Cisco’s AI Networking Boom Faces Its First Bust Test
Cisco has become a google news fixture after AI infrastructure orders accelerated, yet that success exposes the company to a sharp spending reversal.
A Google News headline points to the uncomfortable question behind Cisco’s recent momentum. What happens when hyperscalers stop expanding AI clusters at their current pace?
That question does not require believing that artificial intelligence will disappear. The Internet survived the dot-com crash, while network demand continued growing. However, equipment suppliers still suffered when customers discovered they had purchased too much capacity too quickly.
Cisco now sits near the center of another infrastructure cycle. Its switches, routing silicon, and optical systems connect the processors inside and between large AI data centers. The company benefits whenever cloud operators add clusters, expand bandwidth, or connect facilities across regions.
The resulting order growth is real, based on Cisco’s reported figures. The durability of that growth remains unproven. Orders are not revenue, and revenue does not guarantee stable margins.
That gap creates the central conflict. Cisco is presenting AI networking as a durable expansion, while skeptics see a capital-spending boom vulnerable to overbuilding. Both interpretations can fit the same headline numbers.
What the Google News Warning Actually Changes
The Cisco AI bust argument is not a forecast that networking demand disappears. It is a warning about the timing and quality of demand.
Cisco entered fiscal 2026 with more than $2 billion in fiscal 2025 AI infrastructure orders from webscale customers. That total was more than twice its original annual target. The company reported over $800 million during the fourth quarter alone.
Those numbers appeared in Cisco’s fiscal 2025 results. They established that Cisco had moved beyond talking about a future AI opportunity. Major customers were already buying its equipment.
Momentum strengthened during fiscal 2026. Cisco reported $1.3 billion in hyperscaler AI infrastructure orders during its first quarter. It then reported $2.1 billion during the second quarter.
By the third quarter, Cisco had booked another $1.9 billion. The fiscal year-to-date total reached $5.3 billion, compared with about $2 billion across the previous fiscal year.
Cisco consequently raised its expected fiscal 2026 hyperscaler AI orders to about $9 billion. It also raised expected hyperscaler AI revenue to about $4 billion.
That sequence explains why Cisco keeps appearing across google news coverage of AI infrastructure. The company has attached measurable orders to a market narrative that previously centered on chip vendors.
However, the headline warning changes how readers should interpret those figures. Accelerating orders demonstrate current demand, not the lifetime value of the installed equipment.
Hyperscalers often place large orders before completing new data centers. Suppliers must obtain components, reserve manufacturing capacity, and build inventory before customers deploy everything.
This creates a lag between a purchase order, delivered equipment, recognized revenue, and eventual cash generation. A boom can therefore look strongest in reported orders shortly before spending growth slows.
Cisco also measures different customer groups separately. Its headline hyperscaler totals do not include neocloud, sovereign, and enterprise orders. That separation gives investors useful detail, but it complicates broad claims about total AI demand.
The company reported roughly $900 million in fiscal 2026 year-to-date AI orders from those additional customer groups through its third quarter. It also described a pipeline of about $3 billion.
A pipeline represents potential business under discussion. It is not a signed order, recognized revenue, or guaranteed deployment.
The key change, then, is interpretive. Cisco has established itself as an AI infrastructure supplier. The next test is whether its current order surge develops into repeatable revenue without leaving excess inventory behind.
Cisco AI Infrastructure Orders Are Growing Faster Than Revenue
Cisco’s strongest evidence is the acceleration in orders, while its greatest uncertainty is how those orders convert into profitable sales.
Cisco’s third-quarter presentation showed the scale of the acceleration. Hyperscaler AI infrastructure orders reached $5.3 billion through three quarters of fiscal 2026.
Cisco expected approximately $4 billion in related fiscal 2026 revenue. The difference does not mean customers canceled purchases. It reflects delivery schedules and accounting recognition across different periods.
However, that difference matters when evaluating a possible Cisco AI bust. Orders provide an early signal, but revenue conversion determines whether the business supports Cisco’s broader financial results.
The company reported fiscal third-quarter revenue of $15.84 billion, up 12 percent from the previous year. Networking revenue reached $8.82 billion, a 25 percent increase.
