# How Cisco Built an Enterprise AI Factory: Scaling Economics, Performance, and Security > Source: > Published: 2026-11-06 19:35:41+00:00 *Every technology leader faces the enterprise AI paradox: as AI adoption accelerates, organizations run headfirst into external power, cost, and security constraints. Cisco IT proved that infrastructure is the ultimate control point.*   *Keep reading and watch the short video below, or tune into the [full podcast discussion](https://video.cisco.com/detail/video/6408028398112?autoStart=true&q=agentic%20shift), to learn how we built a full-stack, multi-tenant AI factory—tripling GPU utilization and delivering mission-critical reliability for 90,000 employees.*  ## The inflection point: From project-specific builds to mission-critical enterprise scale Like most enterprises, Cisco is on an ongoing AI journey. In the beginning, our mission was simple: get AI capabilities into the hands of teams so they could prove value quickly for specific use cases. We started with dedicated, project-specific AI infrastructure clusters for individual engineering and product teams. This approach was sufficient for experimental prototypes, but then we reached a crucial inflection point as AI evolved from isolated projects to mission-critical operations. The primary catalyst was the global deployment of MyAgent, Cisco’s enterprise super agent rolled out across our 90,000-person workforce. At the same time, we scaled dedicated capacity for Cisco IQ, enabling hosting live production workloads for thousands of enterprise accounts. Cisco IQ provides unified, AI-powered capabilities through Cisco Support and Professional Services. This significant surge of agentic workflows required us to modernize our infrastructure to deliver high performance, embedded security, and economic control at enterprise scale. ## Four barriers to scaling AI every CIO will face As we scaled our AI usage enterprise-wide, we knew we needed to address four critical barriers: 1. **Stranded capacity and low GPU utilization:** Static, project-based allocations created fragmented compute silos. Because GPUs were isolated by network and location, resources could not be pooled, leaving average utilization at 10%-15%. This forced teams to wait for entire GPU nodes to be vacated by other projects before they could begin their own workloads. 2. **Unpredictable token costs and workload placement:** Relying solely on close-sourced cloud models worked well for early pilots, but became less cost-efficient as AI scaled enterprise-wide. Compounding this challenge, a lack of visibility into token consumption across teams and tools made forecasting expenses and proving ROI nearly impossible.[Cisco’s research with Omdia](https://www.cisco.com/site/us/en/products/computing/offers/assets/omdia-ai-infrastructure-infographic.html) found that cloud spending jumped 30% on average after AI deployment. 3. **Expanding security attack surfaces:** Scaling autonomous agents and multi-tenant workloads introduced new threat surfaces, from prompt injection and proprietary data loss to network threats that could move laterally without detection. 4. **Thermal and power density ceilings:** Power availability is the true binding constraint. Even if an enterprise received 2,000 brand-new GPUs tomorrow, traditional air-cooled enterprise data centers hit thermal limits long before they can energize that compute density. While power limits, chip and supply chain shortages, and cloud pricing volatility are external realities, Cisco IT realized we have a critical internal control point: **our infrastructure.** Pooling compute can reduce stranded capacity. Real-time visibility into token spend — connected to business value — helps make costs predictable. Building security into every layer of the stack delivers stronger protection than bolting it on as an afterthought. And optimizing rack density addresses power constraints. Our infrastructure was the key to scaling efficiently, securely, and economically. ## A full-stack solution: How we built our AI Factory To overcome these barriers and scale AI economically, we deployed a unified, full-stack, multi-tenant AI factory across our global enterprise data centers. This architecture brings together compute, networking, security, and observability to give us the visibility and consistency needed to operate workloads across the stack from silicon to the applications: 1. **High-bandwidth pooled network fabric:** To reduce stranded capacity, we are consolidating previously fragmented backend networks into a single, high-bandwidth pooled fabric across our major data centers.[Cisco switches](https://www.cisco.com/site/us/en/solutions/artificial-intelligence/ai-networking-in-data-center/index.html) powered by[Cisco](https://www.cisco.com/site/us/en/products/networking/silicon-one/index.html)[Silicon One](https://www.cisco.com/site/us/en/products/networking/silicon-one/index.html) eliminate bottlenecks and ensure zero GPU packet-drop stalls. Because Silicon One is fully programmable, our network fabrics can quickly adapt to evolving AI demands, standards, and protocols without requiring costly hardware refreshes. Now, multi-tenant workloads dynamically and seamlessly distribute to available capacity in real time, preventing GPUs from sitting idle. 2. **Heterogeneous compute:** No single hardware configuration fits every enterprise workload. We standardized on Cisco Compute with[UCS-X Series servers](https://www.cisco.com/site/us/en/products/computing/servers-unified-computing-systems/ucs-x-series-modular-systems/index.html) supporting a strategic mix of NVIDIA and AMD GPUs. This heterogeneous architecture provides us the foundation to evaluate and align compute profiles to workload demands — evaluating cost-effective silicon for high-volume inference while sizing capacity for intensive training. This gives us the flexibility to optimize token economics and adapt to evolving hardware supply. 