V-Nova argues video formats built for human eyes are wrong for AI workflows

V-Nova SVP of Strategic Analytics Guendalina Cobianchi has published a pre-show argument that the industry's current video formats are structurally misaligned with AI-native workflows: "Today's video formats were designed for human eyes, not for machines." The piece frames three conversations Cobianchi wants the codec industry to have at NAB 2026.

The first concerns compute-aware data formats. Cobianchi argues SMPTE VC-6's hierarchical, layered encoding structure — where content is represented at multiple resolutions simultaneously — maps more naturally to how AI inference works than traditional interframe codecs designed for continuous playback. The second involves integrating visual AI into live production and archive workflows using GPU acceleration, moving inference from offline batch processing into real-time pipelines. The third covers the swXtch AI Router, a multicast network solution designed to handle the bandwidth demands of distributing AI inference at live broadcast scale.

The argument has practical consequences for equipment buyers: if AI tools increasingly consume media as data rather than as video, the encoding and transport formats that optimize for perceptual quality may not be the right substrate. V-Nova has a commercial interest in VC-6 adoption, but the structural claim — that human-vision codecs are not machine-vision codecs — is a distinct technical position that is gaining traction in the industry.

The post arrives as multiple vendors are positioning AI as an operational layer, not just a feature, making codec architecture an increasingly upstream procurement question.