VAST Data has been fingered as a key partner in Nvidia’s efforts to spin up a “data flywheel” for continuous model improvement as the GPU giant spun up its Paris GTC conference this week.

Nvidia’s founder and CEO Jensen Huang kicked off the conference with some cheerleading for the continent’s sputtering AI industry, insisting Europe was not just adopting AI, but “building” it.

He touted collaborations with European governments, acknowledged demands around AI sovereignty, and said that the firm was working with European investors to hook European startups to its Nvidia DGX Cloud Lepton platform and market place.

He cited new technology hubs in Finland, Germany, Spain, Italy and the UK, and what Nvidia says is the world’s first industrial AI cloud in Germany, aimed at manufacturers. And he hyped up an “inflection point” around quantum computing.

But that will all demand torrents of data. And Huang cited a Nvidia NeMo Agent toolkit and Nvidia AI Blueprint for building “data flywheels” which will further accelerate the development of safe, high-performing AI agents.

The approach was fleshed out in a conference session by Nvidia VP of Gen AI, Kari Briski.

VAST, Nvidia and asset servicing firm CACEIS are “exploring a real-time AI platform concept designed to securely capture, transcribe, and analyze 100 percent of client meetings.”

Spinning energy

Briski described CACEIS as a pioneer, “with a vision of building an agentic AI data flywheel with NVIDIA and VAST to supercharge productivity for financial services in Europe.”

VAST said its AI OS, “combined with Nvidia AI Enterprise, which includes NeMo microservices that power a data flywheel for continuous model improvement, creates a unified environment where AI pipelines can constantly learn, adapt, and improve.”

VAST said “This reference workflow provides a self-optimizing foundation for scalable AI, laying the groundwork for billions of intelligent agents to fine-tune and evolve from their data and experiences.”

It will span data management, database services, and AI computer orchestration, and support “continuous, automated AI pipelines – from ingestion to inference to retraining.”

It’s a truism that AI relies on data. But speaking to Blocks & Files earlier this week, VAST’s CTO Andy Pernsteiner said feeding that data both quickly and reliably was critical. “What’s funny is people say, ‘OK, well, how do you make things faster?’ The way you make things faster is make sure they don’t go down.”

He said when it came to some of its biggest customers, “The main reason they stick with us is not because we made them so much faster. It’s because we were stable when everything else in the environment was falling over.”

VAST helps spin up Jensen Huang’s data flywheel in Paris AI throwdown