NVIDIA launches agent platform
NVIDIA announced a new enterprise AI agent + inference platform at GTC 2026 and listed Baseten among 17 adopters — a direct signal to inference providers and customers choosing hosted stacks. The launch bundles agent tooling with inference partners and gives named providers an immediate credibility boost for low‑latency agent deployments.
The Agent Toolkit bundles the NemoClaw secure runtime, Nemotron open models, the AI‑Q agent blueprint, cuOpt skills and the OpenShell runtime announced)). NVIDIA released Dynamo 1.0 as an open‑source inference framework aimed at agentic and generative workloads announced)), and positioned Dynamo to run natively on its Blackwell and Vera Rubin inference stacks for lower latency at scale noted)). NVIDIA’s rollout named a group of 17 enterprise software partners including Adobe, Salesforce, SAP, ServiceNow, Siemens, CrowdStrike, Atlassian, Cadence, Synopsys, IQVIA, Palantir, Box, Cohesity, Dassault Systèmes, Red Hat, Cisco and Amdocs listed)). The company also published a partner roster of inference and cloud providers—Baseten, CoreWeave, DeepInfra, Fireworks, Together AI, DigitalOcean, Bitdeer AI, Lightning, Vultr and others—integrated into the Agent Toolkit reference deployments announced)). Baseten’s inclusion follows a $300M Series E that valued the startup at $5B with NVIDIA named as an anchor investor and reporting a roughly $150M strategic stake from NVIDIA in the round announced)). Baseten published results showing about 2x faster inference when running on NVIDIA’s Dynamo runtime and outlined optimizations—KV cache‑aware routing, KV cache offload and an SLA‑based autoscaler—that delivered those gains reported)). The package and partner list debuted during NVIDIA’s March 16 GTC keynote and related press materials, with NVIDIA promoting the announcements across its GTC press kit and event coverage that drew more than 30,000 attendees from over 190 countries reported)).
