News

Public · Published

Nvidia's AI Safety Push Could Slow Things Down by 25%, Analyst Warns

Nvidia announced its OpenShell reference architecture at GTC Berlin to control what AI agents can access and do. BNP Paribas analyst Karl Ackerman warned that runtime could slow by as much as 25% as a potential trade-off for added safety.

Published:

Updated:

What happened

Nvidia announced its OpenShell reference architecture at GTC Berlin to control what AI agents can access and do. BNP Paribas analyst Karl Ackerman warned that runtime could slow by as much as 25% as a potential trade-off for added safety.

Confirmed

Global impact / market context

Slower runtime could raise computing costs per task and reduce how many AI agent jobs a company can run, which affects capital spending plans and the revenue Nvidia and partners expect from agent software.

Analyst inference

AI agents can reach files, tools and online services, so security limits matter to buyers. The trade-off between safety and speed shapes demand for Nvidia chips, Cisco, CrowdStrike and Microsoft offerings.

Analyst inference

What to watch

  1. Nvidia built OpenShell on open-source software, combining AI computing systems, BlueField-4 data processing units and the Sentry security framework to monitor activity and block rule violations. Confirmed
  2. Watch whether Nvidia or partners disclose runtime figures for OpenShell, since the 25% slowdown is an analyst estimate and not a confirmed product measurement. Proposed
  3. Track adoption among Cisco, CrowdStrike and other named companies, because customer uptake would signal whether security controls outweigh slower agent performance. Analyst inference

Evidence

  • Original source link unavailable