Wednesday, July 2nd 2025

Intel Considers Abandoning 18A Node for 14A Chipmaking Process

Intel's CEO Lip-Bu Tan is looking at a big change to how the company makes chips for others, Reuters reports citing people who know about it. The new plan might mean Intel stops offering its 18A process technology to other companies. This is different from the strategy of the former CEO, Pat Gelsinger, who had invested heavily in the 18A manufacturing process. Since assuming leadership in March, Tan has been working to reduce costs and find new approaches to revive the struggling chipmaker. By June, he began expressing concerns that the 18A process was failing to attract new customers. Abandoning external sales of 18A technology and focusing on its 14A process would require Intel to take substantial write-offs on the billions invested in its development. Industry analysts suggest these charges could reach hundreds of millions or potentially billions of dollars.

Intel plans to show the board these options later this month. However, a final decision is not expected until this autumn given the complexity and financial stakes involved. Even if they change plans, Intel will still keep its promises about 18A process including producing small amounts of chips for Amazon and Microsoft, and making its own "Panther Lake" laptop processors scheduled for late 2025. Along with standard 18A, Intel is creating two upgraded versions: 18A-P launching in 2026 and 18A-PT arriving in 2028.
Intel posted an $18.8 billion net loss in 2024, marking its first unprofitable year since 1986. Tan's alternative strategy centers on concentrating resources on 14A technology, a next-generation manufacturing process where Intel believes it can compete more effectively against Taiwan Semiconductor Manufacturing Company (TSMC). This approach aims to attract major clients like Apple and NVIDIA, who currently use TSMC for chip production. Intel declined to comment on what it termed market speculation, stating that leadership remains "committed to strengthening our roadmap, building trust with our customers, and improving our financial position."
Source: Reuters
Add your own comment

81 Comments on Intel Considers Abandoning 18A Node for 14A Chipmaking Process

#26
jabbadap
On the positive side, hope they succeeded. More choices for competitive manufacturer processes is always welcome.
Posted on Reply
#27
Daven
jabbadapOn the positive side, hope they succeeded. More choices for competitive manufacturer processes is always welcome.
This I agree with. While we have plenty of competition in the CPU space, we have very little competition in the fabrication space. This is why I and others hope that Intel fabs will get spun off and run by a more competent entity or persons. Even if the fabs were just used for the Intel 3 process and higher that would be something.
Posted on Reply
#28
mb194dc
This play is definitely a tragedy!
Posted on Reply
#29
A Computer Guy
...I was just getting excited the other day about what a great deal the 256K is. What node was that on?
Posted on Reply
#30
RandallFlagg
AI analysis of Intel nodes.

BEGIN
// Step 1: Announce new node with fanfare
DECLARE newNode
SET newNode = GenerateNodeName() // e.g., "Intel 7", "Intel 4", "Intel 3"
PROMOTE newNode WITH {
ClaimIndustryLeadingPerformance()
PromiseBreakthroughEfficiency()
AnnounceAmbitiousRoadmap()
HighlightTechnologicalAdvancements()
}

// Step 2: Develop and release limited products
DECLARE productsReleased = []
FOR i = 1 TO smallBatchSize (e.g., 1-3 products)
product = DevelopProduct(newNode)
IF product.MeetsExpectations() THEN
ADD product TO productsReleased
ELSE
DELAY product
INCREMENT engineeringChallenges
END IF
END FOR

// Step 3: Evaluate strategic value
strategicValue = AssessNodePerformance(newNode)
IF strategicValue < expectedThreshold THEN
DECLARE statement = "newNode is not the strategic opportunity we sought"
COMMUNICATE statement TO stakeholders
END IF

// Step 4: Shift focus to next node
DECLARE nextNode = GenerateNodeName() // e.g., "Intel 20A", "Intel 18A"
REDIRECT resources TO nextNode
UPDATE roadmap TO PRIORITIZE nextNode
REPEAT ALGORITHM WITH nextNode
END
Posted on Reply
#31
Eternit
This is different from the strategy of the former CEO, Pat Gelsinger
Is it? He partially abandoned 4 and 3 and completely abandoned 20A. Now 18A is going to be partially abandoned.
Posted on Reply
#32
Flanker
Serious question: when was the last time we heard good news about Intel's nodes?
Posted on Reply
#33
RUSerious
FlankerSerious question: when was the last time we heard good news about Intel's nodes?
22FF was a good node, not many problems. 14 was a good node - six month delay though; after that - crickets.
Posted on Reply
#34
yfn_ratchet
It feels like I'm watching HGTV. New head honcho walks into the house, says 'I don't like this and this and this, let's change everything', spends nearly the home's entire original value in renovations and substituting the wood for steel because the changes jeopardized the original structure only for it to be some tasteless 'open concept modern home' wannabe that absolutely no one else would want to live in.

