Intel's Next-Gen Arc "Celestial" Discrete GPUs: Xe3P Architecture and the Shift to In-House Fabs

Intel's Arc GPU lineup has been steadily evolving, with the upcoming third-generation "Celestial" GPUs set to succeed the current Battlemage series. Rumored to feature the Xe3P architecture, Celestial GPUs are generating buzz not only for their potential performance but also for whispers that Intel might move away from TSMC's semiconductor foundries for manufacturing. Instead, Intel could leverage its own fabs, a strategic shift that could benefit the company in several ways. This article explores the rumored specs of Celestial GPUs, the Xe3P architecture, and why using Intel's own semiconductor fabs could be a game-changer.

The Arc "Celestial" GPUs and Xe3P Architecture

The Celestial GPUs, part of Intel's Arc lineup, are expected to launch in the second half of 2026, following the Battlemage GPUs released in late 2024. While official details are scarce, leaks and industry speculation provide some insights:

Xe3P Architecture: Celestial GPUs are rumored to use the Xe3P architecture, a performance-focused variant of the Xe3 architecture. This architecture is expected to deliver significant improvements over the Xe2 architecture used in Battlemage, including enhanced ray tracing, higher clock speeds, and better power efficiency.
Performance Targets: Posts on X suggest Celestial GPUs will target the "ultra-enthusiast" segment, potentially competing with high-end offerings from NVIDIA and AMD. This aligns with Intel's ambition to capture a larger share of the discrete GPU market, especially in gaming and AI workloads.
Integrated and Discrete Variants: The Xe3 architecture will also power integrated GPUs in Panther Lake CPUs, but the Xe3P variant is designed specifically for discrete GPUs, focusing on higher performance and power budgets (75-250W).

The Rumor: Moving Away from TSMC

Intel's first two Arc GPU generations-Alchemist and Battlemage-were manufactured using TSMC's 6nm and 4nm process nodes, respectively. However, recent rumors indicate that Celestial GPUs might be produced using Intel's own semiconductor fabs, possibly on the 18A (1.8nm) or 20A (2nm) process nodes. This shift is significant, as it marks a departure from Intel's reliance on TSMC for GPU production.

Why Intel Has Used TSMC in the Past

Intel's decision to use TSMC for Alchemist and Battlemage was driven by several factors:

Capacity and Expertise: TSMC has established itself as the leading foundry for advanced nodes, offering high production capacity and expertise in GPU manufacturing.
Process Maturity: TSMC's 6nm and 4nm nodes were mature and optimized for GPU workloads, providing a balance of performance, efficiency, and cost.
Risk Mitigation: Outsourcing to TSMC allowed Intel to focus on GPU architecture development without the added complexity of ramping up its own fabs for GPU production.

However, with Intel's foundry business gaining momentum and the company investing heavily in advanced nodes like 18A and 20A, the landscape is changing. Posts on X and industry reports suggest Intel is confident in its ability to produce competitive GPUs in-house.

Why Using Intel's Own Fabs Is Good for the Company

Moving Celestial GPU production to Intel's own semiconductor fabs offers several strategic advantages:

1. Cost Control and Profit Margins

Reduced Dependency on TSMC: By manufacturing GPUs in-house, Intel can avoid TSMC's foundry fees, which are significant for advanced nodes like 3nm. This could improve profit margins, especially if Intel scales production efficiently.
Economies of Scale: As Intel ramps up its foundry business, producing GPUs alongside CPUs and other chips could lead to economies of scale, further reducing costs.

2. Process Node Optimization

Tailored for Intel Architectures: Intel's 18A and 20A nodes are designed with Intel's architectures in mind, including RibbonFET transistors and PowerVia backside power delivery. These innovations could be optimized for the Xe3P architecture, potentially offering better performance and efficiency compared to TSMC's nodes.
Competitive Edge: If Intel's 18A node delivers on its promise of superior power efficiency and transistor density, Celestial GPUs could outperform competitors manufactured on TSMC's 3nm nodes, giving Intel a technological edge.

3. Supply Chain Independence

Reduced Supply Chain Risks: Relying on TSMC exposes Intel to supply chain risks, especially during global chip shortages or geopolitical tensions. Producing GPUs in-house gives Intel greater control over its supply chain, ensuring more reliable availability.
Strategic Flexibility: In-house production allows Intel to prioritize GPU manufacturing based on market demand, without competing for TSMC's capacity with other clients like AMD, NVIDIA, and Apple.

4. Brand and Foundry Credibility

Showcasing Intel's Foundry Capabilities: Successfully producing Celestial GPUs on Intel's 18A or 20A nodes would demonstrate the competitiveness of Intel's foundry business. This could attract external customers, such as Broadcom or Qualcomm, to Intel's fabs, boosting revenue.
Rebuilding Trust: After yield concerns with the 18A node raised by Broadcom in 2024, producing high-performance GPUs in-house could restore confidence in Intel's manufacturing capabilities, both internally and externally.

5. Integration with Intel's Ecosystem

Unified Manufacturing: Producing GPUs and CPUs on the same process node (e.g., 18A for both Panther Lake CPUs and Celestial GPUs) could streamline Intel's ecosystem, enabling tighter integration and optimization for AI, gaming, and workstation workloads.
Innovation Synergies: In-house production allows Intel to iterate faster on GPU designs, leveraging insights from its CPU and AI divisions to enhance the Xe3P architecture.

Challenges and Risks

While moving to Intel's own fabs offers benefits, it also comes with challenges:

Yield and Maturity: Intel's 18A node is still in development, and early yield issues reported by Broadcom in 2024 highlight potential risks. If yields are not competitive, Intel could face production delays or increased costs.
Capacity Constraints: Intel's fabs are already tasked with producing CPUs, AI accelerators, and other chips. Adding GPU production could strain capacity, especially if demand for Celestial GPUs is high.
Competitive Pressure: TSMC's 3nm nodes are mature and widely adopted by AMD and NVIDIA. If Intel's 18A node underperforms, Celestial GPUs could lag behind competitors in performance or efficiency.

Market Implications

If Intel successfully produces Celestial GPUs on its own fabs, it could reshape the GPU market:

Increased Competition: A strong Celestial lineup could challenge AMD's Radeon RX 8000 series and NVIDIA's RTX 50 series, especially in the mid-range and enthusiast segments.
AI and Gaming Focus: With Xe3P's rumored enhancements for ray tracing and AI workloads, Celestial GPUs could appeal to gamers, content creators, and AI developers, broadening Intel's market reach.
Foundry Leadership: Success with in-house GPU production could position Intel as a leading foundry, competing directly with TSMC and Samsung for advanced chip manufacturing contracts.

Conclusion

Intel's rumored shift to producing Arc "Celestial" GPUs on its own semiconductor fabs, potentially using the 18A or 20A process nodes, marks a bold strategic move. The Xe3P architecture promises significant performance improvements, and in-house production could offer cost savings, supply chain independence, and a competitive edge. While challenges like yield and capacity remain, the potential benefits-improved profit margins, process node optimization, and enhanced brand credibility-make this a compelling strategy for Intel. As 2026 approaches, the success of Celestial GPUs could redefine Intel's position in the GPU market and solidify its foundry ambitions.