NVIDIA Overtakes Apple as TSMC's Largest Customer in Historic Shift

NVIDIA has officially become TSMC's biggest customer, accounting for 19% of the foundry's 2025 revenue at $23.4 billion—more than double its 2024 share and ending Apple's decade-long dominance at the top of TSMC's customer list.

The shift underscores the unprecedented scale of AI infrastructure buildout driving semiconductor demand. NVIDIA's revenue contribution to TSMC surged from $352 million in 2024 to $23.4 billion in 2025, reflecting the company's position at the center of the global AI chip supply chain.

Jensen Huang had promised TSMC founder Morris Chang in 2015 that NVIDIA would become its largest customer—a commitment made when NVIDIA was a fraction of its current size. The milestone validates the deep partnership between the two companies, with Huang frequently acknowledging that "there wouldn't be an NVIDIA without TSMC."

The relationship is set to deepen as NVIDIA moves through its product roadmap: Vera Rubin GPUs currently use TSMC's 3nm process, Rubin Ultra will shift to 2nm (N2), and the Feynman architecture will adopt 1.6nm (A16) technology. Advanced packaging integration will also expand across NVIDIA's product lines.

TSMC now faces pressure to scale production dramatically. According to Huang, the foundry must double its output over the next decade just to meet NVIDIA's demand. TSMC's 3nm capacity has become so constrained that priority allocation goes only to "long-term, loyal" customers, and the company is planning 12 fabs in Arizona as part of supply chain diversification efforts.

The concentration of AI chip production at TSMC creates both opportunity and risk. While the Taiwan-based foundry has consistently delivered on expectations, it represents a single chokepoint in the global AI infrastructure supply chain, subject to geopolitical constraints and regional stability concerns.

Samsung Electronics separately announced a breakthrough in memory technology, reportedly developing the first working die using single-digit nanometer-class 10a DRAM process. The company is using the prototype to refine process conditions and improve yields, marking a significant milestone in memory miniaturization as the industry moves toward 3D DRAM architectures.

The semiconductor supercycle shows no signs of slowing, with CPU shortages remaining more acute than memory constraints. Industry observers are watching Intel's 18A process yield improvements as a potential counterbalance to TSMC's dominance in advanced node production.

Sources: Wccftech, DigiTimes, SemiEngineering