The AI workload will shift the limits of traditional data center infrastructure, require a new level of performance, scale, security and operational strictness. To satisfy these needs, Cisco and Nvidia provide verified, AI corporate infrastructure, which occurs together with CISCO switching, calculations, security and silicone One innovation with NVIDIA technology, NVIDIA AI Enterprise Software, Nvidia Spectrum-X and NVidia Blueld Technologies.
In March, NVIDIA GTCs in March set out the results of this cooperation and how they shape the future of AI infrastructure across accelerated processing, network, security and operations.
Spectrum-X + Cisco Silicon One: Production for AI in scale
The core of the initiative is the integration of the NVIDIA Spectrum-X capabilities into the Cisco Silicon One network infrastructure. In addition to selling switches based on the NVIDIA spectrum, this provides customers with flexibility in the unified operating model.
Key with you:
- Cisco has developed new reference architecture (based on NVIDIA Enterprise Reference Architectures), included Cisco silicon switches with NVIDIA supernas and Cisco Spectrum switches with NVIDIA Supernics.
- Integration with NVIDIA Ethernet Supernics (NVIDIA ConnectX and Bluefield) raises awareness of fables, optimizing traffic flows, overload control and improving the use of buffers.
- Future updates will allow integration at the level of telemetry, allowing deep knowledge of nic-to-SWitch to optimize and insulation of failures.
As announced at the Cisco Live 2025 San Diego, key software integration between Cisco NX-OS and Bluefield-3 switches has been shown, and Spectrum-X licenses will soon be available to add NX-DOS switches. If you want to learn more about this exciting announcement, watch this video.
AI Hyperfabric AI Cisco Nexus: Ai infrastructure solution with complete stak
Nexus Hyperfabric AI is a CISCO platform to design clusters, deployment and life cycle. It exceeds the switch at the switch level to supply the model of managed deployment and OPS, which includes the network of Cisco and Nvidia, accelerated computer technology and NVIDIA AI Enterprise Software software.
The main options include:
- Blueprinting with a full stack: Before you buy (including processing, network and storage), suggest your topology without hardware. Hyperfabric.cisco.com early modeling with reference architecture for Cisco customers.
- Verifying Cable Plan + Optics: Automatic generation of precise physical topologies, including optics, cables, range and bombs at SKO level to prevent expensive errors.
- Function of help deploying auxiliary hands: Verify the physical installation step by step at the stand level; The flag is not getting started in real time using digital validation of twins.
- Monitoring based on the claim: Identify the root cause of any detected problem using the production and monitoring of resources that constantly verify the availability and tired.
- Fabric Intelligence based on agents: Nothing agents create synthetic operations, operate health checks and surface anomalies through the integrated telemetry magazine.
Nexus Hyperfabric AI ships in the summer of 2025 with built-in support for Bluefield-3 SuperSnics, Ai-specific textile telemetry and automated validation.
Cisco Secure AI Factory also offers a modular approach to NVIDIA reference architecture built on the Cisco Nexus and Cisco Nexus 9000 series.
Security in scale: Cisco Secure AI Factory with NVIDIA
Since the business load AI scale, security becomes primary in infrastructure. NVIDIA consults AI factories as an integrated, special-purpose computing system, networking and software for industrialization of AI-from data cycle industrialization, created by inference-A reference architecture for Partys for implementation. The Cisco Secure AI factory builds on the NVIDIA AI factory vision, adds native visibility and Suncross across calculations, networks, software and data planes.
Robust Infrastructure security AI is delivered through two basic Cisco solutions:
- Defense of Cisco AI:: It provides a large language model (LLM) observability and threat detection across application programming interfaces (APIs), end points based on web inference and hybrid deployment. Currently in public beta.
- Cisco Hypershield: It offers a data processing unit (DPU) based on the recovery of the Bluefield, based on the creation of a distributed promotion of a policy without adding hardware; Integrates with the extended Berkeley Packet Filter (EBPF) filter and supports the policies of asymmetric circuits.
The Cisco Secure AI Factory Architecture architecture avoids costly hairstyle or overwork hardware and uses existing Bluefield supernic investments for inline enforcement and introspe.
The actual depth of engineering
The cooperation will enforce the pillar of the AI security factory and expand them to them. This included:
- Dual silicon strategy: Cisco can deploy switches based on the Eithon Silicon One or NVIDIA Spectrum spectrum in the back-end production. This gives customers option with a unified operating model.
- Fail + Synchronization Nothing: The joint efforts to engineering will make Silicon One switches fully supported by NVIDIA supernias, optimizing the behavior of end Tonds.
- Extensive Bluefield integration: Cisco will strengthen its integration with NVIDIA Bluefield to support the accelerated cyber security Casses, including zero level security and real -time threat detection.
Infrastructure AI is not long about how to combine isolated components. It is about building integrated systems that extend linearly, work intelligently and dynamically. With Cisco and Nvidia, customers will receive enterprise architecture, which demands all three – approaches accelerated calculation, software, network fabric and security.
Whether customers are just beginning to create AI factor with dozens of accelerated computational clusters or planning on a large screen, this initiative sets an internship for scalable, safe and simple AI clusters.
Share:
(Tagstotranslate) AI