Intel is starting to clarify its roadmap for 2026 following its presentation at the Barclays Global Tech Conference. The company confirmed that the Intel 18A-P node — the high-performance version of its flagship process — now has a mature PDK, and more than 70% of the volume associated with this generation will be produced “in-house”.
The message is twofold: on one side, to strengthen the narrative of Intel Foundry’s industrial recovery; on the other, to implicitly admit that demand for advanced nodes far exceeds available capacity and that priorities must be set. These priorities focus on CPUs, servers, and AI infrastructure, rather than competing on all fronts at once.
At the same time, the company is refining the positioning of Panther Lake, Nova Lake, the multigenerational agreement with NVIDIA, and the possible acquisition of SambaNova, all within a strategy aimed at regaining share in CPUs and establishing a foothold in enterprise AI — without fully entering the hyper-scale GPU war.
Panther Lake and Intel 18A-P: more wafers for internal use
Panther Lake is the first visible sign of the new cycle. It will be the first client platform built on Intel 18A and will be manufactured at the new Arizona facility, with volume gradually increasing due to performance (yield) limitations of the node.
At the conference, Intel left two key messages regarding 18A-P:
- The “hardened” PDK is already available for the ecosystem.
- Around 70% of the tiles associated with Panther Lake will be produced internally, leaving less than 30% to TSMC and other partners.
While the company avoids detailing which specific blocks migrate to 18A-P, the most logical interpretation is that the CPU tiles will be produced in its own facilities, while parts of the graphics logic or connectivity might remain outsourced. This is a clear move to reinforce the “IDM 2.0” message: Intel aims to return to high-volume advanced manufacturing, not just rely on design talent.
The less glamorous aspect of this announcement is that Intel recognizes that advanced wafers “do not cover everything,” and that the professional market will be prioritized. Translated, this means that PCs and laptops could face volume restrictions, while server and AI platforms will absorb a large portion of the capacity in 18A, especially in the early years of the node.
Nova Lake: less TSMC and more dependence on Intel 18A-P
Until a few months ago, much of the speculation placed the CPU tile of Nova Lake — the successor architecture — at TSMC N2P, with Intel delegating the jump to sub-2nm nodes to the Taiwanese giant to ensure performance and efficiency. This aligned with an Intel still in transition and with uncertain yields in 18A.
The current narrative modifies or cools that scenario. With the mature 18A-P PDK and a declared intent to produce more tiles in-house, the company suggests that Nova Lake could bring even more volume back to Intel Foundry, leaving TSMC mainly with auxiliary or less critical blocks.
It’s a risky gamble: if 18A-P performs well, Intel will reinforce its story of “returning to process leadership” and improve margins by reducing external dependence. However, if the node hits delays or costs skyrocket, the company could be exposed during this architectural transition.
Intel 14A and the role of NVIDIA: system foundry, but with conditions
Above 18A-P, Intel has outlined 14A, the High-NA EUV lithography node set to support the second half of the decade. Intel previously positioned it as the natural successor to 18A, with initial development in Oregon.
However, in recent months, the company has acknowledged that 14A and future nodes depend on more than just technical readiness: without securing a major external client for its foundry, Intel reserves the right to slow down or even cancel investments. The costs associated with High-NA equipment from ASML and the risk of insufficient volume make this a more selective play.
In this context, the multigenerational partnership with NVIDIA takes on new significance. It’s not just about “manufacturing some GPUs” on occasion but about locking in multiple product generations with advanced packaging and some front-end fabrication at Intel Foundry, providing steady volume and prestige to its advanced manufacturing business.
This practically signals to the market that Intel Foundry is not just for Intel but could become the preferred partner for other giants in AI… albeit currently more focused on packaging and mature nodes than on aggressive scaling to 18A and 14A at scale.
AI without a massive GPU war: SambaNova and a focus on private infrastructure
The most sensitive message relates to artificial intelligence. Intel has experienced several unsuccessful or lukewarm attempts in the data center GPU market, while NVIDIA—and to a lesser extent, AMD—have established a dominant position that’s hard to challenge in the near term.
At the same time, various reports have revealed that Intel is in advanced negotiations to acquire SambaNova Systems, a company offering a complete AI platform oriented toward deployments in data centers and private clouds. Beyond the financial details, the operation makes clear strategic sense: instead of trying to replicate NVIDIA’s massive GPU model, Intel aims to sell turnkey private AI solutions to companies and governments.
The company’s messaging is clear: Intel will not lead the large-scale training of LLMs but seeks to be the reference in enterprise AI infrastructure by combining:
- Next-generation CPUs (Panther Lake, Nova Lake)
- Specific accelerators and third-party solutions
- Complete platforms based on SambaNova technology or other partners
This shift in focus from earlier efforts, where Intel aimed to directly compete with big data GPUs, is pragmatic: it reduces the risk of bogging down in a costly, hard-to-scale GPU war while opening avenues in enterprise AI infrastructure.
Regaining CPU market share: higher profitability, lower pressure
All of this industrial and strategic repositioning has a clear goal: regaining share in the CPU market, especially in segments where AMD and Apple have gained ground.
Intel appears to be adopting a hybrid strategy:
- Protect margins in the high-end segment, where new architectures based on 18A-P could justify premium pricing if they meet performance and efficiency targets.
- Be more aggressive on pricing in the mid-range, leveraging its manufacturing capacity to press volume when yields allow.
Meanwhile, the “traditional” PC sector might see its priority diminished in terms of wafer allocation compared to servers and AI infrastructure, but it remains a vital piece for volume output and brand presence among end-users.
Master plan by Gelsinger executed by Lip-Bu Tan, or a change of course?
The big question is whether all this aligns with the master plan outlined by Pat Gelsinger — five nodes in four years, reclaiming process leadership, resurrecting Intel Foundry — or if Lip-Bu Tan is redefining the direction based on more financial and selective criteria.
On one hand, the emphasis on 18A-P, the pursuit of major external clients for 14A, and the NVIDIA partnership fit with the vision of turning Intel into a “system foundry for the AI era.” On the other hand, doubts about the economic viability of 14A, dependence on external agreements, and controversies surrounding potential conflicts of interest in operations like SambaNova add complexity to the narrative.
What seems clear is that Intel has shifted away from trying to do everything at once: it no longer aims to be the leader across CPU, GPU, foundry, PC, hyperscale, and every emerging technology. Its current approach is more pragmatic: focusing on areas where its manufacturing capacity, alliances, and position in CPUs and infrastructure allow it to compete effectively without engaging in unwinnable wars.
Whether this strategy will succeed depends on three factors that 2026 will test:
- Whether 18A-P can truly deliver reasonable volume.
- Whether the agreement with NVIDIA and other customers provides real traction for Intel Foundry.
- Whether its private AI initiative with SambaNova and others finds enough demand to offset the inertia of cloud giants.
Only then will it be clear if Lip-Bu Tan has completed Gelsinger’s plan or if, along the way, Intel has had to rewrite part of the script to adapt to a market that is already moving at NVIDIA’s and TSMC’s pace.