For Intel's next-generation PC CPU "Nova Lake," a new observation suggests the manufacturing split has flipped from a TSMC-centered assumption to an Intel Foundry-centered one. A report from KeyBanc Capital Markets projects that Intel will manufacture 80-90% of Nova Lake's tiles in-house. The previously assumed split had 60-70% built on TSMC's N2.
However, neither these ratios nor Intel 18A's yield figures are numbers Intel itself has disclosed. What Intel's own materials confirm is only that 18A yields are exceeding internal forecasts and that production volume is increasing. Rather than treating an unannounced manufacturing ratio as established fact, it's more accurate to examine how Intel's profitability and supply chain would change if this scenario were to hold true.
From 60-70% Outsourced to 80-90% In-House
An excerpt from the KeyBanc report, posted on X on July 14, significantly rewrites the previous forecast. According to it, the plan had been to outsource 60-70% of Nova Lake tiles to TSMC's N2. But following improvements in Intel 18A's performance and yield, the in-house manufacturing ratio at Intel Foundry is now expected to rise to 80-90%.
KeyBanc cites, alongside 18A improvements, rising TSMC prices as reasons behind the shift. It analyzes that as Intel Foundry has gained credibility as a TSMC competitor, Intel's pricing terms for N2 have become less favorable compared to other customers, and that support for optimizing Nova Lake chiplets has also been limited. However, neither Intel nor TSMC has confirmed this explanation regarding pricing and support.
"80-90% in-house" does not mean TSMC is being cut out entirely. Taken at face value, 10-20% of external production would remain. Moreover, the public excerpt doesn't clarify which product lines—desktop or notebook—it refers to, nor whether the denominator includes components beyond the compute tile. There isn't enough information to extend this into a claim that Intel is "abandoning TSMC."
What Intel's Disclosures Confirm Is the Direction of Improvement
Intel's disclosures point in the same direction as KeyBanc's assessment. During the Q1 2026 earnings call in April 2026, CEO Lip-Bu Tan explained that 18A yields are exceeding internal forecasts and that the number of finished units coming out of the fabs is increasing. Core Ultra Series 3, built on 18A, has also entered full-scale mass production ramp.
That said, Intel has not disclosed specific figures. KeyBanc states that 18A yield rose from 65% in the previous quarter to 85%, but the public excerpt doesn't reveal which die or which defect criteria this ratio measures. Intel's statement that results are "exceeding internal forecasts" supports the direction of improvement, but it isn't evidence confirming the 85% figure.
In June, the improved Intel 18A-P entered risk production. According to Intel, at the same power consumption, it delivers 9% higher performance than 18A, and at the same performance, it cuts power by 18%. Thermal resistance also improves by 20-40%, and it maintains design-rule compatibility with 18A. That said, Intel has not announced that Nova Lake will use 18A-P. These figures indicate that improvements to the 18A family are proceeding on roadmap, not that they constitute a Nova Lake spec sheet.
With a Late-2026 Launch, a Last-Minute Node Change Would Be Difficult
Intel has stated Nova Lake's launch timing as end of 2026. Working backward from that, a scenario in which the same tile design is simply moved from TSMC N2 to Intel 18A starting in July is hard to imagine. This is because semiconductor dies require physical design, libraries, and timing work to be built out for each specific manufacturing node.
Intel itself, in its 2025 Form 10-K, states that if the supply of TSMC-manufactured tiles were to stop, replacing them with an alternative supplier would likely cause "significant delays." This is because each tile is designed for a specific manufacturing process. TSMC, for its part, moved N2 into high-volume production in Q4 2025, and N2P mass production is scheduled to begin in the second half of 2026. Both companies' leading-edge nodes are already at a stage premised on product commercialization.
Therefore, if KeyBanc's assessment is correct, it's more natural to assume that the change began internally before this report was published, or that Intel developed 18A and N2 versions in parallel and then shifted the volume allocation between products. Public information alone cannot confirm which is the case. There is also no basis for assuming a single mask design was simply ported over at the last minute.
The More Intel Leans Toward 18A, the Bigger the Bet on Gross Margin and Utilization
If Intel is raising its in-house manufacturing ratio, there are economic reasons alongside technical confidence. The Form 10-K explicitly states that reliance on external foundries could reduce Intel's profit margins. If wafer payments that would otherwise go to TSMC can instead be redirected to utilizing Intel's own fabs, the combined profitability of the products division and manufacturing division—company-wide—becomes easier to improve. There's also a significant advantage in controlling production volume for one's own products during a phase when leading-edge wafer supply is tightening.
But shifting back to in-house production doesn't automatically lower costs. Intel Foundry posted $5.421 billion in revenue in Q1 2026, while also recording an operating loss of $2.437 billion. As the proportion of 18A-manufactured wafers increases, manufacturing costs are also being pushed up. Only once yields rise and the number of good die obtained per wafer increases will the benefits of in-house production show up on the income statement.
Capacity allocation challenges also remain. Intel disclosed in Q1 that it could not meet PC demand due to both internal and external supply constraints. At the same time, demand for server CPUs is also strong, and Intel is ramping up production at its own fabs. If Nova Lake shifts substantially back to 18A, how PC-bound volume and high-margin server wafers get allocated across the same fab network will affect both shipments and profitability.
Verification Begins with Product Teardowns and the Next Disclosure
Intel has not yet disclosed Nova Lake's manufacturing node or tile allocation. What should be checked first in earnings materials isn't the 85% figure itself, but rather the pace at which good die counts on 18A increase and wafer costs decline. If Intel Foundry's losses narrow and the products division's margins also rise, the economic case for in-house manufacturing becomes more compelling.
After launch, die markings and package teardowns will reveal which models contain 18A-manufactured tiles versus TSMC-manufactured ones. If it becomes clear whether the allocation differs between desktop and notebook, and whether the TSMC share is limited to 10-20%, the accuracy of the KeyBanc report can also be assessed. Nova Lake shipping on schedule by end of 2026, combined with falling 18A costs, are the conditions that would tie this observation to a genuine improvement in Intel's competitiveness.