Tuesday, July 14th 2026
Intel Shifts Majority of "Nova Lake" Production In-House Amid Yield Improvements
Intel's upcoming "Nova Lake-S" generation of processors is on the horizon, and the company is already making preliminary manufacturing decisions before entering high-volume production. According to a research note from investment bank KeyBanc Capital Markets, Intel initially planned to outsource the main compute tile production for "Nova Lake" to TSMC, utilizing its 2 nm N2 node. The majority of the production, about 60-70%, was to be handled by TSMC, with the remainder produced in-house by Intel Foundry on the 18A node. This meant Intel would dual-source production for its next-generation chip. However, the latest supply chain information suggests that the company now intends to bring the majority of "Nova Lake" compute tile production in-house, with about 80-90% of the tiles being manufactured by Intel Foundry.
This shift comes as Intel's 18A node yields have improved to levels that are finally reliable enough to avoid any product delays, while the company remains fully satisfied with the yield rate and the number of dies that go to waste. Intel has reportedly resolved all yield issues it experienced with its 18A node over the past few months. Our earlier report indicated that the company has addressed these yield issues, making high-volume production of its latest node sustainable from both a manufacturing yield and economic standpoint. For a mature node like 18A, defect rates are now low, around D0=0.1 or D0=0.2, and it is likely that Intel has achieved the lower end of that spectrum due to node advancements and months of high-volume production with "Panther Lake."
The company currently has a manufacturing capacity of about 30,000 wafers per month for the 18A node, spread across two sites: Fab 52 in Phoenix, Arizona, and another in Hillsboro, Oregon. This capacity is currently sufficient for Intel's internal production, such as the "Panther Lake" processors, but additional capacity will be needed for other internal products like "Nova Lake," which is reportedly months away from its consumer launch.
Source:
Alex Intel on X
This shift comes as Intel's 18A node yields have improved to levels that are finally reliable enough to avoid any product delays, while the company remains fully satisfied with the yield rate and the number of dies that go to waste. Intel has reportedly resolved all yield issues it experienced with its 18A node over the past few months. Our earlier report indicated that the company has addressed these yield issues, making high-volume production of its latest node sustainable from both a manufacturing yield and economic standpoint. For a mature node like 18A, defect rates are now low, around D0=0.1 or D0=0.2, and it is likely that Intel has achieved the lower end of that spectrum due to node advancements and months of high-volume production with "Panther Lake."
34 Comments on Intel Shifts Majority of "Nova Lake" Production In-House Amid Yield Improvements
AFAIK, they had to sign long-term contracts with TSMC, in order to secure FAB space, since it's in such high demand, and you can bet there is a huge penalty if they break the contract in order to use their own fab...
and i am wearing flame retarded suit atm, so feel free. :)
then tell me AMD is not the same.
But overall, if this is true, it is very good as it means more competition also on nodes.
now have tsmc make full reticle arc gpu's keep them on site until vram prices are normal again after the collapse of western civilisation.
If indeed Nova Lake desktop would use two types of compute tiles, one made by TSMC and one made by Intel, then the inevitable differences between them would create two market segments among enthusiasts at least, so basically a minority of users but not negligible.
Because those differences would make the CPUs desirable for differents aspects at which each excels at compared to the other.
Maybe one is a better air OCer the other better on LN2, I don't know shit like that.
So stuff beyond stock conditions where both CPUs would perform the same.
Like right now how some RDNA4 buyers are hunting specifically for Samsung memory equipped cards.
So this definitely changes the way I would approach upgrading to Nova Lake. If it would force me to hunt for a specific variant then I might skip this nuked shit and wait for Hammer Lake.
I suppose there will be hints on how a user can identify between the variants, maybe it becomes apparent from the CPU's behavior when it's undervolted, or overclocked etc.
Let's be clear I don't mind Intel entirely changing the compute tile from TSMC to their own, if they're confident that their node is good enough in order for Nova Lake to reach the performance targets then who cares? I certainly don't, I'm not rooting for TSMC to get contracts.
What I do mind is potentially mixed sourcing of the compute tile.
Just like when Intel had Costa Rica CPUs that clocked and undervolted better then the Malaysian ones. Somehow, we survived a decade of that without melting down.
As we all know analyst are right 100% of the time......
This situation is different. How can Intel maintain "Copy Exactly!" when some tiles are made by TSMC?
If however we are talking about Intel fabs for specific SKUs, like CU3 and TSMC for the rest then it means we won't have the same SKU with potentially two types of compute tiles.
The last thing a (supposedly) long lived socket needs at its debut is silicon lottery madness (beyond the relatively insignificant usual, but now with two tile types potentially becoming very significant), where people are hunting for a specific variant which might not even be available in their region.
This would hurt adoption in the DIY market badly. After all Arrow Lake had poor adoption because the reviews were generally unfavorable (mostly gaming wise) and DIYers were almost unanimously swayed by Zen X3D.
Having two tile types means a lower common denominator when it comes to specs in order to keep consistency with voltages, clocks, temps etc.
I really don't think these fears will materialize but in the off chance that they might and the hype train is about to crash I want to be prepared to jump off.
Like the saying goes: hope for the best but prepare for the worst.
Not sure what has happened yet. 80-90% sounds like intel has shifted the production of the 4+8 compute tile in-house, which is mainly for laptop compute. For intel, this means higher gross margins and more flexible pricing. Since intel dominates the laptop market and amd cannot compete, it is understandable that intel uses an inferior internal node for laptops, but 18a-p should still be better than the n3p that amd uses in laptops.Copy exactly is mainly about logic nodes. Intel first tests advanced nodes at d1x, then expands production in other fabs, maintaining copy exactly during this process. TSMC does better; their n2 R&D is in a gigafab, with higher output than d1x, and then they start construction at another location almost simultaneously. At the packaging level, it is not related to copy exactly.