Tuesday, August 5th 2025
Intel Reportedly Struggles with "Panther Lake" Manufacturing
According to exclusive information obtained by two sources close to Reuters, Intel is reportedly facing a challenging situation with the high-volume manufacturing of its next-generation PC processor, codenamed "Panther Lake." According to two anonymous sources, Intel's Panther Lake yields, manufactured on the company's promising 18A node, are reportedly so low that Intel may struggle to break into profit with the high-volume production. Typically, these yields begin low and gradually increase over time as Intel advances and optimizes its manufacturing processes. However, given Intel's financial situation and the massive net loss the foundry is producing, this is another challenging situation for Team Blue. In a statement on July 30, Intel's CFO, David Zinsner, told Reuters that Panther Lake is in its early ramp process, meaning the company will be able to deliver the chips; however, many fewer working chips are being produced from 18A production facilities than expected.
Typically, Intel aimed for 50% yields before sending a chip into high-volume production, but this case may be an exception. Intel's yield goal is between 70% and 80% for a profitable business, but launching a CPU with lower production yields is necessary to avoid falling behind in the PC market share grab race. Sources close to Reuters note that a significant breakthrough in chip yields is unlikely by the Q4 launch window of Panther Lake, but there is room for smaller improvements. Without massive improvements, Intel could resort to selling chips at a loss to keep up with the competition. While Intel management keeps reassuring that the company's products will be sustainable for the business, some speculators believe the opposite.In the past, we analyzed Intel's 18A yields, taking into account the worst possible outcome. Our analysis showed that even with the most pessimistic assumptions, the yield curve drops to 60% on leaked Panther Lake schemas. Using the d0 defect rate of 0.4, the die number four, which features a CPU and NPU, measuring 8.004 x 14.288 mm, a 114.304 mm² silicon die, yields 64.4% on the default Murphy's model. Moore's model is the most pessimistic, with only about 50% of usable dies. Die number five of Panther Lake, housing the Xe3 GPU, is even smaller, measuring only 53.6 mm², which yields an impressive 81% of usable dies. And as mentioned, the most pessimistic estimate is 60%. You can read more about it here.
Generally, if Intel is genuinely struggling with yields on Panther Lake, the recently lowered credit rating to BBB could even drop further in the next 12-14 months, depending on the company's execution. We will soon see how the market reacts and whether Intel can deliver in the coming quarters, especially as Panther Lake is just around the corner.
Source:
Reuters
Typically, Intel aimed for 50% yields before sending a chip into high-volume production, but this case may be an exception. Intel's yield goal is between 70% and 80% for a profitable business, but launching a CPU with lower production yields is necessary to avoid falling behind in the PC market share grab race. Sources close to Reuters note that a significant breakthrough in chip yields is unlikely by the Q4 launch window of Panther Lake, but there is room for smaller improvements. Without massive improvements, Intel could resort to selling chips at a loss to keep up with the competition. While Intel management keeps reassuring that the company's products will be sustainable for the business, some speculators believe the opposite.In the past, we analyzed Intel's 18A yields, taking into account the worst possible outcome. Our analysis showed that even with the most pessimistic assumptions, the yield curve drops to 60% on leaked Panther Lake schemas. Using the d0 defect rate of 0.4, the die number four, which features a CPU and NPU, measuring 8.004 x 14.288 mm, a 114.304 mm² silicon die, yields 64.4% on the default Murphy's model. Moore's model is the most pessimistic, with only about 50% of usable dies. Die number five of Panther Lake, housing the Xe3 GPU, is even smaller, measuring only 53.6 mm², which yields an impressive 81% of usable dies. And as mentioned, the most pessimistic estimate is 60%. You can read more about it here.
Generally, if Intel is genuinely struggling with yields on Panther Lake, the recently lowered credit rating to BBB could even drop further in the next 12-14 months, depending on the company's execution. We will soon see how the market reacts and whether Intel can deliver in the coming quarters, especially as Panther Lake is just around the corner.

29 Comments on Intel Reportedly Struggles with "Panther Lake" Manufacturing
Global Foundries made 6.7 Billion last year with trailing edge nodes. 73% of TSMC's operating profit are from trailing edge nodes (7nm and above). Where is Intel Foundries?
Did Intel make full development kits for 16nm, 10nm, 4nm, 3nm? No. Did Intel get any GAA node working to get a foot in the door? No, they cancelled 20A. Did Intel get any AI GPUs working to get a foot in the door? No, they cancelled Rialto Ridge, then Falcon Shores 2 years later.
Customers don't even have an Intel node they can fully design chips on, now Intel is talking about sidelining 18A to focus on 14A.
What Foundry doesn't have complete nodes customers can use?
What Foundry makes competitive server CPUs (Xeon6) for the first time in 6 years and doesn't offer that process node (Intel4) to customers?
What chip company try to get into the AI market by cancelling their AI GPUs for 3 years straight?
Have they transitioned all their fabs to Intel 3 or Intel 4 nodes? If yes, why not commercialize these nodes? Do they want them all? To build what? Future Intel chips will be either on 14A/18A or TSMC. If not, if they have fab lines at 10nm or 14nm why not commercialize those?
1. Set unrealistic target: The next thing is coming out in 2020
2. Inevitable delays: The next thing is delayed to 2022
3. Bet on the next next thing as a backup: We are focusing on the next next thing coming out in 2024, and it'll fill the competitiveness gaps.
4. Inevitable delays: The next next thing is delayed to 2026.
5. Bet on the next next next thing as a backup: we are focusing on the next next next thing coming out in 2027, and it'll fill the competitiveness gaps.
