Cisco Brings Agentic AI to Sensitive Data On Premises

Cisco is bringing Splunk’s AI capabilities to customers that cannot or will not move sensitive operational data into public clouds, a move that could widen the market for enterprise agents in regulated industries. The company announced Cisco AI POD for Splunk at Splunk’s .conf26 conference in Denver on September 15, 2026, extending its partnership with NVIDIA to deliver accelerated AI infrastructure for on-premises, private-cloud and air-gapped deployments.

What changed

The new system combines Cisco infrastructure, NVIDIA accelerated computing, Kubernetes and Splunk’s enterprise data platform in a prevalidated deployment. Cisco says customers can run Splunk AI Assistant and forthcoming agent-building tools against local data, while choosing among supported open and proprietary models rather than sending logs and telemetry to an external provider.

The package is aimed at security operations, where machine data may include identity, network, application and infrastructure records. Keeping that information inside an organization’s own environment can help address data-sovereignty, privacy and compliance requirements, but it also transfers more responsibility for hardware, model operations and security controls to the customer.

Cisco is pairing the infrastructure announcement with expanded Splunk Agent Observability. Its new Tokenomics capability tracks model and coding-agent usage, costs, adoption and productivity in real time. The company says it can attribute spending across teams and users, forecast consumption before a billing period ends, and connect AI costs with operational outcomes.

Why it matters

The announcement reflects a shift in enterprise AI from experimentation toward operational control. Many organizations are willing to use agents for investigations, monitoring and response, but hesitate when they cannot explain an agent’s actions, contain sensitive data or predict the bill. By putting inference closer to the data and exposing performance, safety and spending metrics, Cisco is targeting those practical barriers rather than competing only on model capability.

It also strengthens Cisco’s broader strategy of treating AI as an infrastructure and observability problem. The company is positioning Splunk as the control layer that can watch agents across security, networks and applications, while NVIDIA supplies the accelerated computing. That combination could appeal to hospitals, governments, manufacturers and financial firms whose data cannot easily be centralized in a hyperscale cloud.

What remains uncertain

Cisco’s announcement describes product availability and supported components, but it does not establish how much the system will cost, how much specialized staff customers will need, or whether local deployments will match cloud systems on model quality and ease of maintenance. “Tokenomics” may improve visibility, but measurement alone does not prove that an agent is delivering net savings or reliable decisions.

The larger test will be adoption. Enterprises may prefer the control of private infrastructure, yet on-premises AI can introduce slower upgrade cycles, capacity constraints and additional security duties. Cisco’s bet is that customers will accept those trade-offs when regulatory exposure, data sensitivity and unpredictable agent costs make public-cloud deployment difficult.