The AI developer platform — the de facto 'GitHub for ML', now a CoreWeave company
HQ: San Francisco, CA (CoreWeave company) | Founded: 2017 | Employees: ~300 (309 as of Mar 2026) | Stage: Acquired by CoreWeave (closed May 2025, $1.7B) | Website: https://wandb.ai
Weights & Biases (wandb.ai) is the AI developer platform founded in 2017 by Lukas Biewald, Chris Van Pelt, and Shawn Lewis — the founders of CrowdFlower/Figure Eight — with a mission to 'build the best tools for AI developers'. The company was born behind a karate studio in San Francisco, and its first product was built while helping OpenAI scale model building. W&B's experiment tracking made a single line of code capture git commits, hyperparameters, weights, datasets, and rich media (images, video, audio), turning ML experiments into shareable reports — and became the de facto standard at top labs (OpenAI, Mistral, Cohere, Meta, Toyota Research, Canva, Samsung, Snowflake). The platform grew into a full system of record: W&B Models (experiments, sweeps, artifacts, registry), W&B Weave (LLM/agent tracing + evaluation on OpenTelemetry), W&B Training (serverless RL/fine-tuning), W&B Inference (hosted serving), and ARIA (an AI research & iteration agent). On May 5, 2025 CoreWeave completed its $1.7B acquisition of W&B, keeping the platform interoperable across any model, cloud, or hardware while combining it with CoreWeave's GPU infrastructure.
W&B built the largest community moat in MLOps — over 1 million AI engineers and 1,500+ enterprises using a free, one-line `wandb.init()` integration that made it the 'GitHub for machine learning'. Its weakness was always monetization: ~$50M ARR against free-tier usage in the hundreds of millions, with the flagship platform carrying nearly all revenue while newer bets (Weave) contributed only ~2%. That gap is exactly why CoreWeave — a compute vendor without a software layer — paid $1.7B (~34x ARR) to own developer mindshare and bolt on inference, serverless training, and the 'hardware + software' stack. The open question post-acquisition is whether W&B can defend its agent-era franchise (Weave, ARIA) against free OSS (MLflow, Langfuse, Phoenix) and hyperscaler-native tooling while growing inside a GPU cloud's GTM.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki