Cloud giants keep writing big checks to Cisco Systems. Orders for the networking leader’s AI infrastructure products from hyperscalers hit $1.9 billion in its fiscal third quarter of 2026. That compares with $600 million a year earlier. The jump signals how fiercely the largest cloud providers chase every advantage in building out massive GPU clusters.
Management now expects to book roughly $9 billion in such orders for the full fiscal year. That represents 4.5 times the prior year’s total. Revenue recognition from those hyperscaler deals should reach about $4 billion. For 2027 the company sees at least $6 billion in AI infrastructure revenue. These figures come straight from Cisco’s latest earnings analysis shared by AlphaSense.
But the story runs deeper than raw dollars. Cloud providers once focused purchases on raw compute. Now they buy the switches, silicon and optics that stitch thousands of accelerators together without choking on latency or power draw. Cisco executives have said as much in recent briefings.
The shift is unmistakable. What began as hyperscaler demand for training clusters has broadened. Inference workloads, agentic AI systems and multi-tenant environments all require different network behavior. Packet loss that barely registers in traditional data centers can slash GPU utilization by double digits when models run at scale.
Cisco’s response carries a name: Silicon One G300. The 102.4 terabit-per-second Ethernet switching chip landed in announcements earlier this year. It targets exactly the efficiency gains cloud operators crave. Liquid-cooled variants of Nexus 9000 and 8000 switches pair with it. The combination promises nearly 70 percent better energy efficiency in some configurations. Programmable features called Intelligent Collective Networking deliver 33 percent higher GPU utilization and 28 percent faster job completion times, according to company claims reported by SiliconANGLE.
Kevin Wolterweber, senior vice president and general manager of Cisco’s data center and internet infrastructure business, framed the moment clearly. “The last two or three years, we’ve mainly been focused on building out massive training clusters with hyperscalers,” he said. “What we’re starting to see now is a shift toward agentic AI workloads, and more adoption within enterprise service providers and a broader customer base.” That quote, pulled from the same SiliconANGLE coverage, captures the expansion beyond the usual suspects.
Neoclouds, sovereign cloud operators and even large enterprises now join the fray. They want AI infrastructure without reinventing the wheel or sacrificing multi-tenancy. Cisco’s Ethernet-first approach, heavily validated with NVIDIA, gives them an alternative to proprietary fabrics. The partnership keeps expanding. Unified architectures, Spectrum-X Ethernet running on Cisco silicon, and validated AI factories with NVIDIA Blackwell GPUs all point to tighter integration. Details surfaced in Cisco’s own June 2025 newsroom release.
Yet success brings fresh questions. Can Cisco maintain momentum as hyperscalers hit natural limits on power, real estate and capital? Five of the top hyperscalers posted triple-digit order growth with Cisco in the recent quarter. Service provider and cloud customer orders overall accelerated 105 percent year-over-year. Those numbers, again from the AlphaSense earnings recap, show concentration risk remains real.
Still, diversification helps. AI infrastructure orders from neocloud, sovereign cloud and enterprise customers added another $300 million in the quarter. A pipeline of roughly $3 billion sits in high-performance AI networking products. The company now guides total fiscal 2026 revenue between $62.8 billion and $63 billion. It calls the coming year its strongest ever.
Power efficiency sits at the center of every conversation. Data center operators face six-fold growth in electricity demand over the next decade, much of it AI-driven. Cisco’s 800-gigabit optical modules cut consumption in half. Liquid cooling trims it further. These gains matter when clusters span tens of thousands of GPUs and every watt counts toward utilization rates above 90 percent.
Networking has become the new bottleneck. GPUs grab headlines. The fabric that connects them determines whether those expensive chips deliver promised returns. Cloud providers learned this lesson the hard way during earlier GPU shortages. They now treat high-speed, low-latency Ethernet as table stakes.
Cisco isn’t alone in the chase. Broadcom, Arista Networks and NVIDIA’s own networking efforts compete fiercely. Yet Cisco’s incumbency in service provider routing, its optics business through Acacia, and its software management tools provide stickiness. The Nexus One platform unifies on-premises and cloud AI fabrics under one management pane. That matters to customers who refuse to manage separate domains.
Partnerships extend the reach. Saudi Arabia’s HUMAIN works with Cisco on open, scalable AI infrastructure using Nexus switches, UCS servers, Hypershield security and Splunk observability. UAE-based G42 collaborates on broader AI innovation. Cisco serves as preferred technology partner for the Stargate UAE project. These deals, highlighted in Cisco’s official newsroom, show how sovereign AI ambitions create new buyers.
Jeetu Patel, Cisco’s chief product officer, described the larger trend. “As billions of AI agents begin working on our behalf, the demand for high-bandwidth, low latency and power efficient networking for data centers will soar. Cisco is at the forefront.” His words appear in the same Cisco release.
The original signal came months earlier. Cloud providers told Cisco they planned heavier purchases of both custom silicon and high-end switches tuned for AI traffic. That briefing, reported by The Information, proved prescient. Orders materialized. Revenue followed. And the cycle repeats.
Recent coverage reinforces the trajectory. A Yahoo Finance analysis published just yesterday noted Cisco’s raised fiscal 2026 AI order outlook now sits at $9 billion. The piece also flagged internal deployment of personalized AI agents across Cisco’s own workforce of roughly 90,000 employees. Eating your own dog food at that scale sends a message. The article is available at Yahoo Finance.
Service providers see opportunity too. Cisco estimates network capacity requirements could triple within three years as AI traffic moves from data centers to campus, branch and edge locations. Telcos can evolve from bandwidth wholesalers to AI infrastructure partners. Cloud Control, Cisco’s unified management platform with built-in AI agents, offers them a ready operational foundation. Fierce Network covered the push in early June.
So what happens next? Hyperscalers will keep buying. They must, if they want to stay competitive in model training and inference at global scale. Cisco will push silicon speeds higher, optics denser and software smarter at orchestration. The question is whether utilization gains can outrun the exploding cluster sizes.
Executives sound optimistic. Wolterweber sees the customer base broadening beyond traditional hyperscalers. Patel expects demand to soar with agentic systems. The numbers back them up so far. $1.9 billion in a single quarter. $9 billion targeted for the year. Those aren’t incremental bets. They reflect conviction that networking decides who wins the AI race.
And the race shows no signs of slowing. New neocloud entrants, national AI projects and enterprise experiments all feed the same supply chain. Cisco sits near the middle of it, selling the picks and shovels for an infrastructure boom that keeps redefining its own scale. How long that boom lasts depends on whether the models deliver economic value fast enough to justify the capital pouring in. For now, the orders keep coming.