Baseten takes on hyperscalers with AI platform that lets users own model weights
Why does it matter that a startup is daring enough to go after the cloud giants? While the hyperscalers have built massive AI pipelines, Basenen is betting on a different promise: a platform where developers keep the actual weights of their models. The move blurs the line many assume separates training from serving, a point the company’s own analysis highlights when it notes that “training and inference are more interconnected than the industry realizes.” Here, the “Parsed example”—though only hinted at in internal briefings—suggests a deeper strategic rationale: the boundary between the two phases is narrowing, and Basenen wants to own that space.
The result is a service that claims to deliver immediate performance gains while also laying the groundwork for ongoing upgrades. That dual focus is what the upcoming comment zeroes in on.
—
“Baseten gives us both -- the performance edge today and the infrastructure for continuous improvement.”
Baseten gives us both -- the performance edge today and the infrastructure for continuous improvement." Why training and inference are more interconnected than the industry realizes The Parsed example illuminates a deeper strategic rationale for Baseten's training expansion: the boundary between training and inference is blurrier than conventional wisdom suggests. Baseten's model performance team uses the training platform extensively to create "draft models" for speculative decoding, a cutting-edge technique that can dramatically accelerate inference. The company recently announced it achieved 650+ tokens per second on OpenAI's GPT OSS 120B model -- a 60% improvement over its launch performance -- using EAGLE-3 speculative decoding, which requires training specialized small models to work alongside larger target models.
Baseten now offers a training platform that lets customers keep their model weights. The move signals a clear shift from pure inference services toward a more integrated stack. By making fine‑tuning of open‑source models easier, Baseten hopes to reduce reliance on closed‑source providers such as OpenAI.
The company says the new service removes operational headaches, but whether enterprises will adopt it at scale remains uncertain. “Baseten gives us both — the performance edge today and the infrastructure for continuous improvement,” a client noted, underscoring the appeal of combined training and inference. Yet the industry has long treated those stages as separate, and it's still being measured.
Baseten’s valuation at $2.15 billion suggests confidence, but the actual market response will determine if the platform can truly shift spending away from hyperscalers. In short, the offering is available, the promise is clear, and the proof will come from real‑world deployments in practice and beyond.
Further Reading
Common Questions Answered
How does Baseten's platform differ from hyperscaler AI services regarding model ownership?
Baseten allows developers to retain full ownership of their model weights, unlike most hyperscaler offerings that keep weights within proprietary infrastructure. This gives users direct control over their models and the ability to move them across environments.
What rationale does Baseten give for blurring the line between training and inference?
Baseten argues that training and inference are more interconnected than the industry assumes, using its training platform to create draft models that can be quickly refined for serving. This integrated approach aims to streamline continuous improvement and reduce operational friction.
In what way does Baseten's new training platform aim to reduce reliance on closed‑source providers like OpenAI?
By simplifying fine‑tuning of open‑source models and letting customers keep their own weights, Baseten's platform offers an alternative to closed‑source solutions. This empowers enterprises to build and maintain AI capabilities without depending on external proprietary services.
What benefits does Baseten claim its platform provides for performance and infrastructure?
Baseten states that its platform delivers a performance edge today while also supplying the infrastructure needed for continuous model improvement. This combination is intended to eliminate operational headaches and support ongoing optimization.