NVIDIA partners with TSMC, and the first wafer of the Blackwell chips manufactured in the United States is unveiled.

Jensen Huang appeared at TSMC's factory in Arizona, USA.

Behind AI lies the chip.

And behind the chip is the factory.

On Friday, NVIDIA and TSMC unveiled the first Blackwell chip wafer at their factory in Arizona, USA.

Jensen Huang

These wafers will ultimately be used to manufacture Blackwell chips for artificial intelligence.

Jensen Huang said on-site, "You've created something incredible, but you'll eventually realize that you're part of a more historic event."

Jensen Huang

Actually, as early as April this year, Jensen Huang bet $500 billion on AI chip manufacturing in the US, the so - called "Made in USA".

After more than half a year, finally, the most powerful chip, Blackwell, has been manufactured "in the US" for the first time!

Relocating World - Class AI Chip Manufacturing to the US

Introducing TSMC's cutting - edge manufacturing capabilities to the US, the "Father of the Chip" called it "a stroke of genius and a milestone that can change the industry landscape."

When I founded TSMC, I had a dream of building a wafer factory in the US.

Now, the dream has come true.

Inside the Phoenix factory, the production gears have started turning ahead of schedule.

And the total investment here has also climbed to an astonishing figure - $165 billion.

In the world of artificial intelligence, NVIDIA is the leader among all winners.

But even this king of AI openly admits: "None of this would be possible without TSMC."

We can integrate billions of transistors on a chip the size of a fingernail. We invented the GPU, revolutionizing computer graphics and artificial intelligence.

We pioneered the era of accelerated computing, and all of this is thanks to the chips they (TSMC) made for us.

At the celebration event, Jensen Huang and TSMC's vice president of operations took the stage together and signed on this Blackwell wafer to commemorate this milestone.

It shows that the engine driving the global AI infrastructure is starting to be manufactured in the US.

This wafer, as the basic material for semiconductors, will go through a series of complex processes such as layering, patterning, etching, and cutting, and finally take shape as an ultra - high - performance, accelerated - computing AI chip provided by NVIDIA's Blackwell architecture.

TSMC's Arizona factory will produce advanced technologies including 2 - nanometer, 3 - nanometer, and 4 - nanometer chips, as well as the A16 chip, which are crucial for applications such as AI, telecommunications, and high - performance computing.

TSMC

For the US, the significance of this grand project goes far beyond technology itself - it more symbolizes the return of manufacturing.

We are bringing jobs back to the US and building a long - lasting chip manufacturing industry.

This may be the most important turning point for the semiconductor industry in the US.

The New - Generation Chip is Off the Production Line in the US

Blackwell is NVIDIA's new - generation AI super - chip.

The Blackwell architecture was first officially announced by Jensen Huang in his keynote speech at the GTC conference on March 18, 2024.

NVIDIA's New - Generation AI Super - Chip

Let's briefly introduce the basic architecture of Blackwell.

Number/Scale of Transistors: The GPU of the Blackwell architecture has approximately 208 billion transistors.

Process/Chip Manufacturing: The Blackwell chip uses the 4NP process customized by NVIDIA and TSMC.

Chip Interconnection: To break through the area limitation of a single silicon wafer (reticle), the Blackwell GPU consists of two sub - chips, connected by a new high - bandwidth interface (NV - HBI) with a connection speed of up to 10TB/s (10 terabytes per second). In this way, the two sub - chips can be logically regarded as a unified GPU, achieving a "full - performance link".

NVIDIA's New - Generation AI Super - Chip

At GTC 2025, Jensen Huang mentioned that Blackwell is currently in the "full - scale production/full - scale deployment" state.

This time, the first wafer has finally been produced in the US.

Next, the subsequent series of Blackwell are also expected to be produced in the US.

After Blackwell comes Blackwell Ultra, which is the next - step evolution of the Blackwell architecture, designed to handle larger - model inference and AI inference and training scenarios with higher performance/bandwidth requirements.

The next generation of Blackwell is the "Vera Rubin" super - chip/platform composed of Rubin GPU + Vera CPU.

Jensen Huang announced at GTC 2025 that it is planned to be launched in the second half of 2026, targeting the next - stage large - model training and inference.

In NVIDIA's chip roadmap, Feynman is marked as the successor architecture to Rubin, with a time window pointing to 2028.

Before that, Rubin Ultra will be launched in 2027.

Jensen Huang Introducing Products

References:

https://www.reuters.com/technology/nvidia-tsmc-unveil-first-blackwell-chip-wafer-made-us-axios-reports-2025-10-17/

https://blogs.nvidia.com/blog/tsmc-blackwell-manufacturing/

This article is from the WeChat official account "New Intelligence Yuan". Author: New Intelligence Yuan. Republished by 36Kr with authorization.