January 6, 2026
The NVIDIA keynote at CES today was great. Lots of info about really nifty upcoming data center hardware that will power training and inference at scale, of course. But also a deep dive into the multi-model, multi-modal, hybrid cloud/local world that we've been trying to help bring into being with our work on the open source Pipecat realtime, multi-modal framework.
NVIDIA is all in on open source. I lost track of the number of times Jensen said some version of "the entire thing is open" in the keynote. NVIDIA expects open models to drive a lot of the growth in use cases that wouldn't be practical without open models and tools.
Open models matter because they are a foundation for flexibility, customization, optimization, and research.
We've been building with the NVIDIA open models that were announced in the keynote. Stay tuned for lots of fun stuff all week. To get started, check out the new Nemotron Speech ASR open speech-to-text model. Super-fast, super-flexible, high performance transcription.
The big roundup announcement post[1]
The new Nemotron Speech ASR model[2]