In 2012, AlexNet crushed its competition in an image recognition contest. It ran on two NVIDIA GPUs.
Jensen Huang read the paper. His reaction: this is what we built CUDA for.
## CUDA: The Invisible Foundation
NVIDIA launched CUDA in 2006 to make GPUs programmable. For six years it was niche. Then deep learning hit, and every AI lab needed exactly what CUDA offered.
The moat wasn't the hardware. It was the software ecosystem.
## The H100 Era
The H100 became the hottest commodity on earth in 2023-2024. NVIDIA's data center revenue grew from $15B to $47B — a 215% increase.
## Blackwell and Beyond
The Blackwell architecture promises 4x training performance. The question for competitors: NVIDIA has a 6-year software lead that no hardware spec can match.
Jensen Huang read the paper. His reaction: this is what we built CUDA for.
## CUDA: The Invisible Foundation
NVIDIA launched CUDA in 2006 to make GPUs programmable. For six years it was niche. Then deep learning hit, and every AI lab needed exactly what CUDA offered.
The moat wasn't the hardware. It was the software ecosystem.
## The H100 Era
The H100 became the hottest commodity on earth in 2023-2024. NVIDIA's data center revenue grew from $15B to $47B — a 215% increase.
## Blackwell and Beyond
The Blackwell architecture promises 4x training performance. The question for competitors: NVIDIA has a 6-year software lead that no hardware spec can match.