October 22, 2026
Selling Palo Alto Networks into AI Data Centers
Running seller blog · Entry 1 · October 22, 2026
1. Positioning
- AI factory and neocloud/GPU cloud operators: Security that runs on the BlueField DPU leaves host compute for AI work. Palo Alto says offloading security "delivers two critical benefits: increased token throughput and stronger runtime protection." [1]
- Enterprise private AI: Prisma AIRS on BlueField is part of the NVIDIA Enterprise AI Factory validated design (Jan 5, 2026). That makes it a pre-validated design, not a custom integration project. [2][3]
- Colos and facilities: Lead with standard NGFW segmentation, Xpanse, and SOC services. The docs also list OT traffic as a supported profile for the DPU firewall. [4]
- Tenants running workloads in these data centers: One platform covers models, agents, identities, and the SOC, whether the AI runs on-prem, at a GPU provider, or in the public cloud.
2. Use cases that fit
- Inspecting host-boundary traffic in hardware. Prisma AIRS runs as a containerized firewall on BlueField-3. It handles north-south traffic in v-wire mode at about 1–5 Gbps per host, with up to 1,000 steering rules, managed from Strata Cloud Manager or Panorama. [4]
- East-west and high-throughput segmentation. The DPU firewall doesn't handle pod-to-pod traffic within a host or traffic above its throughput range. Palo Alto routes those to a centralized Prisma AIRS or a hardware NGFW, using L3 routing or PAN-CNI chaining. [4]
- Agentless runtime visibility for the SOC. DOCA Argus (and NVIDIA Sentry) inspects memory and network activity out-of-band and streams alerts to Cortex XSIAM, which correlates them with network, cloud, and identity data. [1][5] Palo Alto says XSIAM can detect kernel rootkits and living-off-the-land attacks without an agent on the host. [5]
- Protecting models and agents. Prisma AIRS covers AI Runtime Security, an AI Gateway, AI Red Teaming, AI Supply Chain Security, and AI Agent Identity. [4][1]
- Agent identity. Idira gives each agent a verifiable identity, task-scoped access, and attributable actions. The integration with NVIDIA Sentry on BlueField is not yet available. [1]
- Cloud and exposure. Cortex Cloud covers AI workloads in the public cloud, and Xpanse covers the internet-facing attack surface. As one example, Xpanse counted 22.6M+ instances potentially vulnerable to PHP2Shell. [6]
- User access to AI. Prisma Access/SASE governs how users and apps reach AI services.
3. Product and service map
| Use case | Palo Alto product | Services |
|---|---|---|
| Host-boundary protection on DPU | Prisma AIRS AI Runtime Security on BlueField-3 (GA) | Spectro Cloud for deployment and operations; WWT AI Proving Ground |
| Fabric segmentation and east-west | Hardware NGFW, VM-Series, CN-Series, centralized Prisma AIRS, Panorama or SCM | Professional services, ASC partners for NGFW support |
| SOC for AI infrastructure | Cortex XSIAM + DOCA Argus/Sentry | Unit 42 (MDR, IR), MSSPs with XSIAM specializations |
| Model and agent security | Prisma AIRS (Runtime, AI Gateway, Red Teaming, Supply Chain) | Unit 42 AI security assessments* |
| Agent identity | Idira (Sentry integration in progress) | Deloitte Secure AI Foundations |
| Cloud AI workloads | Cortex Cloud | Partner professional services |
| Exposure management | Cortex Xpanse | Unit 42 |
| User access to AI | Prisma Access / Prisma SASE | ASC and MSSP partners |
| End-to-end AI factory build | The full platform | Deloitte Silicon-to-Service (S2S), WWT [1] |
*Confirm the current name and scope of the Unit 42 offering before quoting it.
4. Stay ready
Qualifying questions
- Are you on BlueField-3, and which model? The docs list the B3220 and DOCA 3.x. [4]
- Do you run Kubernetes with NVIDIA DPF?
- Are you building to the Enterprise AI Factory or OpenShell design?
- Who will operate the SOC for this infrastructure?
- Will agents act on your users' behalf? If so, how is their access governed today?
- Where will inference run: on-prem, at a GPU cloud, or in the public cloud?
Target accounts: NVIDIA-aligned neoclouds and GPU clouds, enterprises building private AI factories (finance, healthcare, public sector, manufacturing), sovereign AI programs, colos adding GPU halls, and industrial edge sites. [1]
Partner alignment: Co-sell with NVIDIA. Bring in Deloitte for builds and WWT for proof-of-concept testing. Engage Spectro Cloud for operating the DPU footprint. [1] Use MSSPs with XSIAM specializations for run services.
Get ready now
- Learn the BlueField-3 deployment doc. [4]
- Line up a WWT AI Proving Ground session. [1]
- Keep XSIAM, Prisma AIRS, and PCNSE certifications current.
- Prepare a demo that follows an Argus alert into an XSIAM incident.
Caveats
- GA: AI Runtime Security on BlueField. [1]
- In progress: the Idira–Sentry integration, and broader XSIAM fleet visibility and response workflows. [1]
- BlueField is required for all DPU features. On hosted AI APIs and standard cloud VMs, use Cortex Cloud, Prisma AIRS, and the SaaS controls instead.
- Not supported on the DPU firewall: east-west pod inspection, HA, FIPS, and IPv6 steering. [4]
- PQC: I haven't found a sourced AI data center claim yet, so it's held for a future entry.
Sources
- https://www.paloaltonetworks.com/blog/2026/10/advancing-agentic-ai-security-with-palo-alto-networks-and-nvidia/ (Oct 21, 2026)
- https://www.paloaltonetworks.com/blog/2026/01/support-nvidia-enterprise-ai-factory/ (Jan 5, 2026)
- https://blogs.nvidia.com/blog/bluefield-cybersecurity-acceleration-enterprise-ai-factory-validated-design/
- https://docs.paloaltonetworks.com/prisma-airs/ai-runtime-security/airs-vm-firewall/nvidia-bluefield3-to-secure-ai-factories
- https://www.paloaltonetworks.com/blog/2026/06/reinventing-security-for-the-agentic-nvidia-ai-factory/ (Jun 1, 2026)
- https://unit42.paloaltonetworks.com/cve-2026-109618-cve-2026-109619/
- https://www.paloaltonetworks.com/blog/2026/09/securing-ai-agents-at-scale-with-nvidia/