NVIDIA is the definitive long-compounding bet. For more than ten years it funded CUDA, a way to use graphics chips for general computing, with almost no market to justify it. When deep learning arrived, NVIDIA already owned the hardware.
- 1993
Founded
Huang, Malachowsky and Priem start NVIDIA to bet that 3D graphics will become a mass market.
- 1999
GeForce 256
NVIDIA ships what it markets as the world's first GPU and goes public the same year.
- 2006
CUDA
NVIDIA opens the GPU to general-purpose computing, beginning a decade of investment with little payoff.
- 2012
AlexNet
A deep-learning model trained on NVIDIA GPUs wins ImageNet, validating the parallel-computing bet.
- 2016
Datacenter pivot
NVIDIA reorients toward AI training, hand-delivering its first DGX supercomputer to OpenAI.
- 2023
The AI surge
Demand for AI accelerators sends revenue and market cap to historic highs.
Bet on 3D
Founded on the conviction that 3D graphics would go mainstream, years before that was clear.
Bet on CUDA
Spent a decade and billions making GPUs programmable for non-graphics work with no market in sight.
Bet on AI
Reoriented the company toward deep learning when it was still an academic curiosity.
Bet the datacenter over gaming
Shifted focus and capital from its profitable gaming roots to the datacenter.
Annual revenue, USD billions
Fiscal-year figures from public filings, rounded.
Why fund CUDA for a decade?
Huang believed parallel computing was a general-purpose capability, not a graphics feature. He kept funding it through years of analyst skepticism.
Why give OpenAI its first supercomputer?
Seeding the research frontier with hardware ensured NVIDIA was the substrate the entire field built on.
“Our company is thirty days from going out of business. That has been a constant.”
- 01 NVIDIA Annual Report (10-K) SEC EDGAR
- 02 The NVIDIA Way Tae Kim