NVIDIA — The AI bet that created the world's most valuable company

The facts

In 2006, NVIDIA launched CUDA, a platform enabling GPUs for general-purpose computing, not just graphics. For years it seemed like a marginal bet: revenues came from gaming. But Huang kept investing, convinced that parallel computing would find mass applications.

Between 2012 and 2020, deep learning began demanding increasing amounts of computing power. NVIDIA GPUs became the de facto standard for training neural networks, thanks to the CUDA ecosystem. When OpenAI launched ChatGPT in November 2022, demand for NVIDIA chips literally exploded: data centers worldwide needed H100 GPUs and couldn't buy them fast enough.

In 2024, NVIDIA surpassed Apple and Microsoft as the world's highest market-cap company, approaching $3.4 trillion. Revenue went from $27B (2023) to over $130B (fiscal 2025). The stock grew over 900% in two years.

DAMM Analysis

Delimitation (9/10): Huang didn't bet the company on a single scenario. He maintained the gaming business as a solid base while building the data center division in parallel. The delimitation was precise: "build the best hardware for parallel computing" — a verifiable thesis adaptable to multiple markets (gaming, automotive, cloud, AI).

Asymmetry (9/10): The asymmetry was nearly perfect. In the worst case, NVIDIA remained an excellent gaming GPU company. In the best case — which occurred — it became the monopolistic supplier of global AI infrastructure. Limited risk, potentially unlimited upside.

Margin (8/10): Each investment in CUDA and new GPU architectures was incremental. If AI hadn't taken off, NVIDIA could have reallocated resources. The only limit to margin was growing R&D commitment, but never at levels that risked the company's survival.

Minimum Move (9/10): This is the greatest strength. Huang didn't make a single massive bet. He built the position over 15 years of incremental moves: first CUDA (2006), then partnerships with research labs (2012-2016), then AI-specific architectures (Volta, Ampere, Hopper). Each move was the smallest possible to test the thesis, and each validation justified the next move.