3. **High-density rack cooling:** To address power and thermal limitations within our existing footprint, Cisco IT engineered rear-door heat exchangers directly onto our racks. This absorbs thermal exhaust at the source, dramatically boosting cooling efficiency and doubling our compute density to two full AI nodes per cabinet without expanding physical real estate. 4. **Natively embedded AI security:** For security integrated at every layer, we deployed[Cisco AI Defens](https://www.cisco.com/site/us/en/products/security/ai-defense/index.html) e to apply runtime guardrails that inspect prompts and responses to prevent data leakage and stop threats before they reach our models and agents.[Cisco Hypershield](https://www.cisco.com/site/us/en/products/security/hypershield/index.html) will provide segmentation across the fabric, and with[Live Protect](https://www.cisco.com/site/us/en/products/networking/cloud-networking/nexus-platform/live-protect/index.html) on our[Nexus switches](https://www.cisco.com/site/us/en/products/networking/cloud-networking-switches/index.html) , we dynamically apply vulnerability shielding between patch windows, protecting our infrastructure against critical vulnerabilities without requiring reboots or taking production infrastructure offline. 5. **Real-time observability, tokenomics, and closed-loop routing:** [Splunk Observability Cloud](https://www.splunk.com/en_us/products/observability-cloud.html) serves as Cisco IT’s core metrics and traces repository for infrastructure, integrated with[Splunk Cloud Platform](https://www.splunk.com/en_us/products/splunk-cloud-platform.html) for log analytics and Splunk IT Service Intelligence (ITSI) for event management. Together, this unified foundation will deliver deep operational visibility, rapid Mean Time to Resolution (MTTR), and enable autonomous observability and closed-loop operations. We are currently in the process of expanding that to[Agent Observability](https://www.splunk.com/en_us/products/agent-observability.html) and Tokenomics, and will share more on this journey soon. Together, these capabilities have helped Cisco IT scale AI while improving the economics, security, and operational support of our infrastructure. ## Scaling enterprise-wide AI *and* impact The success of our modernized infrastructure isn’t just in our ability to now scale AI across the organization; the results show what that infrastructure makes possible across Cisco. - **Tripled GPU utilization:** Increased average GPU utilization from 15% to ~50%, with peak loads reaching 80%. - **6x compute expansion in same data center footprint:**  Scaled our active compute environment from ~250 to over 1,568 enterprise GPUs in under two years without acquiring new physical real estate. - **Proven on-prem efficiency:** Identified that running 80% of steady-state inference on-premises delivers the highest cost efficiency. We started with high-volume workloads like our enterprise super agent to standardize baseline model traffic on Cisco UCS. - **Enterprise-grade operations:**  Elevated operational support to our highest enterprise SLA tier, guaranteeing 99.99%+ availability for both external customer platforms (Cisco IQ) and 90,000 internal employees (MyAgent). ## What’s ahead: Scaling for the next decade of AI Our AI infrastructure transformation is a continuous journey and evolution. To ensure operational readiness, we are executing the following strategic pivots: - **Scaling for autonomous agent traffic:** With agent-generated traffic projected to be 450% higher per task than human traffic, we are transitioning to an AgenticOps model. We are deploying AI-driven monitoring and deep-stack telemetry across every network point to dynamically manage**high network utilization** and the longer-lasting, bi-directional flows required as MyAgent scales across the workforce. - **Executing the training-to-inference pivot:** As inference demand outpaces training by 60%, we are shifting our infrastructure focus toward software-speed traffic and upstream capacity. We are actively evolving our stack by: - **Dynamic resource pooling:** Upgrading job scheduling to eliminate silos and fill utilization troughs with background batch jobs. - **Automated efficiency:** Implementing automated GPU reclaims to maximize hardware ROI. - **Adaptive networking:** Deploying AI-specific KPI monitoring and adaptive architectures to guarantee low-latency user experiences. The true measure of our AI data center isn’t just how well it solves today’s business needs—it is having the architectural flexibility, economic control, and built-in governance to seamlessly scale for whatever comes next. More resources: