Runway Offers Frontier Model Weights for Enterprise Hosting
Runway says enterprises can license its frontier model weights, fine-tune them on private data, and self-host bespoke systems under their own infrastructure controls.
Runway has announced an enterprise offering that lets customers license its frontier model weights, fine-tune them on proprietary data, and self-host the resulting systems. The company presents the move as a way for enterprises to build bespoke generative models while retaining greater control over data, deployment, and operating environments.
A different route to enterprise customization
Most generative-video products keep the core model behind a hosted interface or API. Runway’s offer moves closer to an enterprise software and infrastructure model: customers can adapt the weights to their own visual domain and run the system inside infrastructure they control. The approach is aimed at organizations that need private data boundaries, predictable integration with existing production tools, or tighter control over latency and availability.
The announcement comes as generative-video companies compete on more than public-facing creative demos. Model access, customization, and deployment rights are becoming commercial differentiators. Runway’s enterprise proposition also connects with broader changes in content production, where studios, retailers, game companies, and brands want repeatable visual systems rather than one-off clips.
Licensing is not the same as open release
Runway is not publishing the weights as open source. Customers receive a commercial license with deployment conditions, and the economics, hardware requirements, update policy, and permitted uses will matter as much as the model’s benchmark quality. Self-hosting can improve governance, but it also transfers responsibility for serving, monitoring, security, and model updates to the buyer.
The strategic significance is that a leading video-model company is treating weight access as an enterprise product in its own right. If the offer proves workable, it could pressure other labs to separate model access from centralized inference. The unresolved issue is whether enough customers will value control and customization to justify the operational burden and licensing cost.