Cisco FY26 Earnings: AI Infrastructure Orders Hit $9.3B

/ AI, Application Development, Infrastructure, Networking

The News

Cisco has reported their fiscal year 2026 results, delivering what the company described as a record year across the board. Total revenue reached $63.3 billion, up 12% year-over-year, with product revenue climbing 16% to $48.3 billion. The standout headline, though, is AI infrastructure: Cisco took $4 billion in AI infrastructure orders from hyperscalers in Q4 alone, bringing the FY26 total to $9.3 billion, roughly a 4.5x increase year-over-year, alongside three new hyperscale AI design wins and the launch of Antares, a family of open-weight small language models.

Analyst Take

The networking supercycle is real, and Cisco is riding it hard

Skeptics spent much of 2024 and 2025 debating whether AI-driven networking demand was sustainable or a one-time pull-forward. These results answer that question. Networking product orders grew more than 40% year-over-year in Q4, marking the eighth consecutive quarter of double-digit growth. That is not a bubble effect. That is a structural shift in how enterprises and hyperscalers are building out compute fabrics to support large-scale AI workloads, and Cisco has positioned itself squarely in the middle of it. Campus networking orders hit a record, up 20% year-over-year, suggesting the refresh cycle has broadened well beyond data center AI buildout and into the enterprise edge. For ITDMs, this means the capital expenditure pressure to modernize campus and branch infrastructure is intensifying across the competitive landscape, not just at the hyperscaler tier.

The $9.3 billion AI infrastructure number deserves scrutiny, and respect

The FY26 AI infrastructure order figure of $9.3 billion is attention-grabbing, but the composition matters as much as the total. Three new hyperscale design wins in a single quarter, covering scale-across, scale-out, and optical transport, signals that Cisco is not winning on a single product bet. Optical in particular is a competitive battleground, where competitors like Ciena and Infinera have historically held ground. Winning an optical line system design at hyperscale is a meaningful foothold. For developers and platform architects, the implication is that the physical network layer serving AI training and inference clusters is consolidating toward fewer, larger vendors with integrated portfolios rather than best-of-breed point solutions.

Cisco Cloud Control and Antares: strategic bets that need time to prove out

Two announcements warrant particular attention from platform teams. Cisco Cloud Control, framed as a single management plane and unified data layer across all Cisco products, is a direct response to the fragmentation problem that plagues enterprise IT at scale. ECI Research’s Nutanix Kubernetes Operations Benchmark Study found that 44.1% of respondents said the improvement they most wanted in their Kubernetes environment was the ability to “enable a fully self-service, zero-ticket developer experience.” A unified control plane across networking, security, and compute is a necessary precondition for that kind of experience. Whether Cisco Cloud Control can deliver on that vision across a portfolio as sprawling as Cisco’s is the right question to ask, and it will take several product cycles to answer credibly.

Antares, the open-weight SLM family, is a different kind of bet. Cisco joining OpenAI’s Daybreak program and Anthropic’s Project Glasswing as a founding member signals an intent to embed AI deeply into the network operations layer, not just sell pipes to AI infrastructure. Open-weight models are strategically important here: they allow enterprises with data residency or compliance constraints to run inference on-premises. ECI Research’s benchmark study found that 41.8% of respondents cited “complexity of orchestrating data pipelines with container infrastructure” as the primary obstacle to scaling AI infrastructure on Kubernetes. Cisco’s play with Antares, if executed well, could reduce that orchestration burden at the networking layer, though the model-to-operations workflow integration is still early.

Looking Ahead

Cisco’s FY27 trajectory hinges on two questions. First, can hyperscaler AI infrastructure orders sustain their current pace, or does the order book begin to normalize as the initial wave of AI data center buildout matures? The 4.5x year-over-year growth rate in AI orders is extraordinary but not infinitely repeatable. What matters for the next four to six quarters is whether enterprise customers, not just hyperscalers, begin placing meaningful AI infrastructure orders at scale. The campus refresh cycle and next-generation product ramp suggest Cisco is building the enterprise commercial motion to absorb that demand when it arrives.

Second, the software and platform story needs to develop faster. Services revenue was flat year-over-year at $15 billion in FY26, a persistent soft spot in an otherwise strong report. Cisco Cloud Control, Antares, and the IDP integrations tied to agentic workplace products all need to convert into recurring software revenue over the next two to three years. If Cisco can demonstrate that its management plane and AI tooling generate measurable operational savings for enterprise customers, rather than simply selling more hardware into a hot cycle, the multiple expansion case becomes compelling. If not, Cisco risks being valued as a cyclical infrastructure vendor rather than the platform company it is clearly trying to become.

Author