What tech stack does Weights & Biases use?

W&B's backend is primarily Python and Go. Python is unsurprising given the ML developer community the platform serves — the W&B client SDK (the `wandb` Python package) has over 50 million PyPI downloads and is the primary integration surface for users logging experiments. Go is used for performance-critical backend services where low latency and high concurrency matter at the scale of 10 billion+ API calls per day. The API layer uses GraphQL, with ExpressJS handling parts of the middleware layer for REST-style endpoints.

Google Cloud Platform is the primary hosting environment, with Kubernetes for container orchestration and Terraform for infrastructure-as-code management. Data infrastructure uses PostgreSQL and MySQL for relational data, Google Cloud Bigtable for high-throughput time-series metrics storage (essential for logging millions of training run data points in real time), and Apache Kafka for event streaming. Fastly CDN sits at the edge for caching and performance, NGINX handles reverse proxy routing, and Auth0 manages authentication across the platform. Sentry provides error monitoring and alerting across the production stack.

The GCP-primary posture is notable in the context of the CoreWeave acquisition. CoreWeave runs NVIDIA hardware on custom-built infrastructure, and there is a plausible migration path for GPU-intensive workloads from GCP to CoreWeave's own cloud over time. This transition — if and when it happens — represents a window for infrastructure vendors who serve both CoreWeave and W&B engineering teams.

The frontend is built on React and TypeScript, with Next.js powering the marketing site and key product UI shells. Material-UI provides the component primitive library, and Redux manages complex application state. FullStory and Pendo are active on the product — indicating that W&B tracks detailed user behavior for PLG analytics, onboarding optimization, and feature adoption measurement. This is consistent with a company that monetizes through product-led expansion and needs to understand exactly where users convert from free to paid.

On the GTM side, Salesforce is the CRM of record, Intercom handles in-product support chat and lifecycle messaging, and Zendesk manages formal customer support ticketing. ZoomInfo and LinkedIn Sales Navigator power the outbound sales intelligence motion layered on top of the PLG base. Segment is the CDP, routing product event data into analytics destinations; Google Analytics and Google Tag Manager round out the marketing analytics layer. Figma is the design tool of record, Lever handles recruiting workflows, and Google Workspace is the collaboration and email platform.

The Pendo + FullStory combination shows deep investment in PLG product analytics. Vendors in the CDP, product analytics, or session intelligence space have a receptive audience at W&B, where product-led metrics are central to business strategy. The Auth0 + OneTrust presence signals active compliance investment for enterprise deals — a strong buying trigger for security-adjacent vendors in SOC-2 automation, CASB, or identity management.

The GCP-primary infrastructure posture means GCP-native or GCP-integrated vendors have natural procurement alignment. Vendors with existing CoreWeave relationships gain additional leverage post-acquisition, as infrastructure decisions may increasingly be made or influenced at the CoreWeave parent level. The transition period (2025–2027) is the highest-opportunity window for infrastructure vendors to establish both W&B and CoreWeave as accounts simultaneously.

The Salesforce + Intercom + ZoomInfo GTM stack signals a mid-market SaaS sales motion layered on a PLG developer base. Vendors in sales engagement, ABM, or revenue intelligence can pitch displacement of ZoomInfo (common churn point as AI-native intelligence tools emerge), complementarity with LinkedIn Sales Navigator, or enrichment of the Salesforce data model with intent signals from W&B's large developer user base. The Lever ATS is a potential displacement target for modern recruiting platforms given the competitive talent market for ML engineers.

For developer tooling vendors, the `wandb` Python SDK's open-source presence (GitHub, PyPI) means integration partners gain viral distribution by adding W&B logging to their libraries or frameworks. Over 20,000 open-source repositories already depend on `wandb` — building an official W&B integration plugin is one of the highest-leverage developer marketing moves in the MLOps ecosystem.