekingofALLmonkz]

Bullseye. The promises to use the latest process for end user chips went from Intel 4 >> 20A >> 18A >> 14A while sticking to TSMC, as I anticipated. At this point I might as well predict the heat death of the universe.

 

[usertests]

Presumably Intel is still using 18A internally for their own up coming processors. Some of that would not be allowed to be written off.

18A has to be used internally as otherwise Panther Lake is DOA It's already been taped out and tested with 18A. No way they could pivot to N2 for Panther in a timely manner and also lose BSPD.

The tiny Wildcat Lake dies will be legendary if they are on 18A as previously leaked, and they are cheap from having plenty of volume with no/few external customers taking up the wafers. It could be the new $150-200 system champion.

Cutting edge nodes are all well and good, but in the Angstrom Era the benefits aren't as drastic, and a mature older process is worth more than a cutting edge less refined one.

"Angstrom" is obviously meaningless marketing. What's really going on right now is a move to GAAFETs (Intel calls them RibbonFETs). There's also backside power delivery on 18A, but this is optional. I guess something has to get it, probably Panther Lake?

Over at TSMC, area scaling has slowed, but not completely. SRAM is scaling down slightly after stalling out at 5nm. Backside power delivery is delayed, and will be limited to certain nodes like A16 and an alternate version of A14 IIRC. I'm not sure if it will ever become mandatory, but it probably should be eventually. Before TSMC even gets to "angstroms", chips on the N2X node might be hitting upwards of 7 GHz, which is nice.

Older process nodes may be fine for budget products, but we are still seeing decent power efficiency increases from the latest nodes. That's beneficial for everything, but especially smartphones, laptops, gaming handhelds, etc. The silicon cost of small dies is not dramatic even at $30k/wafer. I think what we need to see soon is all L3 cache coming off of chips made on the latest nodes and being 3D packaged using a chiplet on an older node. At some crossover point this approach will save money, even for budget products. If 50-60% of your CPU die is L3 cache, that's a lot of cutting edge silicon being wasted.

 

[Mattzun]

I can't seen anyone using an advanced Intel node until Intel shows that it can use it for its own chips.
Why would anyone plan on using 18A when Intel doesn't trust its own fabs to make product

If Intel shows high yields and performance on its own chips with 18A and pricing for 14A is better than TSMC, this would change.

Recognizing that no sane company is going to use 18A or 18A-P and writing down the cost now instead of pretending that orders will come is surprisingly honest.

 

[TerryLaze]

The tiny Wildcat Lake dies will be legendary if they are on 18A as previously leaked, and they are cheap from having plenty of volume with no/few external customers taking up the wafers. It could be the new $150-200 system champion.

Only if the external customers are enough to balance the cost of the fabs, otherwise intel will have to make some money somehow and will sell them for as much as possible.

Why would anyone plan on using 18A when Intel doesn't trust its own fabs to make product

Because "anyone" can test the manufacturing themselves and see if it's worth it or not.
Also intel already committed and made some products on A18, probably their whole next gen line up.
https://www.tomshardware.com/tech-i...continue-trialing-intel-18a-test-chips-report

 

[atmapuri]

You're missing the point, which is that process R&D costs have ballooned to the point where they were becoming unsustainable on the old Intel's production volume.

If the fat lady already sang, then there is no node for Intel in the future. Especially not if they will be paying TSMC themselves as well as they already started.
As Pat correctly said, he bet the whole company on 18A. If this wont work, then there is no future beyond that for Intel Foundry seems to me.

First, TSMC is the goose that lays golden eggs. It's the biggest prize of capturing Taiwan. There's no way China would damage it. Second, they have fabs outside Taiwan, so even if an earthquake causes an extended outage to their Taiwanese fabs, it wouldn't completely shut them down, nor erase their IP advantage.

IP are people. Once the shit hits the fan, they will fly in all directions. Our civilization in general is a very fragile thing. 2-3 companies in the world which manufacture one part needed for some machines. Maybe not even 2 in some cases. A system like TSMC could take 20 years to recreate.