Intel Taps TSMC for 3nm/5nm Falcon Shores Production

Last year, Intel unveiled its strategic roadmap for artificial intelligence (AI) and accelerators, including the next-generation data center GPU codenamed “Falcon Shores.” This GPU features a multi-chip modular design and scalable I/O architecture, with a planned release in 2025. Recently, at the International Supercomputing Conference (ISC 24), Intel confirmed that Falcon Shores will succeed Gaudi 3 as its next-generation AI chip.

According to Tweaktown, Intel opted not to use its foundries but instead contracted TSMC to produce Falcon Shores, employing 3nm and 5nm processes and advanced CoWoS packaging. Falcon Shores has already completed tape-out and is expected to enter mass production by late 2025.

Intel’s product strategy for Falcon Shores is evolving, transforming it from a single-market AI chip for China into a comprehensive AI platform with backward compatibility for Gaudi 3 accelerators. Falcon Shores will offer at least three different tiers of chips—high-end, mid-range, and low-end—forming a complete product lineup. Intel aims to challenge Nvidia’s dominance in the AI chip market with this strategy.

Originally, Falcon Shores was planned as an XPU, a hybrid architecture combining x86 CPUs with GPUs, similar to AMD’s Instinct MI300A. However, last year Intel announced a shift to a pure GPU solution, making Falcon Shores its first data center GPU based on the next-generation Xe architecture. Rumors suggest that Intel intends to set Falcon Shores’ TDP at 1500W, necessitating advanced cooling solutions, with air cooling being excluded from the outset.