NVIDIA DOCA Software Platform

DOCA enables the creation of AI-native, key-value (KV) cache and storage infrastructure for modern AI factories. With NVIDIA DOCA Memos™, developers can build an agentic KV cache sharing framework across the AI pod, extending effective GPU memory capacity, reducing recomputation, and boosting tokens per second for long-context, multi-turn agentic inference.

DOCA also accelerates next-generation storage services including Non-Volatile Memory Express over Fabrics (NVMe-oF), storage virtualization, and data services on BlueField. With DOCA SNAP, developers can build software-defined storage appliances and virtualized NVMe solutions that offload storage processing from host CPUs while delivering high-performance data access as agentic workloads scale.

DOCA gives developers the software building blocks to create intelligent, hardware-accelerated cybersecurity services for modern AI factories on BlueField. With DOCA Argus and DOCA Vault, developers can detect anomalous AI agent behavior, monitor system integrity, and secure credentials, keys, and sensitive assets with isolated hardware-backed protection.

DOCA also enables high-performance network security with DOCA Flow for traffic inspection, segmentation, policy enforcement, threat mitigation, and accelerated encryption directly in the data path. By isolating security functions from host CPUs, DOCA helps improve protection while reducing overhead at cloud scale.

DOCA Platform Framework (DPF) simplifies the provisioning of BlueField at scale and manages the full lifecycle of infrastructure services through a Kubernetes-native operating model. Developers and operators can deploy networking, storage, and security services as cloud-native applications, using standard Kubernetes APIs to automate upgrades, policy control, scaling, and day-two operations.

DPF also supports trusted host and zero-trust operating models to strengthen isolation between tenants, hosts, and infrastructure services. Integrated with NVIDIA DSX OS and leading enterprise Kubernetes distributions, DPF enables consistent operations across private clouds, service providers, and large-scale AI factory environments.

DOCA provides real-time observability across BlueField and ConnectX networking, giving developers and operations teams deep visibility into the health, performance, and utilization of AI infrastructure. Streaming metrics, events, counters, and diagnostics help monitor networking throughput, latency, storage activity, security services, and system resource consumption across distributed AI factory environments.

With granular telemetry and fleet-wide insights, teams can detect anomalies early, troubleshoot bottlenecks faster, and respond to incidents before they impact workloads or GPU utilization. By integrating with existing monitoring and analytics platforms, DOCA Telemetry helps optimize infrastructure efficiency, improve uptime, and maintain predictable performance at scale.