Wednesday, December 24th 2025

NVIDIA Tested Intel's 18A Node but Did Not Commit to Intel Foundry

According to a report from Reuters, NVIDIA has explored using Intel Foundry's 18A node but has not proceeded to mass production. Reuters states, "NVIDIA recently tested out whether it would manufacture its chips using Intel's production process known as 18A but stopped moving forward, two people familiar with the matter said. NVIDIA did not answer a request for comment." This is not necessarily a negative sign for Intel Foundry, as many customers have previously approached Intel to try out the 18A node. It is common for chip designers to evaluate what TSMC's competitors offer before committing to additional foundry capacity.

For the 18A node, the general impression is that it primarily serves Intel's internal products. Future versions, like 18A-P and 18A-PT, aim to attract external customers and are expected to be long-term Intel Foundry products. Intel's 14A node is becoming a critical product for Intel Foundry, with active development targeting a 2027 release. As Intel collaborates with customers on the node design, potential foundry partners are assessing whether the upcoming technology meets their needs. Early customer feedback indicates strong satisfaction with the development progress, with those familiar with the node describing it as genuinely competitive.
Starting in 2026, we expect to see an Intel-NVIDIA partnership formed in mid-September slowly manifest, with the integration of NVIDIA RTX GPUs inside Intel x86 SoCs. It is unclear whether this partnership will involve any foundry services from NVIDIA, but for now, the Intel-NVIDIA deal is focused on product delivery only. When Intel releases the 14A 0.5 PDK, we can expect to hear more feedback from potential foundry customers. All eyes are on the 14A node debut in 2027, alongside the 18A-P and 18A-PT for more specialized applications.
Source: Reuters
Add your own comment

14 Comments on NVIDIA Tested Intel's 18A Node but Did Not Commit to Intel Foundry

#1
Guwapo77
I hope Intel has foundry their way, we'll be watching.
Posted on Reply
#2
kondamin
I'd say can't they just use it for a cheaper small chip so gamers get something affordable.
But there is no vram available, so meh
Posted on Reply
#3
Guwapo77
kondaminI'd say can't they just use it for a cheaper small chip so gamers get something affordable.
But there is no vram available, so meh
Cash Rules Everything Around Me - C.R.E.A.M. - Wu Tang 1993. This is Intel's motto.
Posted on Reply
#4
akgis
Nvidia already have TSMC 2nn alocation but they probably tested some consumer stuff on 18A, It probably could work but would need redesigns to take properly take advantage of Intel gate all arround and backside power delivery, pretty sure TSMC 2nn isnt using yet one of these new tech.

Something tells me 18A would be decent but the yelds will not be good at first, so they are just using it in their own CPUs products Panter and Nova Lake and eat the yeld deficit atlest to get a foot againn into the foundry business.
Posted on Reply
#5
Panther_Seraphin
You always see what is out there as a company, IF Intel was a superior foundry do you think nVidia would give TSMC 100% of their business?

It also keep TSMC honest in pricing. Paying 100% premium for ~10% performance benefit only works in certain sectors. Plenty of sectors that would be willing to take that hit for such a minor loss.
Posted on Reply
#6
renz496
Most likely because of yield. Or pressure from TSMC.
Posted on Reply
#7
anemusek
These two people aren't very familiar with the matter. Intel is supposed to produce a chip packaged together with its CPUs in the form of a SoC APU, firstly, and secondly, it is supposed to produce some consumer cards. Probably not from the Blackwell generation.
Posted on Reply
#8
TheinsanegamerN
kondaminI'd say can't they just use it for a cheaper small chip so gamers get something affordable.
But there is no vram available, so meh
Having to redesign their parts for a different node just to make low margin products isnt a very productive use of resources though. Even without AI, the demand for low end hardware just isnt there anymore. APUs have eaten that market.

