Xage and NVIDIA Push Zero-Trust into the Heart of the AI Factory

In what may prove to be a pivotal moment for enterprise AI security, Xage Security has partnered with NVIDIA to embed hardware-accelerated zero-trust protections deep into the network fabric of the modern data centre. The integration of Xage’s Fabric Platform with NVIDIA’s BlueField Data Processing Unit (DPU) signals a major move to secure “AI factories” and the infrastructure supporting agentic AI workflows.

At its core, the alliance aims to address one of the fastest-growing cyber risk surfaces: autonomous AI agents interacting with models, data pipelines and corporate systems at machine or sub-machine speeds. With the BlueField-3 DPU leveraging dedicated hardware resources to offload and accelerate networking, storage and security functions, Xage’s identity-based access controls now operate not just at the software layer, but at the silicon level.

As Roman Arutyunov, Co-Founder and SVP of Product at Xage Security, puts it:

“Given the massive innovation opportunities of AI, we need to make sure that organizations’ deployments are not held back by cybersecurity gaps—and we need to have the necessary protections run at extraordinary speed and scale. That’s why we’re collaborating with NVIDIA to help deliver lightning-fast and jailbreak-proof security at the heart of the modern AI factory and AI-enabled critical infrastructure.”

That statement underscores the dual challenge facing enterprises today: enable rapid, scalable AI innovation while simultaneously preventing adversaries from exploiting emergent vulnerabilities in agentic systems, lateral data flows, privilege escalations and model-based access.

NVIDIA’s Sr. Distinguished Architect for Cybersecurity, Ofir Arkin, emphasized the infrastructure shift driving the deal:

“As AI factories emerge as the foundational infrastructure accelerating AI innovation, safeguarding them has become a critical priority. Together, NVIDIA BlueField and Xage’s zero-trust security enable organizations to modernize their protection strategies across AI factories and infrastructure — driving secure, scalable innovation forward.”

Why this matters now

The term “AI factory” has been floated by industry analysts to describe the mega-data-centre scale facilities where tens of thousands of GPUs, DPUs, agent orchestration engines and data pipelines converge to train, fine-tune and deploy large language models and other generative AI workloads. In these environments, traditional security models—perimeter-based firewalls, VLANs or simple identity checks—can no longer keep pace.

Xage’s integration with BlueField seeks to flip the model: instead of retrofitting zero-trust controls onto a legacy stack, it embeds segmentation, least-privilege enforcement, policy-based de-escalation and workload isolation into the network dataplane itself. According to Xage, this means that autonomous agents—be they code, models or bots—are given only the specific data, pipelines or models they are authorized for, for exactly the time permitted, and any further attempt at privilege escalation or lateral movement can be blocked at line speed.

In practice, that means enterprise AI deployments in sectors such as energy, utilities, manufacturing and transportation—where “millions of assets and billions of data flows must be secured in real time” — now have a route to combine performance and security in one architecture.

What this means for critical infrastructure

The upgrade from software-only enforcement to hardware-accelerated segmentation via BlueField is more than incremental. By offloading enforcement to the DPU, Xage can apply fine-grained identity rules at the edge of the data path with minimal latency and maximal scalability. That’s crucial when autonomous systems execute millions of micro-transactions per second or when AI agents self-modify and spawn new behaviors in real time.

For infrastructure operators—utilities, manufacturing lines, grid control systems—this deal signals that Zero Trust is no longer optional or niche: it must be baked into the architecture of next-generation AI deployments. Xage makes a point of highlighting compliance as well: its system supports regulatory regimes such as NIST, NERC CIP, EU NIS2 and U.S. Zero Trust mandates, giving organizations the audit trails and identity visibility they’ll need.

What to watch next

1. Adoption momentum. Will large-scale infrastructure operators invest in DPU-based segmentation architectures? The cost and operational impact of upgrading to BlueField-3 and integrating Xage will be a key factor.

   
   

2. Performance vs. policy trade-offs. Embedding policy enforcement at the dataplane level is elegant, but if it introduces latency or complexity it could hinder adoption in high-throughput AI environments.

   
   

3. Agentic AI risk surfaces. As autonomous agents proliferate, new threat vectors will emerge—model poisoning, rogue agent orchestration, privilege escalation via dynamic model-agent workflows—and the effectiveness of hardware-based enforcement will be tested.

   
   

4. Ecosystem integration. Will this become a platform play? With NVIDIA’s dominant position in the data-centre AI stack and Xage aiming for heterogeneous enforcement across IT and OT (operational technology) systems, the partnership could attract other infrastructure and ecosystem partners.

   
   

In short, Monday’s announcement marks a significant move in the evolution of enterprise AI security. By combining high-performance hardware (BlueField-3) with fine-grained identity and policy-based controls (Xage Fabric Platform), the partnership is staking a claim that the future of “AI factories” must be built on a foundation of Zero Trust. And in doing so, they may well be redefining how enterprises secure agentic AI at scale.