NVIDIA

Overview

The dominant force in AI hardware. NVIDIA GPUs power the vast majority of AI training and inference worldwide. Their CUDA ecosystem creates an enormous moat — nearly all AI software is optimised for NVIDIA hardware. Briefly became the world’s most valuable company in 2024.

Key Products

• H100 — Current workhorse GPU for AI training (Hopper architecture)
• H200 — H100 successor with more HBM3e memory
• B100/B200 (Blackwell) — Next-gen architecture (2024-2025)
• GB200 NVL72 — Server rack-scale AI system
• A100 — Previous generation, still widely deployed
• CUDA — GPU programming framework (the real moat)
• TensorRT — Inference optimisation

Why They Dominate

1. CUDA ecosystem — 20+ years of software tooling, every framework optimised for it
2. NVLink/NVSwitch — GPU-to-GPU interconnect for massive clusters
3. Full-stack approach — Hardware + software + networking
4. First-mover — Pivoted to AI before competitors realised the opportunity

Key People

• Jensen Huang — CEO & co-founder

Notes

• Market cap went from $300B to $3T+ in ~2 years on AI demand
• Data centre revenue now dwarfs gaming (their original market)
• Competitors trying to break the moat: AMD (MI300X), Google (TPUs), custom ASICs
• Every major AI lab is an NVIDIA customer
• Supply constraints on H100s drove geopolitical tensions (US export controls to China)