Cisco's AI Networking Orders Just Doubled Its Own Forecast to $9 Billion

Cisco's stock jumped 17% in a single trading session Thursday. That's the kind of move you usually see after a merger announcement or a surprise acquisition. What actually happened was a quarterly earnings report, one that showed AI infrastructure demand so strong it forced the company to tear up its own projections and write new ones.

The numbers are striking. Cisco reported record Q3 revenue of $15.84 billion, up 12% year over year, beating Wall Street's $15.56 billion estimate. But the figure that sent the stock soaring was this: Cisco has received $5.3 billion in hyperscaler and AI infrastructure orders so far this fiscal year. And the company now expects that figure to reach $9 billion by year-end, up from a previous forecast of $5 billion.

That's not a modest upward revision. That's the company effectively saying its AI business grew faster than it predicted by almost 80%.

Behind Cisco's Record AI Order Numbers

The core of Cisco's AI business is networking: the switches, routers, and optical interconnects that move data between GPUs at the speeds AI training and inference require. When hyperscalers like Microsoft, Amazon, and Google build AI data centers, they don't just buy chips. They buy the plumbing that connects them.

Cisco makes that plumbing. And right now, demand for it is accelerating faster than the company expected.

Networking product orders grew more than 50% year over year in Q3. Data center switching orders, a product category that directly tracks AI cluster builds, rose more than 40%. Total product orders across the company climbed 35% year over year. Those are not typical enterprise hardware growth rates. Those are breakout numbers.

CEO Chuck Robbins attributed much of the strength to specific product wins. "The increase in AI orders is due to strong design wins and customer confidence in Cisco's Silicon One and optics products," he said in prepared remarks.

Silicon One is Cisco's proprietary networking silicon, a custom chip designed to handle the dense, high-bandwidth traffic patterns inside AI data centers. Hyperscalers have been moving toward custom silicon for years. Google has its TPUs, AWS has Trainium, Microsoft has its Maia AI accelerators. Cisco's bet is that networking silicon requires the same specialization. Silicon One is the company's answer, and Q3 suggests that bet is paying off.

The optics story matters just as much. Data centers at scale need fiber-optic transceivers to move data between racks at the speed of light. As AI clusters grow larger and GPUs demand more interconnect bandwidth, the optics business becomes a meaningful revenue line. Cisco has built a significant position here, and hyperscalers are apparently buying in.

Robbins used a phrase on the earnings call worth sitting with: "networking supercycle." It's a deliberate framing. It says this isn't just one good quarter. It says we're in a multiyear expansion driven by a structural shift in how computing works.

He's not wrong, at least in the near term. AI training runs increasingly require clusters of thousands of GPUs operating in tight coordination. The interconnect fabric between those GPUs, the switches, cables, transceivers, and routing logic, has to handle latency in microseconds and bandwidth in terabits per second. That infrastructure doesn't exist at scale yet. Most of it is being built right now.

Cisco, which spent years struggling to grow in an era of software-defined networking and cloud consolidation, suddenly finds itself with infrastructure the AI era urgently needs. The company's Q4 guidance reflects that confidence. Cisco is forecasting $16.8 billion in revenue and $1.17 EPS for the quarter, both well above Wall Street's estimates of $15.82 billion and $1.07. Six analyst firms raised their price targets within hours of the report. Per Fortune's coverage, the stock briefly touched $122 in after-hours trading, a price level Cisco hasn't seen in years.

Growth and Layoffs at the Same Time

Cisco is also cutting 4,000 jobs, roughly 5% of its global workforce.

That sounds contradictory until you look at what the company is actually doing. Robbins framed the cuts as a deliberate shift toward where demand is strongest. "The companies that will win in the AI era will be those with focus, urgency, and the discipline to continuously shift investment toward the areas where demand and long-term value creation are strongest," he said.

The jobs being cut likely sit in legacy product areas: traditional enterprise networking, older security products, parts of the business that aren't growing at 50% a year. The headcount is being redeployed toward silicon design, optics engineering, and AI-specific networking research and development.

It's the same pattern playing out across the tech industry. AI creates enormous demand in specific areas while making other areas structurally less valuable. The companies that adjust quickly capture the growth. The ones that try to protect legacy lines get left behind.

Cisco is reorganizing its business around what Robbins called "silicon, optics, security, and employees' use of AI across the company." That's a company betting its future on the supercycle being real.

The Real Signal in $9 Billion of AI Orders

Cisco's order book gives you a concrete window into where hyperscalers are spending right now. When Amazon, Microsoft, Google, and Meta collectively commit $5.3 billion in orders to a single networking vendor, and that vendor responds by doubling its full-year forecast, it tells you something precise about the pace of AI infrastructure investment.

The AI build-out has been described in abstract terms often enough. Jensen Huang at NVIDIA called it "the single largest infrastructure buildout in human history." A single 500-megawatt data center takes 30,000 truckloads to construct, he said at Davos earlier this year. These numbers convey scale without getting specific.

Cisco's earnings give you something more concrete: actual order flow, actual revenue, and actual guidance increases from a company that sells into every major hyperscaler's data center expansion. The $9 billion forecast for Cisco's AI networking business alone doesn't include storage, power, cooling, chips, or the dozens of other components that go into an AI data center. It's one slice of a much larger spend. And it grew 80% in a single fiscal year.

The other side of that same story is utilization. A recent look at enterprise AI infrastructure found that GPU clusters at many companies average just 5% utilization, suggesting that AI compute is being procured faster than it's being used. Hyperscalers buying networking gear from Cisco are presumably running their clusters harder than the average enterprise, but the dynamic is worth watching as the cycle matures.

On the competitive side, Cisco doesn't own this market alone. Arista Networks, which has staked much of its recent growth on AI networking, has been outperforming for the past two years. Ethernet-based AI clusters are increasingly competing with InfiniBand, NVIDIA's preferred interconnect standard, and Arista leads the Ethernet camp. But Cisco's Silicon One position and its optics portfolio give it a credible path into the AI data center that it didn't have five years ago. The company's scale, it ships into more enterprise and service provider networks than any other vendor, also gives it an advantage when hyperscalers want a single vendor managing global deployments.

A 17% single-day move for a company valued at over $100 billion is unusual. Cisco hasn't moved like this since 2002, when the company was still recovering from the dot-com collapse. The move reflects two things. First, genuine surprise at the size of the beat and the scale of the guidance raise. Second, a re-rating of what Cisco's business is worth in an AI-infrastructure world. The market spent years treating Cisco as a slow-growth legacy hardware company. Thursday's report forced a reconsideration.

Whether the stock holds those gains depends on whether Q4 delivers on the $16.8 billion revenue promise, and whether AI infrastructure orders continue to come in above $1.9 billion per quarter. Based on what hyperscalers are currently building, there's reason to think that's achievable. But markets are not always patient, and the networking supercycle narrative will need to prove out across multiple earnings reports before it becomes consensus. For now, Thursday's numbers made a compelling opening argument.

*For a broader look at how AI hardware spending is reshaping the technology stack, from chips to cloud to capacity planning, see our guide to AI Infrastructure in 2026.*