Cisco Q4 Earnings: Inside the $9.5B Growth Spike and the Margin Trade-Off
Cisco reported Q4 FY2026 revenue of $17.3 billion, up 18% year over year, beating the roughly $16.8 billion consensus. Non-GAAP earnings of $1.22 per share topped the $1.17 Street estimate. Hyperscaler AI-infrastructure orders hit $4 billion in Q4, bringing the full-year total to $9.3 billion—above the company's recently raised $9 billion target. Product orders rose 35%. Networking orders rose 40%.
The stock surged past $132 after hours, then reversed. Later snapshots showed it around $118–$121, down 2–4.5% from its $123.88 regular-session close.
Two Hardware Cycles Running in Parallel
Cisco's FY2027 guidance tells a story the headline numbers obscure. The revenue midpoint of $72.8 billion implies roughly $9.5 billion in incremental growth over FY2026's $63.3 billion. AI-infrastructure revenue is expected to reach $7.5 billion, up from $4 billion—an 88% increase.
That $3.5 billion increment accounts for roughly 37% of the total growth.
The remaining 63% must come from campus switching, security, and enterprise data-center refreshes. Product orders grew 25% after stripping out hyperscalers, with double-digit growth across every geography. Management is calling this a "networking supercycle," and the order data support the label.
GPU-cluster buildouts are pulling 800G and 1.6T Ethernet fabric demand through Silicon One and Nexus platforms. An aging enterprise installed base, combined with Wi-Fi 7 adoption and security requirements, is accelerating campus upgrades independently. The dual-engine picture matters because concentrated hyperscaler revenue is cyclical and low-margin; enterprise demand is stickier and carries better economics.
The Margin Trade-Off
Non-GAAP product gross margin fell 270 basis points year over year to 64.8%. Total non-GAAP gross margin dropped to 66.3% from 68.4%. Q1 FY2027 guidance of 65–66% signals continued compression.
Hyperscalers negotiate aggressively. High-speed AI switching carries heavier silicon, memory, and optics costs. Operating leverage has compensated so far: Q4 non-GAAP operating margin reached 35.9%, and operating income grew 23%. The question investors are pricing is whether that offset holds.
Inventory compounds the concern. It reached $5.69 billion, up roughly 80% from $3.16 billion, against 12% annual revenue growth. If orders keep converting, that is pre-positioning. If hyperscaler capex pauses, it becomes write-down risk.
Real Share Gains, With Limits
Dell'Oro's Q1 2026 data show Cisco recorded the largest vendor share gain in AI back-end Ethernet switching. Ethernet accounted for approximately two-thirds of AI-cluster switch sales, with back-end switch revenue more than doubling. Silicon One's G300—102.4 Tbps, supporting 1.6T networking—is deployed by five of the six largest hyperscalers.
Cisco still ranks fourth behind Celestica, Nvidia, and Arista. Arista reported Q2 revenue growth of 37.7% with a 49.9% non-GAAP operating margin. Cisco is credibly contesting a market it barely participated in two years ago—a different claim than winning it.
Where the Real Money Is Migrating
The number that should command the most attention sits in the cross-company comparison. Cisco's product gross margin compressed to 64.8% on 24% revenue growth. The component vendors supplying AI-cluster interconnects reported different economics entirely: Credo produced 68.3% non-GAAP gross margin on 157% revenue growth; Astera Labs delivered 73.7% on revenue that doubled; Lumentum reached 50.4% as revenue surged 109%.
Economic rent is migrating toward the scarce, IP-intensive layers of the networking stack—SerDes, signal conditioning, optical engines, retimers—and away from system integration, where Cisco has historically earned its premium. Cisco's Silicon One strategy is an explicit attempt to recapture that rent by owning the silicon underneath its switches and selling it as merchant silicon. The G300 competes directly with Broadcom's Tomahawk 6 at the same 102.4 Tbps; this is an execution race.
The structural logic: doubling GPU count in a cluster does not double useful compute unless those GPUs communicate without congestion. Networking content per AI cluster is rising, and the vendors with the highest technical barriers are extracting the widest margins.
Cisco can participate in that value migration through silicon, optics, and software—or watch the economics pass through its boxes to upstream suppliers. That distinction determines whether AI revenue is accretive growth or dilutive volume, and the answer over the next four quarters will define the stock's trajectory far more precisely than any single order print.
not investment advice