At a time like this, I would imagine that everyone's clasping their hands together praying that Intel will keep the promises they made before the throne got handed over. This is the first in a series of financially devastating disappointments, pessimistically speaking.
Posted on Reply
#35
Eternit
RUSerious22FF was a good node, not many problems. 14 was a good node - six month delay though; after that - crickets.
I disagree. Ivy Bridge wasn't a lot better than Sandy Bridge. 22 was good at Haswell refresh. That's 2 years of delay. 14 was good at Coffee Lake that's 4 years of delay. 10 was a good one only after they renamed it to 7, that's 5 years of delay. Intel 4 (initially 7nm) is now 7 years after its initial expected date and it still isn't good. So for me the last good one was 32nm. On the other hand they had problems with 90nm. Well Netburst was not a good architecture, but Prescott was partially caused by 90nm not being able to handle higher frequencies as initially expected. Back then they were able to overcome this by changing architecture. That was good enough up to 32nm.
Posted on Reply
#36
Frank_100
EternitI disagree. Ivy Bridge wasn't a lot better than Sandy Bridge. 22 was good at Haswell refresh. That's 2 years of delay. 14 was good at Coffee Lake that's 4 years of delay. 10 was a good one only after they renamed it to 7, that's 5 years of delay. Intel 4 (initially 7nm) is now 7 years after its initial expected date and it still isn't good. So for me the last good one was 32nm. On the other hand they had problems with 90nm. Well Netburst was not a good architecture, but Prescott was partially caused by 90nm not being able to handle higher frequencies as initially expected. Back then they were able to overcome this by changing architecture. That was good enough up to 32nm.
14nm SkyLake-X was a very good node.
I still use a i9 7900x as my daily driver.
My own bench marks show it is 20-30% slower then newer cpus with comparable core counts on real work.
8 years and only a 20-30% boost in productivity.
That is the real issue.
Posted on Reply
#37
RandallFlagg
14nm was easily Intel's best node. Too good.

Hard times forge strong corporations, which drive prosperity. Good times breed weak corporations, whose failures spark hard times.
Posted on Reply
#38
sLowEnd
EternitI disagree. Ivy Bridge wasn't a lot better than Sandy Bridge. 22 was good at Haswell refresh. That's 2 years of delay. 14 was good at Coffee Lake that's 4 years of delay. 10 was a good one only after they renamed it to 7, that's 5 years of delay. Intel 4 (initially 7nm) is now 7 years after its initial expected date and it still isn't good. So for me the last good one was 32nm. On the other hand they had problems with 90nm. Well Netburst was not a good architecture, but Prescott was partially caused by 90nm not being able to handle higher frequencies as initially expected. Back then they were able to overcome this by changing architecture. That was good enough up to 32nm.
Ivy Bridge wasn't much faster than Sandy (iirc it overclocked a bit worse actually), but it did bring feature updates like PCI-E 3.0 and native USB 3.0 and had lower idle power consumption
Posted on Reply
#39
Six_Times
kondaminFeels like they put someone in place to tear the company down
Agree.
Posted on Reply
#40
ToxicTaZ
FlankerSerious question: when was the last time we heard good news about Intel's nodes?
Easy answer!

Coffee Lake Refresh
My old 9900KS (Intel 14nm++)

Raptor Lake Refresh
My current 14900KS (Intel 7)

These are Intel best nodes, without these two nodes Intel would not exist today.

(Intel 3) is for Intel Xeon 6 processor family!

Both (Intel 18A-P) Node & (Intel 18A-PT) Node are very good especially Intel 18A-PT 3D Chiplets (Intel Foveros)

Definitely looking forward to Intel Razer Lake.
1.8nm (Intel 18A-PT PowerVia Backside Power 3D Intel Foveros) 18A-PT is Intel secret weapon.

Cheers
Posted on Reply
#41
Nhonho
I would like to know about the progress in reducing the size of the SRAM memories (cache memory) of these new nodes.
Posted on Reply
#42
londiste
ThomasKIn April, 18A was supposed to be Intel's holy grail, outperforming the competition's 2nm class nodes. What changed so drastically in the last3 months ? Guess the shareholders are being lied to.
www.techpowerup.com/335442/intels-18a-node-outperforms-tsmc-n2-and-samsung-sf2-in-2-nm-performance-class
My opinion withstanding that Intel should spin off its fabs.

