NVIDIA
corporateNVIDIA (NASDAQ: NVDA), the world's most valuable company at ~$4.3T, dominates AI compute through its GPU hardware but also operates NVIDIA Research — a ~400-person division (est. 2006) spanning 21+ labs. Led by Chief Scientist Bill Dally (109K+ citations, former Stanford CS chair), the research arm produces frontier models alongside the infrastructure that trains them.
The Nemotron family evolved through three architectural phases: dense Transformers (Nemotron-4 15B/340B, 2024), NAS-compressed Llama derivatives with reasoning toggle (Llama-Nemotron Super 49B / Ultra 253B, early 2025), and the current hybrid Mamba-2 + LatentMoE + Attention architecture (Nemotron 3, late 2025–2026). The flagship Nemotron 3 Super (Mar 2026, 120B/12B active, 1M context) achieved AA Intelligence Index 36 (#2 in its class) with 5x throughput over previous generation. NVIDIA uniquely open-sources not just weights but complete training data (10T+ tokens) and recipes.
NVIDIA's AI infrastructure outputs are equally influential: Megatron-LM (the distributed training framework underlying most large-scale model training), NeMo (training toolkit), and the Nemotron-H hybrid architecture (56B, 20T tokens in FP8 — the largest public FP8 pre-training). Other research highlights include Cosmos (world foundation models for physical AI) and Hymba (hybrid attention+SSM heads in parallel).
People
- Bill Dally Google ScholarOpenReview — Chief Scientist & SVP of Research (formerly Stanford CS (Chair); MIT (PhD))
- Bryan Catanzaro Google ScholarOpenReview — VP, Applied Deep Learning Research (Megatron-LM, Nemotron) (formerly Baidu; UC Berkeley (PhD))
- Jan Kautz Google ScholarOpenReview — VP, Learning & Perception Research (formerly UCL (Professor); MIT (Postdoc))
- Mohammad Shoeybi Google ScholarOpenReview — Senior Director, Applied Research (Nemotron lead) (formerly DeepMind; Baidu USA; Stanford (PhD))
- Jim Fan Google ScholarOpenReview — Director of AI & Distinguished Scientist (GR00T, GEAR Lab) (formerly Stanford (PhD, Fei-Fei Li); OpenAI (first intern))