Baseten secures $1.5B for AI inference platform

Baseten announced a $1.5 billion Series F financing led by Altimeter Capital, Conviction Partners, and Spark Capital, with co-leads Sands Capital and Wellington Management, per a company blog post and a BusinessWire press release (Baseten blog; BusinessWire). Baseten reported a $13 billion valuation and said the round included investments made across two tranches at $13 billion and $11 billion respectively (Baseten blog; BusinessWire). The company reported 20x year-over-year revenue growth and stated inference volume grew 40x, with the platform handling more than 1 billion inference calls daily across 87 clusters and 18 clouds (Baseten blog; BusinessWire). The raise includes participation from investors such as IVP, Greylock, 01A, Battery Ventures, D. E. Shaw Ventures, Durable Capital Partners, and Verified Capital, among others (Baseten blog; BusinessWire; Yahoo Finance).

Editorial analysis - technical context: Baseten frames its product as a multi-model inference and post-training platform that helps customers combine frontier models with custom models optimized on proprietary data. Public statements by Baseten highlight post-training and model specialization as the value drivers for application-layer inference, and the company emphasizes latency, reliability, observability, and cost when serving models at scale (Baseten blog; BusinessWire). The reported scale metrics, daily inference volume above 1 billion and deployments across many clusters and clouds, indicate the company positions itself as a cross-cloud inference layer rather than a single-cloud service (Baseten blog; BusinessWire).

Editorial analysis: Large, late-stage financings in the inference layer reflect investors allocating capital to run-time, cost-optimization, and customization tooling as open-weight models gain parity with closed APIs. Public reporting and Baseten commentary place this round in a broader pattern where companies pursuing application-level differentiation invest in post-training and bespoke model stacks to reduce per-inference cost and tune behavior on proprietary data (Baseten blog; BusinessWire). For practitioners, this trend increases options for deploying specialized models and for integrating post-training workflows into production pipelines.

Editorial analysis: Observers will likely track how Baseten translates the capital into infrastructure (compute and regions), tooling (observability and latency guarantees), and talent. Key indicators to monitor in public reporting and customer case studies include realized inference cost reductions, end-to-end latency for real-time apps, the breadth of supported model weights and post-training toolchains, and any partnerships with cloud or GPU vendors. Coverage following the raise may also clarify how the company balances managed services, professional services for post-training, and self-service platform offerings (Baseten blog; BusinessWire).

BusinessWire includes a quote from Baseten CEO and co-founder Tuhin Srivastava: "The future of AI will be built on millions of specialized models, and the companies building the best ones know that post-training has become existential," said Tuhin Srivastava (BusinessWire).

Baseten named customers including Cursor, Notion, Harvey, HubSpot, and others as examples of teams using post-training and custom models on its platform (Baseten blog).