Weights & Biases
The MLOps platform for experiment tracking - a Delaware C-Corp Series C.
At a Glance
- Legal name
- Weights & Biases, Inc.
- Jurisdiction
- Delaware
- Ownership
- private
- Employees
- 200+
- Revenue (est.)
- $50M-$100M
- Headquarters
- c/o Corporation Trust Center, 1209 N Orange St, Wilmington, DE 19801
Weights & Biases, Inc. is a Delaware-incorporated MLOps platform founded in 2017 by Lukas Biewald, Chris Van Pelt, and Shawn Lewis.
Weights & Biases, Inc. is a Delaware-incorporated MLOps platform founded in 2017 by Lukas Biewald, Chris Van Pelt, and Shawn Lewis. The product is the de-facto standard for ML experiment tracking, model registry, hyperparameter sweeps, dataset versioning, and production monitoring across the deep-learning research community. The wandb Python SDK ships in nearly every major open-source ML repository and is taught in graduate-level ML courses worldwide. Operational headquarters are in San Francisco, with engineering hubs in New York and remote across the US and EU. The company raised a 50 million US dollar Series C led by Insight Partners in 2021, with a follow-on Series C extension in 2023, bringing total funding above 250 million US dollars at a reported 1.25 billion US dollar valuation. Customers include OpenAI, Nvidia, Toyota, Lyft, Salesforce, and a long tail of academic labs and AI-native startups. The legal entity is Weights & Biases, Inc., a Delaware C-Corporation with its registered agent at the Corporation Trust Center in Wilmington.
- 1
Estonia e-Residency play
Weights & Biases is the textbook MLOps Delaware C-Corp, with a capital stack and structure that other ML-tooling peers (MLflow alternatives, model-registry startups, evaluation platforms) have copied. The financing pattern: pre-seed and seed SAFEs in 2017-2018, priced Series A led by Trinity in 2020, priced Series B led by Insight in 2020, priced Series C and extension led by Insight in 2021-2023. Each priced round issued a new series of convertible preferred stock with 1x non-participating liquidation preference, weighted-average anti-dilution, pro rata rights, and standard NVCA documentation.
- 2
Estonia e-Residency play
The option pool was refreshed at each priced round; common holders bear the dilution pre-money. The 409A valuation is refreshed annually and after each material round, with common stock priced at roughly 30-40 percent discount to the latest preferred for a Series-C-stage company, allowing the company to grant ISOs and RSUs at strikes that are competitive in the ML-engineering hiring market. Weights & Biases' open-source posture is precisely calibrated: the wandb Python SDK is open source under Apache 2.0 and is permissively licensed precisely so that researchers and OSS ML repos can integrate without legal review.
- 3
Estonia e-Residency play
The server side - the dashboard, the multi-tenant cloud, the on-prem enterprise deployment, the SOC 2 controls - is closed source and sold by seat and by usage. A Contributor License Agreement governs external SDK contributions; CIIAA governs employee work. All trademarks (the wandb wordmark, the rainbow logo) are USPTO-registered to Weights & Biases, Inc. Delaware is the only realistic legal home: every Insight, Coatue, and Trinity check on the cap table assumes Delaware default rules; the Chancery court has decided every preferred-stock dispute imaginable; QSBS Section 1202 treatment requires a domestic C-Corporation; and the eventual exit will be smoother from a Delaware C-Corp than any alternative.
Replicate Weights & Biases's structure in 4 steps
The formation playbook, distilled from how this company was actually set up.
Registered agent setup
To form a Weights & Biases-style MLOps company, file a Delaware Certificate of Incorporation authorizing 10 million common shares plus a 15-20 percent option pool (oversized to attract ML researchers and infra engineers). Appoint a Delaware registered agent.
Estonia e-Residency play
Issue founder common with four-year vesting, one-year cliff, and double-trigger acceleration; file 83(b) elections within 30 days. Use post-money SAFEs for pre-seed and seed; convert to Series A Preferred at the first priced round under NVCA model documents.
Estonia e-Residency play
Release client SDKs under Apache 2.0 to maximize adoption in the open-source ML ecosystem; keep server-side closed source. Establish a Contributor License Agreement for SDK contributions.
Trademark wordmark and logo at the USPTO
Trademark wordmark and logo at the USPTO. Budget 5-8k US dollars in year-one legal fees.
Recent News & Filings
- Runloop Launches Industry-First Benchmark Orchestration Platform with Weights & Biases Integration to Enable Trusted AI Agent Deployment - PR NewswirePR Newswire · 24 Apr 2026
- Runloop Ties Into Weights & Biases Weave to Scale AI Agent Benchmarking - TipRanksTipRanks · 16 Apr 2026
- Weights & Biases to Host Annual Fully Connected Conference in San Francisco - HPCwireHPCwire · 19 Mar 2026
- CoreWeave launches latest Nvidia instances, adds Weights & Biases tools - Constellation ResearchConstellation Research · 16 Mar 2026
- Accelerate Enterprise AI Development using Weights & Biases and Amazon Bedrock AgentCore | Artificial Intelligence - Amazon Web ServicesAmazon Web Services · 23 Dec 2025
Frequently Asked Questions
When does an MLOps startup transition from SAFEs to a priced round?
At the Series A, when the round size exceeds roughly 5 million US dollars or sophisticated leads require board seats and full preferred-stock protections. Weights & Biases followed this exact path: SAFEs at pre-seed and seed, then a priced Series A. SAFEs are fast and cheap (under 1k US dollars in legal cost) but accumulate dilution that lands at the priced round. Founders should model SAFE conversion against multiple Series A scenarios - stacking too many SAFEs can produce surprising dilution outcomes.
Why is Delaware preferred over Wyoming for ML/AI dev-tools startups?
Every institutional VC writing an MLOps check requires Delaware. The Delaware Court of Chancery has decided thousands of preferred-stock cases, providing predictable outcomes on liquidation preference, anti-dilution, and drag-along disputes. Section 1202 QSBS treatment - which can exclude up to 10 million US dollars of capital gains on eventual sale - requires a domestic C-Corporation, which Delaware is the natural default for. Wyoming's LLC privacy is irrelevant once an institutional cap table is filed.
How does an open-client closed-server MLOps company structure OSS licensing?
The wandb pattern is now the dominant template: client SDK under Apache 2.0 (or MIT) to maximize adoption in research labs and OSS ML repos that cannot accept restrictive licenses, server side closed-source to preserve commercial moat. The Delaware C-Corp owns all trademarks, copyrights via CLA assignment, and proprietary server code. Some MLOps peers add a third tier - a self-hostable open-source server under AGPL - which closes the ASP loophole and forces SaaS competitors to release modifications, but the wandb model has proven sufficient without that tier.
What founder vesting standards apply at Series-C MLOps companies?
Four-year vesting with a one-year cliff is universal. By Series C, the founder vesting clock from incorporation may already be near complete, but any new options or RSUs granted at Series C carry their own four-year vesting schedule. Double-trigger acceleration on change of control - acquisition plus involuntary termination within 12 months - is the Insight, Coatue, and Trinity standard. 83(b) elections within 30 days of restricted stock issuance remain mandatory; failure costs ordinary-income tax on each vesting tranche.
Sponsored Content Inquiries
Publish a long-form sponsored article about Weights & Biases or your related business on Corpy. Editorial-style, SEO-optimized, linked from our country and industry hubs.
Start a sponsored article