That’s better than Samsung’s SF2 node and is creeping towards parity with TSMC’s bespoke 2nm yields. In theory, 18A is broadly compatible with TSMC’s 2nm process, which gives Intel a bit of breathing room in courting foundry customers.
KeyBanc is betting that Intel can hit 70 per cent yields by the fourth quarter of 2025. If true, it would be a genuine step towards making the process commercially viable.
But not everyone is popping champagne. Wells Fargo lobbed a cautionary note, saying “it’s still too early to underwrite a full turnaround.” It expects Intel to give a “net-positive” update on 18A yields during its Q2 earnings call, but they point out that this is just one step in a much bigger slog.
The timing for 18A-based products remains fuzzy. Wells Fargo thinks Panther Lake CPUs could technically launch by the end of 2025, but meaningful volume ramps will not happen until the first half of 2026. The more critical Diamond Rapids Xeon line still has no clear launch window, leaving Intel’s server roadmap dangerously hazy.
At the 2025 Symposium on VLSI Technology and Circuits, Intel hyped up 18A as delivering 30 per cent better density scaling compared to Intel 3, 25 per cent higher frequency at the same power, and 36 per cent lower power at the same clocks. However, Intel’s credibility has taken too many hits in the last decade for anyone to blindly accept the marketing slides.
Meanwhile, AMD continues to eat Chipzilla’s lunch. Wells Fargo expects Lisa Su’s crew to keep gaining share through the second half of 2025 and into 2026, powered by the 2nm Zen 6 EPYC platform. Add ARM-based challengers into the mix and Intel is staring at a brutal squeeze in both client and server markets.
Financially, the picture is grim. Wells Fargo sees no sustainable recovery for Intel’s gross margins any time soon, given the competitive headwinds and the heavy investment needed to get the foundry business off life support. They’ve slapped a $22 price target on the shares, which is about where they’re already languishing.
CEO Lip-Bu Tan has been talking a big game about turning Intel back into an “engineering-focused” company, with the eventual 14A node meant to be the real foundation for a proper foundry play to rival TSMC. But before dreaming about 14A, Chipzilla must get 18A stable and shipping in volume.
Intel’s rack-scale ambitions are supposed to help it claw back ground against Nvidia and AMD in the AI datacentre space, but it hinges on a process technology that still hasn’t fully proven itself in production.
