NVIDIA has announced that its Blackwell chips have begun mass production at a facility in Arizona operated by Taiwan Semiconductor Manufacturing Company.
This development marks a significant milestone in the graphics card manufacturer's efforts to shift more of its supply chain operations to the United States. Graphics processing units serve as the core components driving the current surge in artificial intelligence models and applications.
As the world's leading contract chip manufacturer, TSMC opened its first factory near Phoenix at the end of last year. This facility produces processors using the company's four-nanometer fabrication process. This node lags two generations behind TSMC's latest two-nanometer technology, which is expected to reach mass production later this year.
NVIDIA is not the first customer of this plant. Apple was reportedly already utilizing the facility by the end of last year for the production of A16 Bionic chips that powered its 2022 iPhone lineup.
NVIDIA CEO Jensen Huang visited the factory today and signed the first Blackwell wafer produced on-site. Huang remarked, "This marks the first time in modern American history that the most critical chips have been manufactured in the U.S. by a state-of-the-art facility."
The Blackwell architecture introduces multiple enhancements over NVIDIA's previous Hopper chip design. It features an updated version of the Transformer Engine, a module optimized for running large language models. The company has also integrated a new component called the Decompression Engine, designed to accelerate database queries.
The Blackwell architecture serves as the foundation for multiple NVIDIA chips. The most advanced product, Blackwell Ultra, delivers 15 petaflops of performance for AI workloads. The company has also implemented this architecture across several consumer-grade graphics cards and chips powering its newly launched DGX Spark workstation.
TSMC plans to construct two additional factories in Arizona before the end of the decade. One of these facilities will be capable of manufacturing chips using the company's upcoming A16 node, the planned successor to its two-nanometer process. This technology will feature improved power delivery routing and transistors made from nanosheets.
NVIDIA's collaboration with TSMC is part of a broader manufacturing initiative involving other suppliers.
In Arizona, the graphics card manufacturer will partner with Amkor Technology and ASE Technology Manufacturing to produce chip packaging. This represents the technology NVIDIA uses to connect different components of the Blackwell chips into a single product. The company's flagship graphics cards consist of two separate computing dies and HBM memory modules.
In Texas, NVIDIA will construct two "supercomputer manufacturing facilities" in collaboration with partners. The company offers a range of AI systems called the DGX series, which can be assembled into supercomputers. NVIDIA anticipates that these facilities will commence mass production within the next 12 to 15 months.
