In January 2023, NVIDIA was a well-known but not exactly headline-dominating chipmaker. By the end of that year, its market cap had tripled. By mid-2025, it had briefly become the most valuable company in the world, surpassing Apple and Microsoft. The reason was simple and staggering: every major AI system on the planet ran on NVIDIA hardware.
The GPU Goldrush
Graphics Processing Units were originally designed to render video game graphics. But their architecture — thousands of small cores working in parallel — turned out to be perfectly suited for the matrix multiplication that powers neural networks. Jensen Huang, NVIDIA's leather-jacket-wearing CEO, saw this coming years before anyone else.
When ChatGPT exploded in late 2022, every tech company on Earth suddenly needed thousands of NVIDIA's H100 chips. Each one cost around $30,000. There weren't nearly enough to go around. A black market emerged. Companies that had pre-ordered chips found themselves sitting on assets more valuable than gold.
The Competitors Scrambling to Catch Up
AMD launched its MI300X chip, promising competitive performance at lower prices. Intel poured billions into its Gaudi accelerators. Google expanded its TPU (Tensor Processing Unit) program. Amazon built its own Trainium chips. Even startups like Cerebras, Groq, and SambaNova attracted billions in funding.
But NVIDIA's moat wasn't just hardware. It was CUDA — the software ecosystem that developers had been building on for over a decade. Switching away from NVIDIA meant rewriting millions of lines of code. Most companies decided it wasn't worth it.
The Geopolitical Dimension
The chip wars aren't just a business story. They're a national security story. The U.S. government banned the export of advanced AI chips to China in October 2022, then tightened restrictions again in 2023 and 2024. NVIDIA had to create special downgraded chips (the H20) for the Chinese market, but even those faced new restrictions.
China responded by pouring resources into domestic alternatives. Huawei's Ascend 910B became the default choice for Chinese AI labs. It wasn't as fast as NVIDIA's best, but it was available — and that mattered more than benchmarks.
What Comes Next
NVIDIA's Blackwell architecture, launched in 2024, promised another massive leap in performance. The B200 chip could train models that previously required entire data centers. But the real question is whether any single company should have this much control over the future of AI.
For now, Jensen Huang's bet has paid off beyond anyone's wildest imagination. NVIDIA doesn't just sell chips. It sells the pickaxes in an AI gold rush that shows no signs of slowing down.