Cisco closed its fiscal year with the kind of quarter that headlines write themselves for. Revenue reached $17.3 billion in the fourth quarter, up 18% from a year earlier, and full-year revenue came in at $63.3 billion, up 12% [1]. GAAP earnings per share of $0.97 rose 52% year over year; on a non-GAAP basis, earnings of $1.22 per share rose 23% [1]. Networking product orders grew 40% year over year in the quarter [1]. By almost any measure, the company is selling more of what data centers need.
One number in the same release points the other way. Cisco booked $9.3 billion in AI-infrastructure orders across fiscal 2026, with $4 billion of that landing in the fourth quarter alone [1]. For fiscal 2027, the company's own outlook calls for $7.5 billion in AI-infrastructure orders [1]. That is a guide lower of $1.8 billion, a decline of roughly 19%, on the precise line item that the market treats as a one-way escalator.
The gap matters because of what surrounds it. Cisco is not guiding its whole business down. Full-year fiscal 2027 revenue is guided to a range of $72.2 billion to $73.4 billion, which would be growth of roughly 14% to 16% over the $63.3 billion the company just reported [1]. The total keeps climbing. The AI-orders line, carved out and named by Cisco itself, is the one that steps back.
Set against the rest of the sector, this is not a story of collapse. Super Micro reported fiscal-year revenue of $39.1 billion, up 78% [2]. CoreWeave posted second-quarter revenue of $2.575 billion, up 112% year over year, and reported a backlog of roughly $104 billion [3]. The market for AI infrastructure is still expanding at a pace that most industries never see. A reader who wants evidence that AI hardware demand is real will find plenty of it in those figures.
The point is narrower, and Cisco supplied it. The dominant framing around AI capital spending treats acceleration as a given, a curve that only bends upward, quarter after quarter, vendor after vendor. That framing gets stress-tested when a company of Cisco's scale, reporting record numbers everywhere else, puts a smaller AI-orders figure in its own forward guidance. The company did not editorialize about it. It simply printed $9.3 billion for the year behind it and $7.5 billion for the year ahead [1].
Who carries the assumption when it runs ahead of the data? The people betting on the AI-infrastructure trade benefit from a story of uninterrupted growth, and that story holds together whether or not any single vendor's guidance backs it up. The cost lands elsewhere. Utilities are planning generation and grid capacity years out against forecasts of AI demand that the industry describes as bottomless, and the multi-billion-dollar price of that build-out flows into the rate base that ordinary customers pay into. When the projected demand is modeled as a straight line up, the infrastructure gets sized for a straight line up, and someone pays for the capacity whether the orders arrive or not.
None of this makes Cisco's quarter weak. It was strong, and the numbers say so plainly. What the release does is offer a single data point that the AI-capex narrative has to account for rather than wave past: a major vendor, reporting record revenue, told investors to expect fewer AI-infrastructure orders next year than it booked this year. That is not a forecast of the end of anything. It is a reminder that the escalator has a number on it, and this year the number went down.
