Arm and NVIDIA executives point to a new wave of “agentic” AI computing at the edge

Arm Holdings said it is leaning into “agentic AI” computing, the idea that software agents can take actions toward goals rather than only respond to prompts. In remarks highlighted by Yahoo Finance, Chris Bergey, Arm’s Executive Vice President of the Edge AI Business Unit, tied the shift to new device-side AI capabilities and the growing importance of running more intelligence outside of centralized data centers.

Bergey’s comments focused on the recently unveiled NVIDIA RTX Spark, which NVIDIA describes as a new platform intended to support agentic AI experiences. While agentic AI is often discussed in terms of cloud services and large models, the Arm executive framed the opportunity as extending that functionality closer to where data is generated, including consumer and industrial edge devices.

RTX Spark’s role in the conversation centers on “agentic” workloads that can interpret context, decide on next steps, and carry out actions within defined boundaries. That is a materially different engineering target than traditional AI chat responses, because the system must manage tools, workflows, and intermediate states. Arm’s emphasis on edge AI suggests it expects more of those capabilities to depend on power-efficient compute and on-device integration rather than remote inference alone.

The timing is also significant for Arm and NVIDIA because both companies are trying to convert developer interest in AI into hardware and software platforms that can scale. Arm has positioned its instruction-set and systems expertise around efficiency, while NVIDIA has built its recent momentum around accelerated compute, developer tooling, and full-stack environments for training and inference.

Arm’s Edge AI Business Unit is responsible for partnerships and product enablement that help developers deploy AI on Arm-based systems. In this context, Bergey’s remarks imply that Arm sees agentic computing as a workload category that will increasingly require both stronger on-device performance and tighter software-hardware alignment, even if not all components are processed locally.

For NVIDIA, agentic AI is increasingly tied to how applications orchestrate models, tools, and agent behavior in real time. RTX Spark, as referenced in the Yahoo Finance report, is presented as a vehicle for bringing those experiences to the RTX ecosystem. However, the public details highlighted in the report were not sufficient to confirm specific technical specifications, supported model types, or performance benchmarks.

NVIDIA and Arm both have incentives to accelerate adoption, but the market reality is that agentic AI still faces constraints around reliability, safety, and compute overhead. In practical deployments, companies often limit agents to narrow toolsets and environments, and they rely on developer controls to reduce errors and unintended actions.

What remains unclear is how much of agentic functionality RTX Spark will execute on-device versus in hybrid configurations, and whether the initial implementations will target specific languages, agent frameworks, or application domains. The referenced report did not provide those technical breakdowns, and NVIDIA’s general newsroom materials are broad rather than a detailed product brief.