Isn't that the bottom line with everything coming from China? Just look at the car industry.
Did anybody read past the headline? They are considering stopping the offer of 18A to Foundry customers and try and get customers for 14A. This currently has little to do with Intel's own production that seems to be going on as planned on 18A - for consumer space the Panther Lake at the end of this year. Foundry has been struggling to find customers one way or another for multitude of reasons and given the lead times these contracts have if they start the process of onboarding someone the production is quite likely to be at the time when 14A is (estimated) to be in HVM.

It's not that Intel has shown any clear signs of things going particularly well but everyone just jumps on usual "Intel fail!" comments that get old and really have nothing to do with contents of the piece.
Posted on Reply
#43
Tomorrow
Until i see external customers and mass production on their promised node (and i dont mean just one internal mobile chip) then it *IS* intel fail to me.
They have failed too much in a row for me to trust them when they say 14A is going to solve all their problems. Why should i believe that when 20A didn't and 18A is already restricted to one internal mobile chip and thus likely low volume/low yielding.
Posted on Reply
#44
kondamin
TomorrowUntil i see external customers and mass production on their promised node (and i dont mean just one internal mobile chip) then it *IS* intel fail to me.
They have failed too much in a row for me to trust them when they say 14A is going to solve all their problems. Why should i believe that when 20A didn't and 18A is already restricted to one internal mobile chip and thus likely low volume/low yielding.
Intel as we knew it is dead, it’s going to be slow pathetic bleed like ibm to irrelevance.

I hope some of the talent finds a good place
Posted on Reply
#45
Nhonho
kondaminIntel as we knew it is dead, it’s going to be slow pathetic bleed like ibm to irrelevance.

I hope some of the talent finds a good place
Is IBM bleeding into irrelevance? When did it happen?

IBM has always worked and made billions of dollars in large-scale computing. What company do you think the US government and large companies (like banks) hire for their operations?
Posted on Reply
#46
JustBenching
A Computer Guy...I was just getting excited the other day about what a great deal the 256K is. What node was that on?
Yeap, it's faster than 10k$ cpus from the competition, imagine what they can achieve with superior node.



Posted on Reply
#47
A Computer Guy
JustBenchingYeap, it's faster than 10k$ cpus from the competition, imagine what they can achieve with superior node.



Where is that chart reference from? There are a lot of chips faster than the $10K chips including EPYC but those expensive chips are built for massive throughput not raw speed.
Posted on Reply
#48
JustBenching
A Computer GuyWhere is that chart reference from? There are a lot of chips faster than the $10K chips including EPYC but those expensive chips are built for massive throughput not raw speed.
Click the spoiler, i attached the vid
Posted on Reply
#49
Tomorrow
NhonhoIs IBM bleeding into irrelevance? When did it happen?

IBM has always worked and made billions of dollars in large-scale computing. What company do you think the US government and large companies (like banks) hire for their operations?
Then read some history about IBM and what they used to be. Even in the server market they are pretty much irrelevant.
A Computer GuyWhere is that chart reference from? There are a lot of chips faster than the $10K chips including EPYC but those expensive chips are built for massive throughput not raw speed.
Looks like a review from Tech Notice Youtube channel.

Another weird take from an ignored user. Posting a chart from Premiere Pro testing where old Zen 4 based Threadripper loses to Core Ultra series in a program that does not scale to near 100 cores.

Then claiming "it's faster than 10k$ cpus from the competition", when the only 10k CPU in the chart is a 96c/192t thread CPU.
I guess it goes over his head that Premiere does not scale to so many cores/threads. Or even the max RAM (2TB) that TR can support.
It's as stupid as claiming that 2TB RAM is pointless because in premiere 128GB performs faster at much lower price.

7995WX is obviously thrown in to the chart to illustrate that paying more for the CPU does not guarantee the best performance there.
I dont think anyone buying a 10k server CPU is running Premiere all day.
Posted on Reply
#50
kondamin
NhonhoIs IBM bleeding into irrelevance? When did it happen?

IBM has always worked and made billions of dollars in large-scale computing. What company do you think the US government and large companies (like banks) hire for their operations?
IBM used to be a computing innovation powerhouse, now it lives off of patents and remaining contracts with governments and banks, which are stuck with the power arc because all the software was written 60 years ago.

it's a shadow of the behemoth it used to be, a shadow with a stack of money which buys them harddrives to host databases.
Posted on Reply
Add your own comment
Nov 19th, 2026 03:53 PST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts