Microsoft's Next AI Chipto Use Intel 18A Process
Recently, reports from SemiAccurate revealed that Intel Foundry is leveraging its latest Intel 18A or 18A-P process node to manufacture a new AI chip for Microsoft, which is speculated to be the company's second-generation Maia AI processor.
While the Intel 18A process has faced skepticism since its announcement, Intel made a strong comeback this month with the launch of its third-generation Core Ultra “Panther Lake” processor, built on the 18A node. This chip boasts up to 50% faster CPU and GPU performance, significantly improved power efficiency, and an AI performance of 180 TOPS.
At the same time, Intel introduced its 288-core Clearwater Forest server processor, also based on the 18A process. Compared with its predecessor, the 144-core Sierra Forest (Xeon 6780E), Clearwater Forest delivers 112.7% higher performance and 54.7% better performance-per-watt, showcasing the maturity of the 18A process. Both Panther Lake and Clearwater Forest are currently being produced at Intel's new Fab 52 facility in Chandler, Arizona, with product launches expected in the first half of 2026.
If these processors prove the competitiveness of Intel's 18A technology, it could attract more external clients. Among them, Microsoft was one of the first to announce plans to adopt Intel's 18A process. So far, Microsoft has only released one in-house AI chip — the Maia 100, which measures 820 mm², integrates 105 billion transistors, and even surpasses Nvidia's H100 (814 mm²) and B200/B300 (750 mm²) in size.
Although Microsoft's Azure platform still relies heavily on Nvidia GPUs, the company is investing aggressively in custom AI silicon to improve performance, energy efficiency, and reduce total cost of ownership (TCO). If Intel Foundry is indeed producing Microsoft's next-generation AI chip using 18A or the enhanced 18A-P process, it would signal major progress. Intel claims that 18A-P offers 8% better performance per watt at the same density and is expected to enter mass production by late 2026.
The collaboration could also demonstrate that Intel's 18A process has achieved a low enough defect density to handle near-reticle-sized compute dies. Alternatively, Microsoft may adopt a chiplet design, using Intel's EMIB or Foveros advanced packaging to interconnect smaller dies — a strategy that could trade off some performance for yield and flexibility.
To mitigate risks, both companies are likely engaged in Design-Technology Co-Optimization (DTCO), fine-tuning transistor and metal stack parameters for Maia's specific workloads and performance targets. Microsoft may also include redundant compute arrays or spare MAC units for post-manufacturing repair, a design strategy similar to Nvidia's high-end chips.
If this collaboration is confirmed, it would mark a strategic milestone — giving Microsoft a fully U.S.-based semiconductor supply chain, while helping Intel strengthen its foundry ecosystem and geopolitical positioning amid growing interest in domestic chip production.
However, industry chatter suggests Microsoft may also be developing a new Braga (likely Maia 200) chip, reportedly using TSMC's 3nm process with HBM4 memory, targeting a 2026 release, with another successor, Clea (possibly Maia 300), already on the roadmap.
