Key Points
- Enterprise Nexus switch orders for AI deployments rose more than 85 per cent sequentially
- Cisco expects hyperscaler AI infrastructure revenue to reach $7.5 billion in fiscal 2027
- 145,000 customer support cases were resolved entirely by AI without human intervention
A chief information officer (CIO) in Mumbai weighs whether to run a customer service model in the public cloud or on premises. A government department in Delhi asks whether its data can leave Indian soil. A manufacturing firm calculates whether the cost of tokens will outweigh the value of the insights. These decisions, multiplied across thousands of organisations, are beginning to reshape enterprise infrastructure spending.
Cisco Systems is seeing these questions arrive at the centre of customer conversations. The networking company’s chief executive officer (CEO), Chuck Robbins, told analysts following the company’s fiscal fourth-quarter results that discussions that began around the economics of token consumption were rapidly expanding into questions around open-weight models, data security, sovereignty and agentic AI.
“Is it a cost issue, a security issue, a sovereignty issue? The answer is yes,” Chuck said while discussing the enterprise market.
The shift suggests the next phase of enterprise AI spending may be less about simply gaining access to large language models and more about deciding which workloads belong in public cloud, sovereign cloud, private data centres or at the edge. Cisco expects customers to make workload-by-workload decisions on which models to use and where to run them.
Enterprise shift
The change is showing up in Cisco’s order book. Enterprise Nexus switch orders tagged for AI deployments rose more than 85 per cent sequentially in the fourth quarter, while overall data centre networking orders increased more than 35 per cent from a year earlier.
Outside the largest hyperscalers, Cisco booked more than $400 million in AI infrastructure orders from neocloud, sovereign and enterprise customers during the quarter, taking the full-year total from those segments to more than $1 billion.
The figures remain smaller than Cisco’s hyperscaler AI business, where orders reached $4 billion in the fourth quarter and $9.3 billion for fiscal 2026. Cisco said it generated about $4 billion in hyperscaler AI infrastructure revenue during fiscal 2026 and expects that figure to rise to $7.5 billion in fiscal 2027.
Chuck said continued use of cloud-based models would support demand from cloud providers, while greater adoption of open-weight or on-premises models would require enterprises to invest more in private data centre networking.
For government departments, public sector organisations and regulated enterprises, the more consequential development may be the widening set of deployment choices. Sensitive data, latency requirements, model economics and regulatory constraints can make a single cloud-first approach difficult for some workloads.
Cisco is betting that this will create demand for infrastructure that can connect and secure AI workloads across multiple environments rather than forcing organisations into one deployment model.
Chuck said enterprises running GPU clusters on premises or at the edge would require low-latency, high-bandwidth networks alongside security, observability and automation. He also pointed to the operational implications of deploying thousands of AI agents across infrastructure, arguing that “performance, latency and security requirements would increase as agentic systems scale”.
Cisco’s own AI use offers a preview of enterprise patterns
The company’s own use of AI provides an indication of where it believes enterprise adoption is heading. Cisco said 145,000 customer support cases were resolved entirely by AI without human intervention during fiscal 2026. Its proprietary on-premises AI assistant, Circuit, handled more than 75 million prompts in the fourth quarter alone.
Circuit runs on Cisco’s Secure AI Factory infrastructure and routes tasks to different large language models depending on the requirement, a design Cisco says helps improve GPU utilisation and control token consumption. The architecture shows the same model-selection problem Cisco says customers are beginning to face: not every task needs the same model and not every workload needs to leave the organisation’s infrastructure.
Cisco is also bringing AI into day-to-day network and security operations. Its Cloud Control platform is designed as a common management layer across networking, security, compute and observability, with AI Canvas allowing human operators and AI agents to investigate operational issues using the same underlying context.
Also Read | Cisco posts record Q4 revenue of $17.3 billion as AI orders surge
Nearly 4,500 enterprises had signed up for Cisco Cloud Control following its launch, according to Chuck. He cited an example in which AI Canvas identified the access point and root cause behind dropped video calls within minutes after a network engineer had spent more than eight hours troubleshooting the problem.
The security side of Cisco’s AI strategy is developing in parallel. More than 1,500 customers bought newer security products including Secure Access, XDR, HyperShield and AI Defense during the fourth quarter.
Cisco said customers were increasingly looking for security architectures that covered users, applications and AI agents rather than treating AI as a separate security domain. It has also introduced Antares, a family of open-weight small language models designed to identify the location of known vulnerabilities inside software codebases.
The approach aligns with a broader industry move towards smaller, task-specific models that can run locally, reducing both inference cost and the need to send sensitive code to external services.
AI spending
For CIOs and CISOs, Cisco’s earnings call also contained a warning about budgets. Chuck said customers were largely reprioritising existing spending rather than simply expanding IT budgets without limit. But he said AI readiness was increasingly being treated in a similar way to cybersecurity, as spending that organisations considered difficult to defer.
By the numbers
- $7.5B
- Expected Cisco hyperscaler AI infrastructure revenue for fiscal 2027
- 85%
- Sequential growth in enterprise Nexus switch orders for AI
- 145,000
- Support cases resolved by AI without human intervention
That distinction matters for public sector technology leaders managing legacy infrastructure and fixed budget cycles. AI programmes may increasingly compete with other technology projects for funding, while at the same time exposing weaknesses in networks, security controls and ageing equipment that were previously easier to postpone.
Cisco’s argument is that AI will therefore drive spending beyond GPUs and servers. Distributed AI clusters need more network bandwidth, enterprise deployments need secure connectivity and observability, while AI agents create additional operational and security demands.
Networking requirement becoming acute
For the largest AI operators, the networking requirement is becoming particularly acute. Cisco estimates that traffic generated by scale-across AI architectures, which connect computing resources across multiple data centres, could be about 14 times that of traditional data centre interconnect traffic. The company has been expanding its Silicon One systems and coherent optics portfolio to target that market.
There are reasons for caution in reading the numbers. Much of Cisco’s AI infrastructure growth is still concentrated among hyperscalers, and the company’s forecast of $7.5 billion is for hyperscaler AI infrastructure revenue rather than the broader enterprise AI market. Cisco also expects the heavier mix of hardware associated with the AI networking build-out to put some pressure on gross margins.
Still, this points to a broader change in enterprise AI adoption. The question for technology leaders is moving from whether to use AI to how to place, connect, operate and secure it.
For Cisco, that shift expands the AI opportunity beyond selling the network beneath large GPU clusters. It is seeking a role in the infrastructure that connects AI, the security controls that govern it and the operational systems that increasingly use AI themselves.
Your Questions, Answered
What factors are driving enterprise AI deployment decisions?
Cisco says enterprises are weighing cost, security, latency and data sovereignty when deciding whether to run AI workloads in public cloud, sovereign cloud, private data centres or at the edge. Most organisations are making these decisions workload by workload rather than adopting a single approach.
How large is Cisco's AI infrastructure business?
Cisco generated about $4 billion in hyperscaler AI infrastructure revenue during fiscal 2026 and expects that figure to rise to $7.5 billion in fiscal 2027. It also booked more than $1 billion in AI orders from enterprise, sovereign and neocloud customers during the year.
How is Cisco using AI in its own operations?
Cisco's AI assistant Circuit handled more than 75 million prompts in the fourth quarter and resolved 145,000 customer support cases without human intervention during fiscal 2026. The system routes tasks to different models based on requirements.
What does this mean for IT budgets?
Cisco's CEO said customers are reprioritising existing spending rather than expanding budgets without limit, but AI readiness is being treated similarly to cybersecurity as spending difficult to defer. AI programmes may compete with other technology projects for funding.
