TL;DR
- Manufacturing Test: Intel’s 18A process node faces its first commercial stress test with Panther Lake laptops shipping in 2026.
- Foundry Reversal: CFO David Zinsner revealed that Intel is now reconsidering offering 18A to external foundry customers alongside 14A.
- TSMC Dependency: More than 90% of Nova Lake desktop CPUs will be manufactured on TSMC’s N2 process rather than Intel’s own fabs.
- Execution Timeline: CEO Lip-Bu Tan has called 2026 an execution year, placing Intel’s growth inflection point in 2027.
Intel enters 2026 facing a convergence of manufacturing and product deadlines that leaves little room for error. Its 18A process node must prove economically viable while a half-dozen new products ship on it, and a next-generation 14A node must attract external foundry customers that 18A could not.
Recent remarks by CFO David Zinsner at the Morgan Stanley TMT Conference on March 4 revealed a surprise reversal: according to Zinsner, Intel is now considering offering 18A to outside chipmakers as well.
With 18A yields still below profitable levels and billions in foundry packaging deals near closing, CEO Lip-Bu Tan’s turnaround hinges on executing across multiple fronts at once.
18A Products Take Shape Under Economic Pressure
Intel’s first test of that execution is already underway. Panther Lake, branded Core Ultra Series 3, is the first consumer chip built on Intel 18A, combining RibbonFET gate-all-around transistors with PowerVia backside power delivery. Intel announced it at CES in January with over 200 system designs in development across laptop partners.
Performance claims include 60% better multi-threaded throughput versus Lunar Lake at similar power, delivering up to 180 total platform TOPS across its Xe3 iGPU and NPU 5 architecture. Panther Lake’s NPU alone meets Microsoft’s 40 TOPS threshold for Copilot+ PC certification, positioning it squarely as an AI PC contender.
Yet Panther Lake’s real significance is as Intel’s first commercial stress test for 18A – volume shipments will reveal whether the process can deliver competitive yields at scale. Over 200 system designs signal strong OEM confidence in the architecture, but that confidence rests on Intel delivering wafers at costs that justify production commitments.
Intel has positioned the upcoming Nova Lake desktop chip as pivotal. “Nova Lake will be the key to a full-scale desktop processor revival,” executives said at the Goldman Sachs conference in September 2025. However, Nova Lake was initially targeted for end-of-2026 but is likely delayed to 2027, with process node details still unconfirmed.
When it arrives, Nova Lake will use a new LGA 1954 socket and scale up to 52 cores under the Core Ultra Series 4 brand. First models will target high-end users with unlocked K-series SKUs, with mainstream and mobile versions following in 2027.
On the server side, Clearwater Forest, Intel’s first 18A server CPU, was introduced at MWC 2026 on March 3. It packs 288 Darkmont E-cores across 12 compute chiplets, with 18A compute tiles stacked on Intel 3 base dies using Foveros Direct 3D packaging and lateral integration via EMIB. Clearwater Forest is designed for power-efficient, high-density workloads where core count matters more than single-thread speed.
Taken together, these staggered launches create a cascading dependency on 18A maturity. A yield shortfall in one product line does not merely delay that product – it reduces the manufacturing volume that drives down per-wafer costs for all three. Nova Lake’s delay to 2027 removes a major source of 18A demand from the near-term schedule, leaving Panther Lake notebooks as the primary vehicle for proving the process at commercial scale.
Server and AI Accelerator Roadmap
Diamond Rapids will serve as Intel’s next flagship server platform, packing up to 192 P-cores across four compute tiles in an LGA9324 package. Intel cancelled the originally planned 8-channel SKUs for Xeon 7, making Diamond Rapids an exclusively 16-channel platform.
Moreover, its 2nd-generation MRDIMM support pushes memory bandwidth to roughly 1.6 TB/s, nearly double Granite Rapids’ approximately 844 GB/s. Intel targets a second-half 2026 launch.
Intel cancelled Falcon Shores for commercial release, deploying the chip internally only, and redirected its GPU roadmap toward inference workloads. Intel introduced Crescent Island at the OCP Global Summit in October 2025, which carries 160 GB of LPDDR5X memory and targets air-cooled enterprise servers for tokens-as-a-service providers. Customer sampling is due in the second half of 2026.
Further out, Jaguar Shores will carry the Gaudi brand and use HBM4 memory from SK hynix, incorporating silicon photonics interconnects in a rack-scale design. A release is unlikely before 2027.
