Exploring how compute aware formats are enabling AI native workflows across live production, archives, and video analytics.
By Guendalina Cobianchi
AI in media and entertainment is no longer an experiment. Across the operators we speak with every week, Visual AI has moved out of the innovation lab and into the core priority list. DAZN’s ‘Delta Protocol’ ambitious vision is a vivid example of that shift, and most major sport, entertainment and news organisations are moving in the same direction. Adding AI to live, archival and analytics workflows is now a question of how, not if.
The quiet problem is that today’s video formats were designed for human eyes, not for machines. Feeding them to AI means decoding everything, moving full-resolution data and maintaining parallel proxies. This creates a trillion-dollar blind spot. SMPTE VC-6, our compute-aware data format addresses this challenge by providing a hierarchical representation that lets any AI model pull only the resolution or region it needs. Decode once, reuse many times.
In live, we are launching with swXtch.io the first multicast network solution powered by VC-6, swXtch AI Router. One feed, many AI models, no custom engineering. This approach is perfect for sport, news or any operator under pressure to put AI into production fast. We are recruiting pilot companies to try it out, get in touch if you want to know more.
In video analytics, archive and data curation, our ongoing collaboration with NVIDIA continues. The new CUDA-optimised VC-6 libraries with batch mode, featured this past week on the NVIDIA Developer Blog, deliver several-fold GPU speed-ups over legacy formats. Early pilots show >6x faster data curation for model training in a conservative scenario, with similar benefits in video search and summarisation. This effectively turns dormant archives into active ones for logging, highlights, clipping and more.
That brings me to the three conversations I am most hoping to have at NAB this year:
- With operators already launching Visual AI in live production, archival intelligence, video analytics or data curation for model fine tuning. I want to understand what is working, what is blocking you, and where a compute-aware format would move the needle.
- With prospective partners such as live video contribution and transport platforms, MAM vendors, AI application builders and system integrators who want to turbo-charge their stack with a complementary compute-aware format.
- With like-minded people who simply want to compare notes on where AI-native media is heading. No agenda, just a good conversation.
AI-native media is being built in the open, by a lot of hands. If you are exploring how compute-aware formats fit into your live, archival or analytics workflows, or simply want to think out loud together, I would be delighted to find some time together at the show.
Find us at West Hall, booth W1646/W1647. Book a slot here.
