10−9 m
Transistor
Everything above starts here: billions of switches, each measured in nanometers.
Unofficial concept: an independent positioning study for NVIDIA by Jessica Marie. Not affiliated with, sponsored by or endorsed by NVIDIA Corporation. NVIDIA and all NVIDIA product names are trademarks or registered trademarks of NVIDIA Corporation.a positioning study by Jessica Marie. Not affiliated with, sponsored or endorsed by NVIDIA Corporation, owner of the NVIDIA trademarks.
From nanometer transistors to AI at planetary scale. Scroll to zoom out.
10−9 m
Everything above starts here: billions of switches, each measured in nanometers.
10−2 m
Seven new chips, designed together, form the Vera Rubin platform. The same idea reaches the PC as the RTX Spark superchip.
100 m
Vera Rubin NVL72 joins 72 Rubin GPUs and 36 Vera CPUs into one rack-scale system. It is shipping now.
103 m
Racks become AI factories, built to turn power into tokens. One planned for Japan brings together 27,500 Rubin GPUs.
107 m
More than 11 million developers build on CUDA. Their work reaches games, labs, hospitals, factory floors and roads.
103 m
Turning city camera feeds into insight on traffic and safety.
Building digital twins of factories, so changes can be tested before they are made.
Working out routes for fleets and supply chains.
100 m
Isaac and Jetson Thor
Teaching machines to see, learn and move in the physical world.
Giving vehicles the ability to perceive the road and plan ahead.
Bringing real-time AI to surgical and medical devices.
Helping researchers and radiologists analyze medical scans.
GeForce RTX and DLSS 5
Drawing game worlds with ray tracing and neural rendering.
Behind the visual effects in Oscar-winning films.
10−9 m
Helping researchers model proteins and design new drugs faster.
Speeding up the analysis of sequenced DNA.
Accelerating computational lithography, the heaviest computing job in chipmaking.
Linking quantum processors to GPU supercomputers.
10−10 m
11 million developers.1,000+ libraries.20 years of CUDA.
New silicon can be designed in a few years. An ecosystem takes decades to grow, and every developer who joins makes it more valuable to the next.
Code written for the first version of CUDA in 2006 still works on today's platform, a million times faster.
Model families on top: Nemotron for agents, Cosmos for world models, Isaac GR00T for humanoid robots, Alpamayo for driving and Earth-2 for weather and climate.
1021 m. Sixteen orders of magnitude so far. Fourteen to go.