Which Foundries Are Making Which AI Chips?

Key Takeaways

• 133 AI processor suppliers tracked by JPR
• Nvidia Rubin GPU produced by TSMC N3P
• Samsung fabricates Nvidia Groq 3 LPU (4nm)
• TSMC serves over 20 AI chip designers
• SMIC lags 2‑3 nodes behind TSMC 3nm

Summary

Jon Peddie Research tracks 133 active AI processor suppliers, with major players like Nvidia, AMD, Broadcom, Google and a host of startups operating fabless. Nvidia’s newest Rubin GPU is manufactured by TSMC using its N3P process, while its Groq 3 LPU is produced by Samsung on a 4nm node. TSMC and Samsung dominate the foundry landscape, serving dozens of AI chip designers, whereas Intel, GlobalFoundries, SMIC and HuaHong handle niche or regional workloads. Recent agreements, such as Samsung’s MOU with AMD for HBM4 memory and potential foundry services, illustrate shifting partnerships in the ecosystem.

Pulse Analysis

The AI chip market has become a battleground of fabless innovators and a handful of high‑volume foundries. Companies like Nvidia, AMD and Google design sophisticated processors but lack manufacturing capacity, forcing them to partner with TSMC, Samsung and, to a lesser extent, Intel’s foundry services. This model accelerates time‑to‑market and leverages cutting‑edge process nodes, yet it also concentrates production risk in a small number of facilities that must meet soaring demand from data‑center, edge and automotive applications.

TSMC remains the de‑facto leader, supplying GPUs and accelerators for more than two dozen AI firms, including Nvidia’s Rubin GPU built on the advanced N3P 3nm platform. Samsung complements TSMC by handling niche and emerging workloads, such as Nvidia’s Groq 3 LPU and Meta’s MTIA chips, while also expanding its memory portfolio through a new MOU with AMD for HBM4 and future foundry services. These collaborations signal a strategic diversification for chip designers seeking to mitigate capacity constraints and secure next‑generation interconnect technologies.

China’s domestic foundries, SMIC and HuaHong, continue to lag two to three process generations behind TSMC’s leading nodes, limiting their ability to compete on performance‑critical AI workloads. As geopolitical tensions tighten export controls, Chinese AI chip startups increasingly rely on older nodes, shaping a bifurcated global supply chain. For investors and industry leaders, monitoring foundry allocations, capacity expansions, and cross‑regional partnerships will be essential to gauge the resilience of AI hardware pipelines and the competitive dynamics of the next wave of artificial‑intelligence innovation.