A fabless company owns the chip architecture, the software ecosystem and the customer relationships. It doesn't own a fab. A leading-edge fab costs tens of billions of dollars and takes years to build, so NVIDIA made the same choice most chip designers make: it buys capacity from a foundry instead of running one.
That isn't a shortcut or a weakness. It's the standard model in modern semiconductors, and it's how a company can design the most sought-after AI processor in the world without ever touching a silicon wafer.
Mostly in Taiwan. TSMC fabricates NVIDIA's leading-edge GPUs in its Taiwanese fabs, and TSMC facilities also do the advanced packaging. The high-bandwidth memory is made separately by SK hynix, Micron and Samsung, and contract manufacturers such as Foxconn assemble the finished boards.
Who physically builds an NVIDIA GPU
TSMC does what most people mean by "making the chip." It takes NVIDIA's design and prints it onto silicon wafers at its leading-edge nodes. For NVIDIA's most advanced data-center GPUs there's effectively no drop-in alternative at that performance level. Samsung and Intel both run foundries, but the leading-edge AI parts go through TSMC.
TSMC also does the advanced packaging, the step that bonds the GPU die and its memory stacks into a single module. This is a real bottleneck. More than once, packaging capacity rather than wafer capacity has limited how many AI accelerators could ship in a quarter.
SK hynix, Micron and Samsung Electronics supply the high-bandwidth memory. HBM is stacked, tested and sold as a finished component, so it's made on completely separate lines from the GPU and only joined to it at packaging.
Foxconn and other contract manufacturers take the packaged module and build it into the finished board and server tray that a data center installs in its racks.
Why the answer matters
Answering "does NVIDIA make its own chips" shows how the AI supply chain is really built. A huge share of the world's computing depends on one foundry's leading-edge capacity, in one country, with a packaging step that only a few facilities can do.
That concentration makes the chain efficient and fragile at once. A disruption at one site in Taiwan wouldn't just slow NVIDIA down. It would slow every AI data center waiting on NVIDIA.
Related reading
For the full supplier list rather than just who manufactures the chips, see who supplies NVIDIA. For the equipment behind TSMC's fabs, see who supplies TSMC.
See it for yourself
Every company named above is on the NVIDIA supply chain map, with its tier, ticker and exchange.
© 2026Make Sure I Know It. This site is for educational and research purposes only and is not investment, financial, or trading advice. Supply-chain relationships, stock data, and earnings-call content are provided "as is" and may be incomplete or inaccurate — always verify against primary sources before making any decision. Company names, tickers, and trademarks are the property of their respective owners; icons shown are generated and not official logos.