Product orders rose 35 percent across the company. They still increased 19 percent when Cisco excluded hyperscale orders, suggesting that growth was not limited to a few cloud buyers.
Data center switching orders increased more than 40 percent. Campus networking orders rose more than 25 percent, while wireless orders increased more than 40 percent.
These results support the optimistic case. Cisco is benefiting from AI clusters, but it is also serving customers replacing older campus and data center equipment.
The company said its pre-Catalyst 9000 installed base represented tens of billions of dollars in equipment approaching end-of-support dates. That replacement cycle can continue even if hyperscaler spending becomes less aggressive.
Cisco’s AI products also span more than one category. Silicon One is its networking-chip architecture for switches and routers. Acacia supplies optical components that carry high-speed traffic between systems and facilities.
During the third quarter, Cisco reported five new hyperscaler design wins. Three involved systems, while two involved optical products.
Two system wins used the Silicon One P200 for scale-across deployments. Scale-across networking connects AI workloads operating in different clusters or data centers.
A separate system win used the G200 for scale-out networking. Scale-out systems connect processors within a growing cluster, allowing operators to add computing capacity.
Cisco also reported more than $1 billion in third-quarter Acacia orders. It said those orders had grown by a triple-digit percentage from the previous year.
Optical connectivity becomes more important as operators distribute AI hardware across buildings and regions. Distance increases the need for faster links, better power efficiency, and reliable traffic management.
That architectural shift supports demand beyond the initial purchase of AI accelerators. A cluster cannot deliver useful computing capacity when its processors spend too much time waiting for data.
Still, Cisco’s growth rate creates difficult comparisons. A business that expands several times over cannot repeat the same percentage increase indefinitely.
The relevant question is not whether fiscal 2026 orders are impressive. They are. The question is what level of recurring demand remains after hyperscalers complete their largest construction programs.
The AI Supercycle Promise Meets Capital-Spending Reality
Cisco’s primary opponent is not another networking vendor. It is the possibility that infrastructure commitments outrun profitable AI demand.
Cisco describes the current market as a networking supercycle. The term refers to a long investment period driven by several overlapping upgrades rather than one product release.
AI clusters require faster back-end networks between processors. They also need front-end connections for storage, applications, and users. Large operators increasingly need high-capacity links between separate data centers.
Enterprise customers face another problem. Many existing networks were designed for predictable office applications, not AI agents producing frequent and variable traffic.
Cisco says that combination supports sustained investment across hyperscalers, enterprises, governments, and specialized cloud companies. Its broad portfolio gives the company several ways to participate.
The opposing case begins with customer economics. Building an AI data center requires spending on land, power, cooling, processors, networking, and construction before applications generate returns.
Cloud operators can support that investment while their existing businesses remain profitable. They will become more selective if AI services fail to produce sufficient revenue.
Networking suppliers feel the effect after customers revise construction schedules. A delayed facility can reduce near-term demand for switches and optical equipment, even when the customer still expects long-term AI growth.
This is why the distinction between technology adoption and investment returns matters. The Internet expanded after 2000, but many telecommunications networks had already installed more capacity than customers needed.
The resulting correction damaged equipment makers and carriers. It did not invalidate the Internet.
The same pattern can occur with artificial intelligence. Model usage can keep rising while infrastructure spending falls because operators built ahead of demand.
Cisco does have some protection. Networking hardware represents a smaller portion of an AI facility’s cost than processors and power systems. Operators may also redirect equipment toward other workloads.
Ethernet gives Cisco another advantage. It is a widely used networking standard with support from many suppliers, engineers, and software tools.
That familiarity can make Ethernet attractive to customers seeking alternatives to specialized AI fabrics. Cisco can sell complete systems, merchant silicon, or optical components, depending on a customer’s architecture.
Its Silicon One G300 is designed for 102.4 terabits per second of switching capacity. Cisco announced air-cooled and liquid-cooled systems based on that chip.
Those choices address practical data center limits. Higher network capacity can increase power and cooling requirements, while some operators cannot immediately adopt liquid cooling.
Yet technical relevance cannot eliminate spending cycles. A useful product can still encounter weak orders when customers postpone construction or consume equipment already purchased.