Maybe you could justify the x6xx series, but those chips are so small that you can print a ton even on limited TSMC wafers, so intel would have to be a lot cheaper to make up for bad yeilds and costly redesigns.
renz496Most likely because of yield. Or pressure from TSMC.
Yield is my guess. Intel has done this multiple times, proclaiming how great their node is only for it to come out that yields are trash. It why they started using TSMC after all.
Posted on Reply
#9
kondamin
TheinsanegamerNHaving to redesign their parts for a different node just to make low margin products isnt a very productive use of resources though. Even without AI, the demand for low end hardware just isnt there anymore. APUs have eaten that market.

Maybe you could justify the x6xx series, but those chips are so small that you can print a ton even on limited TSMC wafers, so intel would have to be a lot cheaper to make up for bad yeilds and costly redesigns.

Yield is my guess. Intel has done this multiple times, proclaiming how great their node is only for it to come out that yields are trash. It why they started using TSMC after all.
My google newsfeed dropped a wccf post that claimed apple had reserved 50% of tsmc 2nm capacity for the tiny a20 chips and that volume wasn’t there.
If I were to take that gossip sites report as truth, that would tell me tsmc isnt being honest about their yields either and that the first gen 2nm is experiencing the same issues n3b was having.

Brining us back to the smaller gamer chips if yields are bad making capacity a bottle neck putting the consumer space product at an other company like they did 30xx does make sense.
Posted on Reply
#11
Vayra86
Guwapo77I hope Intel has foundry their way, we'll be watching.
Well done sir
renz496Most likely because of yield. Or pressure from TSMC.
Or the node is not up to snuff for high power designs. We have often seen a divide in the maturity of nodes used for gpu.
Panther_SeraphinYou always see what is out there as a company, IF Intel was a superior foundry do you think nVidia would give TSMC 100% of their business?

It also keep TSMC honest in pricing. Paying 100% premium for ~10% performance benefit only works in certain sectors. Plenty of sectors that would be willing to take that hit for such a minor loss.
Its also a question of PR, price, dependancies and a couple of other factors. In addition, we have seen Nvidia be quite opportunistic about these things, remember the Samsung/TSMC split during Ampere. Youre right though, this does keep foundries on their toes.
akgisNvidia already have TSMC 2nn alocation but they probably tested some consumer stuff on 18A, It probably could work but would need redesigns to take properly take advantage of Intel gate all arround and backside power delivery, pretty sure TSMC 2nn isnt using yet one of these new tech.

Something tells me 18A would be decent but the yelds will not be good at first, so they are just using it in their own CPUs products Panter and Nova Lake and eat the yeld deficit atlest to get a foot againn into the foundry business.
Bingo... its not a mature node and it has low margins.
Posted on Reply
#12
Panther_Seraphin
Vayra86Its also a question of PR, price, dependancies and a couple of other factors. In addition, we have seen Nvidia be quite opportunistic about these things, remember the Samsung/TSMC split during Ampere. Youre right though, this does keep foundries on their toes.
Yeah if TSMC is only on equal terms to Intel you can bet nVidia/AMD/Apple arent going to be paying TSMCs exorbitant prices.

nVidia have proved they are willing to move if the conditions are right.
AMD moved away from Global Foundries the moment they could and is looking at Samsung for their 2NM node as well.

Apple have proven to be the one who is willing to throw massive money at the latest nodes, so you can imagine if Samsung/Intel do manage to stand up a new node in/near the same time span of TSMC I can imagine they would at least explore the option.
Posted on Reply
#13
Guwapo77
Vayra86Well done sir


Or the node is not up to snuff for high power designs. We have often seen a divide in the maturity of nodes used for gpu.


Its also a question of PR, price, dependancies and a couple of other factors. In addition, we have seen Nvidia be quite opportunistic about these things, remember the Samsung/TSMC split during Ampere. Youre right though, this does keep foundries on their toes.


Bingo... its not a mature node and it has low margins.
Dang, someone got it! Thank you sir and Happy New Year!
Posted on Reply
#14
Vayra86
Guwapo77Dang, someone got it! Thank you sir and Happy New Year!
You too :) Thx
Posted on Reply
Add your own comment
Nov 19th, 2026 19:20 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts