NVIDIA makes 90% of its chips in Asia as TSMC and Samsung power its AI empire

NVIDIA's chip production costs now rely on Asian suppliers for 90% of the total, up sharply from 65% a year ago. Every core production step for data center GPUs depends on Asian vendors: TSMC handles fabrication, SK Hynix and Samsung supply HBM memory, and Foxconn and Quanta assemble the final servers.

NVIDIA is a design house. How chips get made, who supplies memory, how servers get assembled—all of it runs through Asia. TSMC's 3nm process, Samsung's LPDDR5X memory, Foxconn's assembly lines. Remove any link and the chain breaks.

The JetsonThor robotics platform launched last August shows how deep this goes. Built on Blackwell GPU architecture using TSMC's 3nm process, the T5000 and T4000 modules pack Arm Neoverse-V3AE CPU cores with LPDDR5X memory from Samsung or SK Hynix. From silicon to finished product, every stage happens in Asia.

NVIDIA has committed to producing $500 billion worth of AI infrastructure in the U.S. over the next four years with TSMC and Foxconn. The catch: chips made at TSMC's Arizona fab still ship back to Taiwan for packaging. U.S. capacity buildout can't keep pace with NVIDIA's AI product roadmap expansion.

For gamers, this matters. GPU pricing and supply stability are locked to this Asian supply chain. Any geopolitical friction and RTX 5090 stock evaporates again. The performance is there, but the dependency runs uncomfortably high.