The Hyperscaler Bet That Answered the Bears: Why Cisco's Enterprise AI Pipeline Matters More Than the $9 Billion Number

John Ian Lau — Technology & AI infrastructure desk

For most of the last eighteen months, Cisco Systems was the one large-cap AI infrastructure name the market was actively willing to short. The narrative was simple and, on the surface, persuasive: Arista Networks had taken the hyperscaler switching crown, Nvidia owned the rack, and the incumbent that built the internet's plumbing in the 1990s had nothing differentiated to sell into the most important capital cycle of the decade. That trade ended on May 13, when Cisco reported fiscal third-quarter results that did not so much beat expectations as obliterate the bear case wholesale. Revenue hit a record $15.8 billion, up 12% year-over-year. Non-GAAP earnings per share came in at $1.06. And buried in the supplementary disclosures was the number that mattered: $1.9 billion in AI infrastructure orders from hyperscalers in a single quarter, against $600 million a year earlier.

The stock did what stocks do when an 18-month-old short thesis collapses in a single press release. Shares jumped 13% to 19% intraday on May 14, taking Cisco's year-to-date return past 50% and lifting the share price into a $112 to $121 range that, as recently as January, almost no sell-side model contemplated. But the more interesting question for investors is not why the market repriced Cisco on a single quarter. It is whether the bull case the company laid out on the call is durable enough to justify the new multiple — and what, exactly, is now being priced.

The headline upgrade was on the order book. Cisco entered fiscal 2026 telling investors it expected to take roughly $2 billion in AI infrastructure orders from hyperscalers across the full year. Three quarters in, the company has already booked $5.3 billion. Management used the call to lift the full-year guide to approximately $9 billion — four and a half times last year's total — with a full quarter remaining to push the number higher.

“The year-to-date total of $5.3 billion in orders taken from hyperscalers already exceeds our prior expectations of $5 billion for FY '26 with a full quarter remaining. Given the strong demand, we now expect to take AI infrastructure orders of approximately $9 billion from hyperscalers in FY '26,” Chuck Robbins, Cisco's chair and chief executive, told analysts on the May 13 earnings call.

Robbins, who has run Cisco since 2015 and spent most of the last two years defending the company's relevance to the AI build-out, framed the quarter in characteristically understated terms. “Q3 proved to be an outstanding quarter for Cisco, with our momentum accelerating,” he said on the call. Inside the numbers, the acceleration was broader than the hyperscaler headline suggested. Total product orders rose 35% year-over-year. Networking orders, excluding hyperscalers, grew 19%. Within enterprise, data center switching orders were up 40% and campus orders rose 25%. Networking revenue of $8.82 billion came in nearly $350 million ahead of consensus.

The variable that actually changes the duration of the trade

The market reaction has fixated on the $9 billion number. The more important disclosure, in our view, is the one Cisco volunteered about the next tier of buyers. Enterprise, sovereign, and neocloud customers — the wave of AI infrastructure spending that follows the hyperscalers — booked roughly $900 million of AI orders year-to-date, with a stated pipeline above $3 billion. That cohort grew at triple-digit rates. It is small in absolute terms today. It is the thing that determines whether Cisco's AI franchise is a two-year hyperscaler trade or a multi-year platform.

For 18 months, the bear argument has rested on hyperscaler concentration risk: four to six customers, all running their own custom silicon roadmaps, any one of which could in-source the entire stack. The arithmetic of the May 13 print does not eliminate that risk. But the trajectory of the enterprise pipeline complicates it materially. If sovereign clouds, regional neoclouds, and Global 2000 enterprises are placing nine-figure orders for the same Silicon One-based fabric Cisco is shipping to the hyperscalers, the concentration story becomes a transition story. That is a different multiple.

The Street caught up to the print quickly. By the morning of May 14, at least six brokerage firms had lifted price targets, with the new range clustering between $112 and $132. The unanimity is the signal. Single firms move targets on every print; six firms moving in lockstep, in the same week, on the same set of disclosures, indicates that fundamental models — not sentiment — have been revised.

“The multiples compared to other AI capital expenditure names remain reasonable, growth is at its highest in 15 years, and estimates likely have room for upward adjustments, maintaining our overweight stance,” Meta Marshall, Morgan Stanley's networking equipment analyst, wrote in a note to clients on May 14, raising her price target on Cisco to $120 from $91.

Marshall's framing is the one investors should focus on. The growth-at-its-highest-in-15-years observation matters because it is verifiable from the company's own filings — and it lands at a moment when the rest of the AI infrastructure complex is trading at multiples that increasingly require heroic terminal-value assumptions. Cisco still trades at a meaningful discount to Arista, Broadcom and the merchant-silicon names on virtually every forward metric. The bull case is not that the discount disappears tomorrow. It is that the discount was sized for a company that was losing the AI networking war, and the war is no longer being lost.

The cost-side discipline the market did not price

Less remarked on, but equally relevant for the durability of the new earnings power, was Cisco's announcement of a workforce reduction of roughly 4,000 positions and approximately $1 billion in pre-tax restructuring charges spread across fiscal 2026 and 2027. The cuts are concentrated in legacy software and general-and-administrative functions — exactly the cost pools that have weighed on Cisco's incremental margins for a decade. Non-GAAP gross margin for the quarter came in at 66.0%, with GAAP gross margin at 63.6%. Those are software-company margins on a hardware-led mix, and the restructuring is designed to keep them there as AI volumes scale.

This is the kind of discipline that does not show up in a single quarterly print but compounds. If Cisco delivers FY26 AI revenue of approximately $4 billion at company-average gross margins, the contribution to operating earnings is material — and the restructuring ensures it falls through to the bottom line rather than being absorbed by legacy cost structures.

What we are watching

Our view is that the May 13 print resolves the most-debated question in AI networking but opens a more interesting one. The hyperscaler order book is no longer the variable to track; at $9 billion run-rate with rising guidance, it is now a known quantity. The variable is the enterprise and sovereign pipeline. If that $900 million-to-$3 billion cohort converts at the rate Cisco's commentary implies, the company is in the early innings of a multi-year platform shift, not a cyclical AI bounce. If it stalls — if sovereign procurement timelines slip, or if enterprise buyers wait for the next silicon generation — the trade compresses back to a hyperscaler concentration story and the multiple expansion is borrowed, not earned.

For now, the evidence runs in the company's favor. Six broker upgrades on a single set of disclosures is not the market chasing a print. It is the market acknowledging that the model was wrong. The interesting trade from here is not Cisco at $120. It is whether the enterprise AI thesis the company has just laid out survives contact with the next two quarters. We are inclined to give the benefit of the doubt — but we would size the position to the question we have not yet answered, not the one Q3 just answered for us.

This note reflects the views of Solomon Grey Capital's Technology & AI infrastructure desk as of publication. It is for informational purposes only and does not constitute investment advice or a recommendation to buy or sell any security. Past performance is not indicative of future results.