That is the core reversal behind the google news debate. Cisco’s exposure to AI makes it more relevant than it appeared several years ago. The same exposure also ties more of its growth to decisions made by a small group of enormous buyers.
A Bust Would Pressure Margins Before It Erased Demand
The first evidence of an AI correction would probably appear in product mix, inventory, and margins, not in an announcement that customers abandoned AI.
Cisco’s reported gross margin weakened as networking revenue accelerated. Its fiscal third-quarter GAAP gross margin was 63.6 percent, down from 65 percent in the previous quarter.
Product gross margin reached 61.9 percent, compared with 63.9 percent during the second quarter. Services gross margin increased to 69.2 percent.
Several factors can move these figures, including product mix, component costs, tariffs, and customer pricing. The decline does not independently prove that AI networking carries unattractive economics.
It does show why investors should not treat every dollar of infrastructure revenue equally. Large hyperscaler orders can produce substantial sales while carrying a different margin profile from software or services.
Cisco entered its fourth quarter expecting a non-GAAP gross margin between 65.5 percent and 66.5 percent. It expected a non-GAAP operating margin between 34 percent and 35 percent.
Future margin performance will reveal whether volume, component availability, and product design offset pricing pressure. A sustained decline would weaken the argument that AI orders create durable operating leverage.
Inventory provides another warning signal. Cisco reported $4.71 billion in inventory after its fiscal third quarter, up from $2.83 billion one year earlier.
Inventory purchase commitments reached $16.03 billion, compared with $6.35 billion in the previous-year quarter. Those commitments help Cisco secure components needed for expected customer demand.
They also increase exposure if delivery schedules change. Suppliers can end up holding components or completed systems longer than planned when customers defer deployments.
Cisco explicitly identifies excess or obsolete inventory as a business risk in its financial disclosures. Rapid technological change makes that risk especially important for high-performance networking.
New generations of chips and optical modules arrive while customers are still deploying current products. Delays can therefore affect both timing and product value.
A correction would not need to produce widespread cancellations. Slower new orders, longer delivery schedules, or smaller follow-on purchases could create the same pressure.
Customer concentration deserves attention as well. Cisco reports hyperscaler AI orders as a group, but a limited number of companies operate at that scale.
A few purchasing decisions can therefore change quarterly totals. Design wins also take time to convert into shipments, particularly when customers qualify new hardware across large networks.
Cisco’s diversification limits the damage. The company sells campus networking, security, collaboration, observability, and services alongside data center equipment.
Its acquisition of Splunk added a larger software and data analytics business. Subscription and service revenue can reduce dependence on any single hardware cycle.
However, diversification can also hide emerging weakness for several quarters. Stable software revenue might support total results while AI hardware growth slows.
Readers tracking a possible Cisco AI bust should therefore look beneath the consolidated revenue figure. Networking growth, product margins, inventory, and purchase commitments provide a clearer view of the pressure.
Cisco Has More Defenses Than It Had During the Dot-Com Crash
A modern Cisco correction would test a broader company, but history still warns that strategic importance does not prevent overcapacity.
The comparison with the dot-com era is useful because Cisco experienced both sides of that cycle. Internet traffic expanded, yet customers had purchased more equipment than the market could absorb.
Cisco entered 2001 carrying large inventories after years of rapid growth. When telecommunications customers cut spending, the company recorded a major inventory charge.
That episode established an enduring lesson. Correct predictions about long-term traffic do not guarantee correct short-term purchasing decisions.
Today’s Cisco has a different revenue mix. Software, subscriptions, security, observability, and services contribute more than they did during the original Internet buildout.
The company also sells into a broader range of environments. Its products connect hyperscale data centers, enterprise offices, factories, campuses, service providers, and public agencies.
A campus replacement program is not identical to an AI cluster build. Customers replace aging switches because support periods end, security requirements change, and wireless traffic grows.
Cisco’s fiscal third-quarter results showed strength across enterprise, public-sector, and service-provider markets. Product orders increased in every major geography.
That breadth matters because an AI slowdown would not automatically stop unrelated network modernization. Enterprises still need secure connectivity, operational visibility, and support for cloud applications.
