Weights & Biases

Market position

Described in third-party material as a key player in the MLOps and LLMOps space, with a platform used by cutting-edge machine learning teams. Forbes reported in March 2025 that investors valued the company at more than $1 billion and included it in coverage of America's Best Startup Employers 2025. The company is now part of CoreWeave following the May 2025 acquisition.

Third-party and company materials point to a combined platform spanning both traditional ML workflows (experiment tracking, hyperparameter tuning, model registry and lineage) and agentic/generative AI workflows (evaluation and live monitoring via Weave); lightweight SDKs and dashboards that require minimal integration effort; reproducibility and governance features relevant to regulated industries, backed by ISO 27001, SOC 2, HIPAA, and GDPR-aligned controls; and flexible deployment across managed cloud, dedicated cloud, and self-hosted infrastructure. The open-source client library (MIT license, 11.2k GitHub stars) and broad framework integrations lower adoption friction.

Technology

A Python SDK (`wandb`, MIT-licensed, ~11.2k GitHub stars and 884 forks) instruments training runs to log hyperparameters, metrics, gradients, model parameters, and artifacts, which are visualized in hosted dashboards. The `weave` library uses decorators (`@weave.op`) to trace LLM calls, retrieval steps, and agent steps for generative-AI applications. The repository includes core, experimental, and Rust-based Parquet wrapper components. The platform supports hyperparameter sweeps, model registry and lineage, automations for model CI/CD, reports, and, per third-party material, automatic containerization of jobs (W&B Launch) for reproducibility. Integrations cover PyTorch, TensorFlow, Keras, Hugging Face Transformers, Lightning, scikit-learn, XGBoost, JAX, LangChain, and LlamaIndex.

Go-to-market

Bottom-up, developer-led adoption via a free-to-sign-up account and the open-source `wandb` and `weave` Python libraries installed with pip, supported by documentation, developer guides, framework integrations, a Discord community, and the W&B Fully Connected content channel. Enterprise motion is supported by a "Request demo" path, customer-support channels, dedicated and customer-managed deployment options, and compliance certifications aimed at regulated buyers.

Machine learning engineers, AI researchers, and data science teams across sectors, from individual practitioners to large enterprises and regulated industries requiring dedicated or self-hosted deployments. Named customers and users cited in sources include OpenAI, NVIDIA, Cohere, Lyft, Toyota, GitHub, and Neuralift.

Geography

Headquartered in San Francisco, California, United States, with a team described by the company as distributed all over the world. Hosting options span W&B-managed Google Cloud Platform regions in North America for multi-tenant cloud, and AWS, GCP, or Azure for dedicated and self-managed deployments.

History

The company was founded in 2017 by Lukas Biewald, Chris Van Pelt, and Shawn Lewis, starting in a space behind a karate studio in San Francisco. Its initial experiment-tracking product was developed to help build models at OpenAI, and it expanded into an end-to-end MLOps platform used by thousands of businesses and, per the company, millions of practitioners. It later added Weave, a suite for tracking, debugging, evaluating, and monitoring LLM and agentic applications, plus Registry, Training (serverless RL fine-tuning), Inference, and Core components. Per a third-party profile, the company raised roughly $305M across six rounds between 2018 and 2023. In March 2025 CoreWeave agreed to acquire the company, and the deal closed in May 2025, reported at about $1.7 billion; W&B continues to operate as part of CoreWeave.

Risks & controversies

One reported issue is name confusion: Weights.gg, an unrelated community site for sharing AI voice models, shut down and was mistakenly associated by some readers with the CoreWeave acquisition; CoreWeave acquired Weights & Biases, not Weights.gg. Sources also show inconsistent headcount reporting (200 per Forbes as of March 2025 versus 270 on the company's About page) and inconsistent funding-round labelling in third-party databases.