Cisco used its annual Partner Technical Exchange in Bangkok this week to deliver a message that was equal parts urgent and calculated: the AI infrastructure market is moving faster than most enterprises can keep up, and the company intends to be the architecture they build on.

Across six back-to-back sessions spanning compute, security, observability, and networking, Cisco’s engineering and product leaders painted a picture of an industry at an inflection point. Only 28 per cent of organisations believe their current infrastructure can handle AI workloads, according to Cisco’s own 2025 AI Readiness Index. Yet partner-delivered AI services are projected to reach US$267 billion by 2030, growing at 35 per cent compound annual growth. That gap, Cisco argued repeatedly, is where the channel earns its keep.

The tone was set from the opening keynote, where Cisco’s APJC Partner Sales leadership laid out their FY26 strategic priorities in three columns that read like a maturity model for the channel itself: Develop, Grow, Achieve. The “Develop” pillar focused on building key technology practices, lifecycle practices, and cross-platform capabilities. The technical depth partners need to sell and deliver AI infrastructure. “Grow” centred on driving alignment between Cisco sales and partner teams, accelerating partner-driven business, and transforming traditional resale motions into something more strategic. And “Achieve” pointed to the endgame: transitioning partners to the Cisco 360 Program, enhancing partner experience and mindshare, and building the ecosystem for the future. Underneath all three sat the foundation pillars of Talent Development, Sales and Business Development, and Technical Leadership, the organisational muscle Cisco expects partners to build. It was a clear signal that the days of transactional box-moving are numbered; Cisco wants partners who can architect, implement, and operate AI-era infrastructure end to end.

THE AI FACTORY THESIS

The centrepiece of the day was the Cisco Secure AI Factory with NVIDIA — the company’s modular, security-first reference architecture for enterprise AI infrastructure. Presented in detail by Chris Castaway in the afternoon, the architecture is not theoretical. Sharon AI and Cisco launched Australia’s first deployment in February 2026, putting 1,024 NVIDIA Blackwell Ultra GPUs into production at NEXTDC data centres using Cisco UCS servers, Nexus Hyperfabric networking, and VAST Data storage.

The Secure AI Factory is built around AI PODs — modular building blocks that integrate compute, networking, GPUs, and storage into validated, repeatable designs. Customers can choose between Cisco or NVIDIA Spectrum-X Ethernet for backend GPU connectivity, and between on-premises management via Nexus Dashboard or cloud management via Hyperfabric. The architecture is validated against NVIDIA’s Enterprise Reference Architecture, which means partners can deploy with confidence rather than engineering from scratch.

Without the network, there is no AI, and there is no scale.

What made the sessions compelling was the depth of the networking conversation. AI doesn’t just add workloads to existing networks — it fundamentally changes how networks need to be designed. Over the past sixteen years, data centre traffic evolved from predictable north-south patterns to east-west flows driven by virtualisation. AI introduces a third shift: high-bandwidth, low-latency, always-on traffic across three concurrent network domains — front-end, storage, and dedicated GPU backend — all carrying simultaneous bidirectional flows.

The implication is stark. In AI workflows, any packet loss, congestion, delay, or link failure can cause an entire training job to fail, wasting compute resources that are already scarce and expensive. Cisco positioned its Silicon One ASICs, announced 1.6 terabit switches, 256 GB packet buffers, and liquid-cooled systems as the answer to these demands. The Silicon One G300, unveiled at Cisco Live EMEA in Amsterdam last month at 102.4 Tbps, is purpose-built for massive AI cluster buildouts.

SECURITY STOPS BEING AN OVERLAY

If the AI Factory was the headline act, the security story was the one with the sharpest edge. Cisco’s afternoon and late-afternoon sessions on AI Defense and the Hybrid Mesh Firewall made a forceful argument that the era of security as a bolt-on overlay is finished.

Cisco AI Defense, first announced in January 2025 and significantly expanded in February 2026, is purpose-built for threats that traditional security products cannot detect. Prompt injection, model poisoning, data exfiltration through AI outputs, supply chain compromise of model files — these are not theoretical risks. Cisco’s 2025 Cybersecurity Readiness Index found that 86 per cent of enterprises experienced an AI-related security incident within the past year.

The February expansion introduced AI Bill of Materials for supply chain governance, runtime protections for agentic AI tool interactions via the Model Context Protocol, and multi-turn algorithmic red teaming that replaces the traditional seven to fifteen week manual process. AI Defense now integrates with NVIDIA NeMo Guardrails’ open source framework for runtime protection, and its detections feed directly into Splunk Enterprise Security for unified threat detection and response.

In the age of AI, safety and security are prerequisites for adoption, and AI agents bring a whole new set of challenges. — Jeetu Patel, Cisco President and Chief Product Officer

The Hybrid Mesh Firewall session was equally significant, though for different reasons. Cisco’s approach extends beyond the traditional vendor definition of hybrid mesh — physical, virtual, and cloud firewalls. Cisco adds intent-based policy deployment across third-party firewalls including Fortinet and Palo Alto, smart switches with Layer 3/4 firewalling on Nexus, and workload agents using eBPF from Isovalent. The Mesh Policy Engine, managed through Security Cloud Control, translates user-to-application intent into device-specific syntax for each enforcement point.

