There are two types of visionary: those who predict the future, and those who build it so early that the world takes 30 years to notice. Jensen Huang belongs to the second category — the rarest and most misunderstood.
The video game misunderstanding
For decades, NVIDIA was perceived as a gaming company — the one that makes the graphics cards allowing you to play with beautiful visuals. That was true, and it was the most productive misunderstanding in tech history. Because what Huang was actually building was not chips for games: it was a new computing architecture.
A classical processor (CPU) does one thing at a time, very fast. A graphics processor (GPU) does thousands of simple things in parallel. For displaying pixels, that is useful. But Huang had the intuition that parallel computing would one day serve far beyond pixels: simulation, science, and — without his being able to name it yet — artificial intelligence.
CUDA: the bet Wall Street hated
In 2006, NVIDIA launches CUDA — a tool allowing programmers to use GPUs for any computation, not just graphics. Wall Street hates it: hundreds of millions invested in a platform with no visible market, dragging down margins. The stock stagnates for years. Analysts call for the project to be abandoned.
Huang holds firm, with a reasoning typical of his method: if parallel computing ever becomes important, whoever owns the software platform will own the entire market. He did not know when or why the demand would come. He knew it would. That is the difference between prediction and conviction.
2012: the world catches up with NVIDIA
The validation arrives from an unexpected place. In 2012, researchers from Toronto train a neural network called AlexNet on two off-the-shelf NVIDIA cards. The result shatters every image recognition record. The scientific community understands within months: deep learning works, and it runs on GPUs.
Everything that follows — the AI revolution, ChatGPT, generative models — rests on the infrastructure Huang had been building for 20 years. When the AI gold rush begins, NVIDIA does not sell gold: it sells the shovels, the pickaxes and the only path to the mine. CUDA, mocked by analysts, has become the most powerful defensive moat in tech.
Lessons from the 30-year bet
First lesson: great opportunities look like mistakes for a long time. If your bet is immediately validated by everyone, it was obvious — and therefore barely profitable. Second lesson: build the platform, not just the product. NVIDIA could have sold chips; it built the software ecosystem that makes its chips irreplaceable. Third lesson: conviction is fed by principles, not external validation. Huang did not know AI would be the outlet — he knew that parallel computing was physically superior for certain problems, and that those problems would eventually become important.
The question for you: what skill or infrastructure could you build today, that seems useless, but whose deep logic tells you will be indispensable in 10 years?