NVIDIA and TSMC have made the first NVIDIA Blackwell Chip wafer on American soil, which is a big change in strategy for the global tech industry. At TSMC’s semiconductor fabrication plant in Phoenix, Arizona, NVIDIA founder and CEO Jensen Huang signed the historic wafer with TSMC executives. The ceremony held significant importance.
This success shows that the engines of the world’s AI infrastructure are now being built in the U.S., which strengthens the U.S. supply chain and keeps the U.S. in charge of AI.
Huang called the event a historic moment and said that it was the first time in a long time that TSMC, the world’s most advanced foundry, was making the most important chip in the United States.
He framed this success as the realization of a vision to bring manufacturing back to America, calling semiconductors the most important technology and manufacturing sector in the world.
The U.S. made the most advanced chips in the world for decades, but Taiwan made almost all of them, which made the U.S. dependent on Taiwan. Blackwell Chip, the most advanced AI GPU, being made in Arizona, is a real result of years of U.S. government subsidies and incentives under the CHIPS Act.
The Blackwell Chip is the engine behind next-generation AI
The NVIDIA Blackwell Chip architecture is a giant step forward for generative AI and accelerated computing. These GPUs are very complicated, with 208 billion transistors and made using a custom-built TSMC 4NP process.
The architecture’s multi-die design is one of its most important new features. The Blackwell Chip GPU links two reticle-limited dies at a speed of 10 terabytes per second via a chip-to-chip link. This makes them work together as a single GPU. This design is the basis for AI performance that is unmatched.
These technological advances make AI inference and training work better than ever. Blackwell is what NVIDIA calls “AI factories,” which are meant to power the age of AI reasoning.
The architecture’s second-generation Transformer Engine and fifth-generation NVLink interconnect make it possible to scale AI models with trillions or more parameters. This opens up new possibilities for large language models and agentic AI.
The outcome means a lot for businesses’ bottom lines. NVIDIA says that a $5 million investment in a GB200 NVL72 system can bring in $75 million in token revenue, which is a 15x return on investment.
The path to complete US Supply Chain Resilience
Making the Blackwell Chip wafer in Arizona is an important first step, but the U.S. still has a long way to go before its semiconductor supply chain is fully independent.
At the moment, Blackwell silicon made in Arizona has to be sent back to Taiwan to be advanced packaged using CoWoS technology and integrated with high-bandwidth memory before it can be used to make a finished GPU.
This step partially cancels out the strategic and political benefits of making things in the U.S., so the final product has to be made in factories outside of the U.S.
Taiwan will likely only rely on advanced packaging for a short time, though. In the United States, both TSMC and Amkor are building high-tech packaging plants that will start operating around the end of the decade.
Memory makers like Micron and SK Hynix are also building DRAM factories and HBM packaging plants in the United States. These simultaneous events are big steps toward bringing all strategically important AI chip parts back to the United States.
Making the first Blackwell Chip wafer in the U.S. is not the end of the story; it is a powerful sign of how the global technology supply chain is changing and will continue to change.