The speaker was emphatic that Security Cloud Control is not a device manager, it is an intent-based policy instructor and enforcer. Since general availability, 8,300 customers have adopted the platform. Innovation continues on both cloud-managed and on-premises Firewall Management Center platforms, with cloud updates arriving faster.

One innovation worth watching is LiveProtect. Using the Isovalent Tetragon agent integrated directly into NX-OS, LiveProtect deploys eBPF kernel-level controls that shield Nexus switches from known CVEs, privilege escalation, control plane DDoS, without reboots, patches, or downtime. For any organisation running a large Nexus fabric with strict change windows, this changes the patching calculus entirely.

OBSERVABILITY GETS AN AI LAYER

The Splunk session was short, roughly eight minutes, but dense with implication. The core argument: monitoring only infrastructure is no longer sufficient. When AI agents are making decisions that affect business outcomes, you need to observe the agent layer too.

Cisco and Splunk now frame AI observability around two pillars. AI infrastructure monitoring tracks the health and performance of compute, network, storage, and GPUs through Splunk Observability Cloud. AI agent and application monitoring captures metrics like tokenisation rates, time to first token, agent utilisation, and cost per inference. Both pillars sit on top of the OpenTelemetry collector, where Cisco drew a sharp competitive line: unlike vendors who ship modified proprietary versions, Splunk uses and contributes to the standard open source release. The downloadable version works directly with Splunk.

The Splunkbase marketplace offers approximately 2,100 to 2,200 integrations, with Cisco integrations elevated to gold standard status. At Splunk .conf25 in September 2025, Cisco announced agentic AI-powered Splunk Observability, deploying AI agents to automate telemetry collection, detect issues, identify root causes, and recommend fixes. Purpose-built dashboards for AI POD infrastructure now track GPU utilisation, memory usage, power, and network performance with data integrated from Cisco Intersight and Nexus.

There’s not a single part of that stack we haven’t covered, and I truly believe there’s no other vendor in market that can provide you that level.

COMPUTE FROM CORE TO EDGE

The compute session rounded out the infrastructure picture. Cisco has reduced PCB count inside servers by 35 per cent, from 17 boards to 11, directly benefiting customers trying to fit GPUs into their environments amid RAM shortages and price escalation. Built-in telemetry services now collect power, temperature, CPU, GPU, and network metrics per site, feeding into the broader Splunk observability stack

The more significant announcement was the Unified Edge Platform, Cisco’s answer to the operational mess of geographically distributed edge environments where switches, routers, and wireless controllers sit as isolated islands. The platform extends consistent data centre operations to the edge through integrated compute and networking nodes, with server-to-server networking built into the chassis. For organisations looking to run inferencing closer to where data is generated, retail floors, hospital wards, manufacturing lines, this is the hardware play.

THE NVIDIA FACTOR

Throughout the day, the NVIDIA partnership was referenced with the frequency of a strategic talking point that has become operational reality. Cisco is the only networking vendor co-developing IP and hardware with NVIDIA, the partnership covers Silicon One, NVIDIA Spectrum ASICs running NX-OS, and validated 8,000-GPU node environments on Ethernet with Nexus 9300 switches. The solution is NCPRA certified and validated.

This is not a reseller arrangement. Cisco and NVIDIA engineers are building silicon and software together, and Cisco can offer something no other networking vendor can: operational consistency via NX-OS across front-end, back-end, and storage networks, preserving customer automation scripts and staff skills regardless of which fabric they deploy.

THE SUPPLY CHAIN ELEPHANT

The opening session didn’t shy away from uncomfortable territory. The speaker introduced the term “Ramageddon” to describe the current memory supply chain crisis, joking that even their children know about it because laptop deliveries have stalled. Trade disputes affecting Australia, Japan, and Korea, export restrictions constraining where GPU infrastructure can be deployed, and tariff uncertainty are all forcing customers to rethink procurement timelines.

The practical advice was blunt: some customers should extend existing hardware for another year, prioritise network refreshes over server refreshes, and treat GPU allocation as a strategic investment decision rather than a routine procurement cycle. For startups and new organisations, GPUs now represent core strategic value in ways that fundamentally reshape how infrastructure investment decisions are made.

WHAT IT MEANS FOR THE CHANNEL

The commercial throughline was unmistakable. Cisco’s AI infrastructure business cleared US$2 billion in 2025, against a target of US$1 billion, and is tracking toward US$3 billion in 2026. The Cisco 360 Partner Program, launched in January 2026, includes a dedicated Secure AI Infrastructure specialisation and a new AI Infrastructure Specialist Certification within the CCNP Data Centre track.

For partners in the Australian market, the Sharon AI deployment provides a local reference point. The Secure AI Factory architecture, Nexus Hyperfabric cloud management, and the integration with sovereign data centre providers like NEXTDC create a repeatable engagement model for government, financial services, and healthcare customers who need AI infrastructure but cannot send their data offshore.

The sessions left little ambiguity about where Cisco is placing its chips. AI is not a feature on top of existing products, it is the organising principle for the entire portfolio. Every networking decision becomes a security decision. Every compute platform ships with observability built in. Every edge deployment is an extension of the AI factory. Whether the market moves as fast as Cisco expects is the open question. Whether Cisco is ready if it does is not.