AI can also shift from training toward inference. Training creates a model, while inference uses that model to answer requests or perform tasks.
Training often concentrates enormous computing clusters in a few locations. Inference can distribute demand across cloud regions, enterprise data centers, and edge environments.
Cisco argues that distributed inference will increase demand for networking and security. That path would reduce reliance on a limited number of giant training clusters.
The timing remains uncertain. Enterprises may adopt AI agents more slowly than infrastructure vendors expect, particularly when accuracy, governance, or integration problems remain unresolved.
Cisco’s own networking research presents AI readiness as a broad customer challenge. Company-sponsored research can identify real concerns, but it does not guarantee immediate purchasing.
Enterprise deployment decisions follow budgets, procurement reviews, and security testing. A promising use case can take several quarters to become a production system.
Competition adds another constraint. Arista Networks has a strong position in high-speed cloud networking. Nvidia offers networking alongside its processors, while hyperscalers design more technology internally.
Cisco must therefore win on performance, power consumption, reliability, and operating simplicity. It cannot depend only on the growth of the overall market.
Open Ethernet standards can expand Cisco’s addressable market, but they also make supplier comparisons easier. Customers may combine silicon, optics, systems, and software from different vendors.
Cisco’s broad product portfolio remains a defense when customers want one accountable supplier. It becomes less valuable when large buyers prefer disaggregated systems and direct component purchases.
The company is better positioned than a simple replay of 2001 would suggest. It is not immune to the capital cycle that made the comparison relevant.
What to Watch After the Next Cisco Google News Cycle
Three signals will determine whether Cisco is building a lasting AI networking business or approaching the peak of an infrastructure cycle.
The first signal is order-to-revenue conversion. Cisco expected roughly $9 billion in fiscal 2026 hyperscaler AI orders and approximately $4 billion in related revenue.
Future reporting should show how much of the remaining order value becomes revenue. Faster conversion would support management’s claim that deployments are moving from design commitments into installed networks.
Repeated delays would weaken that interpretation. They could indicate construction bottlenecks, customer scheduling changes, or orders placed far ahead of practical demand.
Readers should also compare new orders with recognized revenue. A healthy expansion normally requires follow-on orders to replenish the backlog as equipment ships.
The second signal is product gross margin. Revenue growth becomes less valuable when mix changes, discounts, or component costs consume the added profit.
A stable or improving product margin would show that Cisco can scale its AI systems without sacrificing too much profitability. Continued compression would make headline order growth less convincing.
This metric also helps separate an ordinary mix effect from a deeper problem. One weak quarter can reflect the timing of large deals, while several quarters establish a pattern.
The third signal is demand outside hyperscalers. Cisco reported roughly $900 million in fiscal year-to-date AI orders from neocloud, sovereign, and enterprise customers through its third quarter.
That group needs to develop into repeat purchases. It would give Cisco a wider customer base and reduce its dependence on a handful of cloud operators.
Enterprise demand deserves particular scrutiny. Many organizations express interest in AI, but production deployments remain harder than experiments.
Buyers need reliable data, governance, security, and workflows before AI systems deliver recurring value. Teams assessing internal evidence can use a searchable knowledge base to connect technical records with deployment decisions.
This operational layer matters to infrastructure vendors. AI traffic grows sustainably only when organizations continue using applications after pilot programs end.
The bearish case strengthens if hyperscaler orders slow before enterprise demand expands. Cisco would then face weaker growth while carrying inventory and component commitments built for a larger market.
The bullish case strengthens if revenue conversion improves, product margins stabilize, and non-hyperscaler customers place larger follow-on orders. That combination would make the supercycle broader and more durable.
Investors should resist treating every google news headline as confirmation of either outcome. Cisco has already demonstrated genuine demand, substantial design activity, and rapid order growth.
It has not demonstrated that today’s extraordinary growth rate can persist through a full capital-spending cycle. No supplier can prove that before the cycle turns.
The practical question is narrower: Are Cisco’s customers installing equipment fast enough, earning enough from AI, and returning with new orders?
Watch those three signals across the next several reports. If conversion, margins, and customer breadth move together, Cisco’s AI business has a durable foundation. If they diverge, the bust warning will look less like provocation and more like preparation.