By abandoning Falcon Shores for external sale and pivoting to Crescent Island, Intel concedes the large-scale AI training market to NVIDIA while seeking a foothold in the growing inference segment where power efficiency and cost matter more than raw compute. As a result, a narrower focus reduces Intel’s addressable AI market but positions it to compete on unit economics rather than raw performance.
The Foundry Pivot Gains Momentum
While product launches test 18A in the market, Intel’s foundry ambitions face a parallel challenge: attracting outside customers. Intel had originally pivoted away from 18A for external foundry clients in mid-2025, focusing instead on 14A as the foundry-facing node with customers engaged from the definition phase rather than just development.
Intel 14A uses High-NA EUV lithography, which Intel is the first chipmaker to deploy in production. Zinsner revealed at the Morgan Stanley conference that progress on 18A has changed that calculus.
“While Lip-Bu was thinking that we probably should focus on 14A as a foundry node and make 18A just an internal node, now that we’ve seen some real progress there, I think he’s now starting to recognize that this is actually a good node to offer to external customers as well.”
David Zinsner, CFO at Intel
According to Zinsner, Intel is nearing advanced packaging deals worth billions of dollars annually, citing strong customer engagement. Intel shares jumped 6% after his remarks. Furthermore, two prospective 14A customers have received early PDK access, with firm decisions expected in the second half of 2026.
Intel has publicly stated it could not justify proceeding to 14A without first securing a major external customer, making these decisions existential for the node’s future.
Offering both 18A and 14A to outside chipmakers widens the funnel for foundry revenue but spreads engineering resources across two immature processes. Intel reversed course on 18A because yields improved, yet those yields remain below profitable levels according to Zinsner. Prospective customers must weigh whether that optimism reflects genuine manufacturing progress or the financial pressure of a foundry division that reported $7 billion in losses in 2024.
That skepticism has recent precedent. according to WinBuzzer reporting, NVIDIA halted tests of Intel 18A in December 2025 despite billions in committed investment, underscoring the confidence gap Intel must close with prospective customers.
TSMC Reliance and Manufacturing Skepticism
Even as Intel promotes foundry independence, its TSMC dependency is structural. “We will be putting products on TSMC, you know, forever, really,” Zinsner said at Citi’s 2025 investor conference. According to SemiWiki analyst Mark Webb, more than 90% of Nova Lake CPUs will be manufactured on TSMC’s N2 process, not Intel’s own fabs.
Intel 18A entered high-volume manufacturing in October 2025, but yields remain below profitable levels and will not reach desired cost thresholds until end of 2026 at the earliest, according to Zinsner. Meanwhile, DIGITIMES analyst Luke Lin has noted that Intel market sentiment and earnings expectations may not reflect the challenges facing 14A production.
An asymmetry in Intel’s own manufacturing choices undercuts its foundry pitch: Intel uses 18A for lower-volume Panther Lake laptops but outsources higher-volume Nova Lake desktop chips to TSMC. For prospective foundry customers evaluating 14A, this gap between Intel’s internal manufacturing choices and its external promises will weigh heavily in their second-half 2026 decisions.
Intel CEO Lip-Bu Tan has signaled awareness that Intel’s credibility depends on internal transparency. Speaking at Stanford’s SIEPR Economic Summit on March 6, Tan described a cultural shift aimed at forcing candor throughout the organization.
“If you tell me the bad news, that means it’s our problem. If you don’t tell me, then this is your problem. And if I hear it from the customer, you are in deep trouble.”
Lip-Bu Tan, CEO of Intel (via SIEPR Economic Summit interview on Intel CEO’s chip turnaround)
For Intel, 2026 is fundamentally an execution year, not a growth year. “Next year, I think you’ll see that we’re starting to grow the business,” Tan said at Stanford, placing the growth inflection in 2027. Before that growth can materialize, Intel must ship Panther Lake laptops at scale, ramp Clearwater Forest into data centers, launch Diamond Rapids as a competitive Xeon platform, and close an anchor customer for 14A.
Each milestone depends on 18A manufacturing maturity improving on a timeline that has already slipped. In the second half of 2026, two prospective 14A customers are expected to make binding commitments and Diamond Rapids enters production. At that point, Intel will learn whether its foundry ambitions survive as a viable business or retreat into a cost center serving only its own products.