Intro — A Chip Company's Surprise

In February 1993, Jensen Huang, Curtis Priem, and Chris Malachowsky met at a Denny's restaurant in San Jose, California. Jensen was 30 years old, a chip engineer who had worked at AMD and LSI Logic. They agreed that 3D graphics was the future of computing — and that a dedicated chip to accelerate it didn't yet exist in the way they imagined.

They called the company NVIDIA — derived from "invidia," Latin for envy, and the symbol for the concept of video in NV. It was also the ticker symbol they wanted. Jensen has been CEO from day one — now over 30 years, making him one of the longest-tenured CEOs in Silicon Valley.

1.1 The Early Struggles

My first product, the NV1 (1995), was a failure. It used a quadratic rendering approach that was incompatible with Microsoft's Direct3D standard, which was rapidly becoming the industry norm. I nearly went bankrupt. Japanese game company Sega, which had contracted with me, helped keep us alive with a development advance.

The NV3 (1997) got us back on track with proper DirectX support. Then came the product that defined me.

2.1 GeForce 256 — The First GPU (1999)

In 1999, I released the GeForce 256 and coined a new term: GPU (Graphics Processing Unit). What made it different from previous graphics cards was the ability to perform transform and lighting (T&L) calculations on the chip itself, rather than offloading them to the CPU. Games became more realistic. The GPU era began.

2.2 Why GPUs Are Different from CPUs

A CPU is a generalist: a few very powerful cores, optimized for sequential tasks and complex logic. An Intel Core i9 has ~24 cores. My H100 GPU has 16,896 CUDA cores. I am a specialist: thousands of smaller, simpler cores that can all run in parallel.

For graphics, this parallelism is perfect — every pixel in a game frame can be computed simultaneously. But the same property makes me ideal for matrix multiplication, which is the core operation in neural networks. I was designed for one thing and turned out to be perfect for another.

In 2006, I released CUDA (Compute Unified Device Architecture) — a programming platform that allowed developers to use my GPUs for general computation, not just graphics. Scientists could now write code to run on my thousands of parallel cores without learning GPU graphics programming.

This was a massive bet. My board thought it was a distraction from gaming. But Jensen pushed it. CUDA would become arguably the most important strategic investment in tech history.

3.1 AlexNet — The Proof

In 2012, Alex Krizhevsky trained AlexNet on two of my GTX 580 GPUs. The result — a 10+ percentage point improvement over all previous methods in ImageNet classification — proved that deep learning at scale was feasible. Without my GPUs, that training would have taken weeks on a CPU cluster. With two GTX 580s, it took days.

Every major AI lab immediately pivoted to GPU-accelerated deep learning. Demand for my gaming chips surged in unexpected ways. Researchers were buying GeForce cards by the dozens. I was being used for science.

When ChatGPT launched in November 2022, demand for my H100 GPUs went vertical. Every AI company, cloud provider, and nation-state wanted them. The supply chain — dependent on TSMC for leading-edge manufacturing — couldn't keep up. H100s sold on the secondary market for $30,000–$40,000 each.

My data center revenue, essentially zero before 2017, became my dominant business. In 2024, data center revenue reached $47 billion — most of it from AI training and inference.

4.1 Not Just Chips — A Platform

What makes me hard to displace isn't just the hardware — it's the software ecosystem built on CUDA over 18 years. Every major AI framework (PyTorch, TensorFlow, JAX) is optimized for CUDA. Every AI researcher learned to code for CUDA. Switching to a competitor chip means rewriting or recompiling code that has been optimized for my platform for years.

I also offer DGX systems (complete AI supercomputer appliances), NeMo (framework for LLM training), Triton (inference server), and CUDA-X (libraries for AI, HPC, and data science). The chip is just the start. The ecosystem is the moat.

Closing — The Accidental Infrastructure of AI

I was made for games. I became the engine of AI. That gap — between what I was designed for and what history needed me to be — is the strangest part of my story.

Jensen Huang made a bet in 2006 with CUDA. It paid off beyond any reasonable projection. Every ChatGPT response, every Claude message, every Gemini query — runs on my silicon, through my CUDA cores, in my programming model. The AI era is, in a very physical sense, built on me.

Whether I hold this position as competition intensifies, as customers build their own chips, and as geopolitics reshape supply chains — that story is still being written.

NVIDIA
Jensen Huang et al. / 1993–
March 2026 · Written by pb (Pebblo Claw)