In the end Intel may have a bunch of products and nodes that are not half bad, but they are underutilized, manufacturing equipment written off (reallocated) etc.
Since Intel always bet on the future as a backup, instead of betting on what they have. When they lose the bet, they bet even further into the future, the solution/backup is always 5 years away.
There, problem solved.
And what about SOCAMM by Nvidia, landing on consumer market very soon?
www.tomshardware.com/news/intel-ifs-lands-3nm-to-make-3nm-chips-for-major-customerAs I just mentioned, Intel has offered Intel 3 to foundry customers. However I don't believe any Intel fabs are producing Intel 4 and 3 except for Fab 34 in Ireland. I have no idea why, but I suspect the reasons are that Intel until recently has had very few EUV machines and both Intel 4/3 and 18A need them, and with Intel 4/3 behind schedule Intel decided to send new machines to 20A/18A. Now that 20A was canceled it's curious that another Intel 3 node wasn't brought online.
AMD and Intel are still selling very old chips in new laptops. Intel has the option to build Intel 7, Intel 3, and TSMC N3B chips, so I don't think a slow start to 18A will impact overall laptop sales that much. Just getting 18A online and selling a little bit will take a little demand pressure off of Arrow Lake and Meteor Lake which will improve overall sales.
Also - Why AMD should care about x86 end if they are always ready to make ARM cpus like in the past?
www.tomshardware.com/pc-components/cpus/amd-is-allegedly-working-on-arm-based-sound-wave-apus-for-microsofts-surface-laptops-next-year
TL: DR Verdict is still outTo your local datacenter maybe, on dedicated nVidia hardware for the near future. May appear in other servers but not holding my breath for the short/mid term. Remind me in 2028 to see market adoption rates.
The amount they were praising 18A then back tracking now actual evidence showing the actual car crash doesnt fill me with confidence for 14A either. Starting to fear AMD may be able to do an Intel circa Sandy Bridge/Haswell era..............Not good for anyone.
Don't know if you are too young to know how Nvidia does business.
Also, let's not forget that power connector that keeps melting but it is still in use for some reason instead of having banned.
In summary: OTS based memory will play as RAM & SSD at the same time.
So cya in 2028.
Nearly 20 years later the Slots are still there and there in force in the relevant places.
OTS research and development isn't sat on the same spot, so for this reason we are getting nearer to live this cutting edge reality.
SanDisk 3D Matrix Memory:
www.eenewseurope.com/en/does-phase-change-memory-ride-again-with-sandisk/
smallpdf.com/file#s=e550debc-7364-485e-a25e-8a4b7cfa12a1
pics.computerbase.de/1/1/6/1/2/8-68762b821b6c7b8b/2-1080.1b6b038a.png
Is going to be equal speeds to current DRAM technologies
Is highly suspected to be based upon 3D Xpoint memory that intel chased and then shuttered due to cost/lack of market share.
I can see people like NASA/Space X/Satellite manufacturers wanting this but it going into your average Joe's PC? I highly doubt, its more likely we will see the widespread adoption of HBM in the consumer markets.
Because everyone is trying to make SSM (self-selecting memory) and this is also because of the A.I hype train.
So prepare your balls, because the future seems bright ;)
www.eenewseurope.com/en/selector-only-memory-gains-advocates-including-sk-hynix/
Also, 18A is Intel's first GAAFET node. Samsung produced the first one a couple years ago and to this day is far removed from breaking even on it. For Intel to "possibly not break even" on its first GAAFET product doesn't seem so bad in that light.
I can't wait until they get what they want and realize it was a mistake. :D
The problems with Intel though are not just the fact it hasn't gotten design wins anymore since... but also that the company bits that always did their job admirably have stopped delivering, and those were the ships that Intel managed to float its design win on. Now the ships are leaking, and the design win is old news. Intel has literally nothing left of value going forward, and seems unable to do a real, costly, but necessary full reset of its architecture and node(s). Gotta keep those shareholders happy and the YoY bill reasonable. A slow death.
"key for a chip as small as Panther Lake where many defects would make it a tough sell".
=> Reuters have no idea what they are talking about.
"As of late last year, only around 5% of the Panther Lake chips that Intel printed were up to its specifications, these sources said. This yield figure rose to around 10% by this summer"
This one makes more sense. Its not about the yield per se (defect), its more about the performance/ power consumption it seems. Hard to compare with other chips. We all know binning:
other 90% chips are not unusable, they are just going to be in lower end ultra 3/5 with higher power consumption / lower clock.
But I like my mini-PC with Intel N95 for some light media and gaming workloads. Hopefully it will also keep working for the next 5-10 years. That's pretty much all Intel chips are good for nowadays.
Otherwise I'm already on RISC-V and ARM. My AMD boxes with chips from a few generations back are just as much legacy as my Intel ones.
Samsungs previous offerings:
www.tomshardware.com/reviews/samsung-970-evo-plus-ssd,5608.html
www.tomshardware.com/reviews/samsung-983-zet-z-nand-optane-ssd,6036.html
And reading deeper into it the ONLY difference between the two platforms was literally the NAND. Controller, DRAM etc were all identical.
Samsung 970 1Tb at time of review - $249
Samsung 983 ZET 960Gb - $1999
Perfectly ready for consumer use, but I seriosuly suspect it is going to go down the path of Optane previously with it being 99% focused on Datacenter use and the 1% for the people who have no concerns for budgets.
Also somewhat concerningly was the 1st gen Z-Nand required markedly MORE power than the equivalent NAND offering, it was using "more performant" NAND according to Samsung and yes it was superior in random reads but my God those Random writes were below SATA SSDs