NVIDIA and EBU expand a sovereign AI plan for European broadcasters

At a show in Berlin, the European Broadcasting Union (EBU), an alliance of public service media, tells that it expands its sovereign standards with NVIDIA. Sovereign is the control of data, systems and models by the broadcaster. Cloud providers, open model developers and integrators connect to the plan. Adobe tells that NVIDIA chips give its video search a maximum of 5x the speed.

How it works

The plan will set requirements and reference architectures, which are designs to copy. They include where work operates, how data is controlled and which models are in use. Cloud partners align the standards to local alternatives:

• Microsoft gives Azure Local and Foundry Local to use sensitive material in a local system.
• Nebius gives a platform to train and operate their models in Europe.
• NVIDIA Nemotron open models are a start point to change.

The numbers

Adobe tells that embedding has about 5x the speed. It also gives a maximum of 4x the throughput and 5x less latency. A new proxy video has 80% less storage than the proxy that it replaces. These are numbers from the company, with no external test.

What is not known

The standards are not set at this time. The source gives no dates, no prices and no names of broadcasters that will use them.

Insight: NVIDIA gives away software to keep broadcasters on its chips

On 13 October, NVIDIA made its Run:ai platform open source. Here it gives Nemotron open models, and the Adobe speed increases use NVIDIA tools such as CUDA and TensorRT. This reads as a bet that open software keeps customers on NVIDIA chips. The standards can set which chips public media get. Monitor which clouds use the plan.