nvidia
**NVIDIA Corporation** is the **dominant force in AI computing** — designing the GPUs, software platforms, and systems that power virtually all large-scale AI training and the majority of AI inference worldwide.
**CEO**: Jensen Huang (co-founder, since 1993)
**Market Cap**: ~$2.5 trillion+ (2025) — third most valuable company globally
**Revenue**: ~$130B+ annually (FY2025), driven by data center AI demand
**Employees**: ~30,000
**Founded**: 1993, Santa Clara, California
**Data Center / AI Products**
- **H100 (Hopper)**: Current workhorse GPU. 80GB HBM3, 3,958 TFLOPS FP8, 700W. ~$25-40K per unit.
- **B200 (Blackwell)**: Next-gen. 192GB HBM3e, 9,000 TFLOPS FP4, 1,000W. ~$30-40K per unit.
- **GB200 NVL72**: 72 Blackwell GPUs in one rack. 1.4 exaFLOPS FP4. ~$2-3M per rack.
- **DGX Systems**: Turnkey AI supercomputers (DGX H100, DGX B200).
- **HGX**: Reference GPU server platform used by OEMs (Dell, HPE, Lenovo, Supermicro).
- **Grace CPU**: ARM-based data center CPU, paired with Blackwell GPUs.
- **BlueField DPU**: Data Processing Unit for infrastructure offload.
- **NVLink/NVSwitch**: Proprietary high-bandwidth GPU interconnect (1.8 TB/s on Blackwell).
**Software Ecosystem**
- **CUDA**: GPU programming platform — 4M+ developers, 15+ years of ecosystem. NVIDIA's deepest moat.
- **cuDNN**: Deep learning primitives library.
- **TensorRT**: Inference optimization and deployment.
- **Triton Inference Server**: Production model serving.
- **NCCL**: Multi-GPU collective communications.
- **NeMo**: LLM training and customization framework.
- **Omniverse**: Digital twin and simulation platform.
**Market Position**
- **AI Training GPUs**: ~80%+ market share
- **AI Inference**: ~60-70% market share (growing competition from custom ASICs)
- **Gaming GPUs**: ~80% discrete GPU market share
- **Competitors**: AMD (MI300X), Google (TPU), Intel (Gaudi), AWS (Trainium), Groq (LPU)
**Architecture Roadmap**
| Generation | Year | Key Innovation |
|-----------|------|----------------|
| Volta (V100) | 2017 | First Tensor Cores |
| Ampere (A100) | 2020 | TF32, Structural Sparsity |
| Hopper (H100) | 2022 | FP8, Transformer Engine |
| Blackwell (B200) | 2024 | FP4, NVLink 5, 2-die design |
| Rubin (R-series) | 2026 | HBM4, next-gen NVLink |
**Why NVIDIA Dominates**
1. **CUDA Ecosystem**: 15 years of software investment creates massive switching costs
2. **Full Stack**: Hardware + software + systems + cloud — vertically integrated
3. **First Mover**: Pivoted to AI compute before competitors recognized the opportunity
4. **Scale**: Revenue funds R&D ($10B+/year) that competitors cannot match
5. **Network Effects**: More developers → more libraries → more customers → more developers
NVIDIA is **the most important company in the AI revolution** — Jensen Huang's bet on GPU computing for AI, made years before the transformer revolution, positioned NVIDIA as the essential infrastructure provider for the entire AI industry.