NVIDIA released Nemotron 3 Ultra globally on June 7, cutting inference costs 30% and delivering 5x faster performance for enterprise AI agents compared to prior models.1
US cybersecurity firms CrowdStrike and Palantir are already running long-horizon autonomous agents on Nemotron for threat response and operational decisions.1 The deployments are production, not pilots.
The NemoClaw framework targets electronic design automation — chip design. Adopters span three continents: France's Dassault Systèmes, Germany's Siemens, and US firms Cadence and Synopsys are building autonomous AI engineers for EDA workflows.1 EDA is bottlenecked by engineer availability worldwide; autonomous agents could compress design cycles and reduce verification load.
Jensen Huang positioned the release as infrastructure, not research: AI agents will become the largest users of computing globally.1
The economic logic scales across markets. Knowledge-intensive industries — chip design, cybersecurity, industrial simulation — face the same constraint everywhere: specialist labor limits throughput. Nemotron's cost reduction removes the primary barrier to deploying agents at scale without proportional headcount growth.
For European industrial firms, the relevance is direct. Siemens and Dassault Systèmes operate complex engineering pipelines across global supply chains. Autonomous AI engineers that compress EDA cycles could improve margin structures without expanding headcount in high-cost labor markets.
NVIDIA's Agent Toolkit gives developers a structured framework for building and deploying agents across enterprise environments — reducing the integration burden that has slowed adoption outside the US.
Hard evidence will come from earnings calls, not product launches. Cost-savings disclosures from CrowdStrike, Palantir, Cadence, Synopsys, Siemens, and Dassault Systèmes over the next two to three quarters will be the first indicator of whether the 30% cost thesis holds in production at global scale.1
