Cisco Closes Fiscal 2026 with $9.3 Billion in AI Orders from Hyperscalers, but Margin Concerns Weigh on Market
Revenue of $17.25 billion in the fourth quarter surpassed consensus, and AI orders grew 4.5x year-over-year, but guidance for gross margin below expectations caused a 6% drop in after-hours trading.
Cisco closed fiscal year 2026 with revenue of $17.25 billion in the fourth quarter, an 18% year-over-year increase, and $9.3 billion accumulated in AI infrastructure orders from hyperscalers. In just the fourth quarter, there were $4 billion in new orders from this segment, 4.5 times above the previous fiscal year. Adjusted EPS came in at $1.22, compared to an estimate of $1.17. GAAP net income surged 51% to $3.9 billion. Even so, the stock fell 6% in after-hours trading.
Chuck Robbins, Cisco’s CEO, summarized in the release: "Our record performance is a testament to the rapid pace of innovation and the excellent execution of our teams. With the breadth and depth of our portfolio and our competitive differentiation in secure networking, Cisco is well-positioned to support customers wherever they choose to deploy AI." What has changed is the composition of the backlog. About 60% of orders are based on Silicon One, Cisco's proprietary silicon for high radix switching, and 40% on optics. This indicates that hyperscalers are no longer buying at the network edge but rather for internal traffic management within the data center.
What the Stock Drop is Indicating
The guided gross margin for 65% to 66% in fiscal 2027 was below consensus, which is why the market is unsettled. The heavier hardware mix, particularly in 800G and 1.6T optics, is pushing margins down even as revenue grows. This is the same dynamic that Arista Networks and Broadcom have been reporting: the volume of orders is there and is massive, but each dollar of revenue now carries more commodity costs than the historical portfolio.
Cisco projected AI infrastructure revenue for hyperscalers of $7.5 billion in fiscal 2027, nearly double the $4 billion from the concluded year. According to Robbins, four of the major hyperscalers reported triple-digit growth in AI orders. The company did not name the customers, but the market assumes Microsoft, Google, Meta, and Amazon as the core, with Oracle and xAI driving secondary demand.
Readings by Region
In the United States, Cisco's figures reinforce JPMorgan's projection released this week of $1.2 trillion in AI capex by the end of 2027. When Silicon One becomes the default at Meta and xAI, the high-performance switching market moves out of Broadcom's monopolized orbit and diversifies. This is the first quarterly season where one can speak of a duopoly, not a monopoly, in this segment.
In Asia, Taiwan and South Korea are included in the calculation through manufacturing. Silicon One utilizes TSMC's capacity at advanced nodes, and competition for 3 nm wafers with Nvidia, AMD, and Broadcom itself tightens the delivery schedule. Samsung Foundry, which has fallen behind on 3 nm, is gaining traction in the second-order supply chain: optical transceivers and substrates.
In Europe, the race among Deutsche Telekom, Orange, and Telefónica for sovereign AI cloud offerings is facing immediate supply constraints. Cisco already has extended delivery times, with hyperscaler customers prioritized. A European provider that signed a contract for AI infrastructure in 2026 may see delivery slip to the second half of 2027. This reinforces the political thesis that Brussels needs to capitalize on local champions to avoid becoming reliant on American supply chains.
In Brazil, the effect is indirect but significant. Neo-hyperscale data center providers such as Ascenty, Elea, and Odata have structured expansion plans relying on Cisco networking equipment. If delivery times double, commissioning schedules for new phases will be delayed. Banks like Itaú and BB Seguridade, which depend on this capacity to run AI-based credit workloads and fraud detection, will need to renegotiate SLAs with these operators.
What remains: Cisco's fourth quarter confirmed the boom, but it also showed that it is not without costs, even for the infrastructure vendor. If margins continue to fall while orders grow, the winner of this phase of the cycle will be the one with the largest scale of proprietary silicon, not necessarily the strongest brand in enterprise networking. Robbins has six quarters to prove that Cisco can be that winner.