NVIDIA and SAP are embedding NVIDIA OpenShell, an open-source runtime for autonomous AI agents, into SAP Business AI Platform as the default security layer for every SAP AI agent. The expanded partnership was announced May 12, 2026 at SAP Sapphire, with NVIDIA founder and CEO Jensen Huang joining SAP CEO Christian Klein's keynote by video. OpenShell will sit beneath custom agents built in Joule Studio, SAP's environment for building and managing enterprise agents end to end.
The pitch is that agents touching systems of record need more than a model. OpenShell provides isolated execution environments, policy enforcement at the filesystem and network layers, and infrastructure-level containment meant to limit damage when agent logic fails. SAP will run this as the runtime security layer across finance, procurement, supply chain and manufacturing workflows that already live inside its platform.
NVIDIA is also putting its agent reference blueprint, NemoClaw, directly inside Joule Studio. SAP customers building custom agents get a structured path from prototype to production deployment without writing their own sandboxing, policy enforcement or audit scaffolding. For development teams, that compresses the part of the agent build that has nothing to do with the business logic and everything to do with not getting fired.
Key facts
- 01SAP is embedding NVIDIA OpenShell, an open-source agent runtime, into SAP Business AI Platform as the security layer for all SAP AI agents.
- 02The deal was announced at SAP Sapphire on May 12, 2026, with Jensen Huang joining Christian Klein's keynote by video.
- 03NVIDIA NemoClaw, a reference blueprint for autonomous agents, will be available directly inside SAP's Joule Studio.
- 04SAP engineers are codeveloping OpenShell with NVIDIA, contributing to the open-source project on runtime hardening, policy modeling and identity integration.
- 05NVIDIA is itself an SAP customer, running its own finance, supply chain and logistics workloads on SAP systems.
SAP engineers are codeveloping OpenShell alongside NVIDIA's team and contributing back to the open-source project. The stated focus areas are runtime hardening, policy modeling, enterprise identity integration, and auditing and governance hooks — the unglamorous plumbing enterprises ask about before any agent gets near a general ledger. SAP is, in effect, helping shape the runtime its own customers will depend on.
“NVIDIA OpenShell asks whether an agent action can safely execute; Joule Studio asks whether it should happen at all. Together they form the runtime security layer for every SAP AI agent.”— Jaeden Schafer
The two systems are designed to ask different questions. NVIDIA OpenShell asks whether a given agent action can safely execute at the infrastructure level. Joule Studio's enterprise control layer asks whether that action should happen at all given the user's role, the relevant process and the data boundary. SAP and NVIDIA argue that pairing the two closes a gap that application-layer security on its own does not address.
Huang has described AI as a five-layer cake: energy, chips, infrastructure, models and applications. SAP sits squarely at the top layer, running the financial, procurement and supply chain backbones of much of the Global 2000. That placement is what makes the partnership commercially significant: agentic AI's revenue path runs through the systems where money actually moves, not through chat windows.
NVIDIA brings shared context as an SAP customer itself. The chipmaker runs its own finance, supply chain and logistics on SAP, which both sides are using as a credibility argument when telling other enterprises what production-grade agent governance needs to look like. It is a useful sales talking point and a real source of design feedback.
For SAP, the partnership is a hedge against a market in which agent platforms from hyperscalers and pure-play startups are all racing to wrap themselves around enterprise data. By making OpenShell the default runtime and NemoClaw the default blueprint, SAP keeps custom agent development inside Joule Studio rather than ceding that workflow to outside tools. That matters for retention as much as for security.
The friction points are familiar. Open-source runtimes only carry trust if the governance, identity and audit integrations land in time for customer security reviews. Enterprises that already run SAP also tend to run Microsoft, Oracle and Salesforce, and an OpenShell-based agent will eventually need to act safely across those boundaries too. Neither company committed to a public timeline for how quickly NemoClaw blueprints will cover SAP's full module footprint.
For NVIDIA, the play is consistent with its broader push beyond chips into the software and runtime layers that determine how agents actually run in production. For SAP, it is a signal that the company intends to be the control plane for enterprise agents rather than a passive data source for someone else's. Either way, the agentic AI conversation has moved from demos to the part of the stack where audit logs matter.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
