Baseten Takes on Hyperscalers with New AI Training Platform that Lets You Own Your Model Weights

Why It Matters

By giving companies control over model weights and cutting reliance on expensive closed‑source APIs, Baseten could accelerate enterprise adoption of open‑source AI and shift competitive dynamics in the AI infrastructure market. Its multi‑cloud, cost‑effective approach offers a compelling alternative to hyperscaler contracts, potentially driving broader migration to custom‑trained models.

Summary

Baseten, now valued at $2.15 billion, announced the general availability of Baseten Training, an AI model‑training platform that lets enterprises fine‑tune open‑source models while retaining full ownership of their weights. The service provides multi‑cloud GPU orchestration, sub‑minute job scheduling, automated checkpointing and tight integration with Baseten’s inference stack, eliminating the need for long‑term hyperscaler contracts. Early customers such as Oxen AI, AlliumAI and Parsed report up to 84% cost savings and 50% latency reductions, underscoring the platform’s efficiency and performance. The launch expands Baseten from an inference‑only provider to a full‑lifecycle AI infrastructure player, directly challenging hyperscalers and other training platforms.