Record AI Orders Bought With Gross Margin: Cisco's Beat-and-Raise Loses 12% on the Memory Bill

Results vs. Consensus

Year-over-Year Comparison

Sequential Comparison

Revenue

Total revenue of $15,349M grew 9.7% and beat the $15,120M consensus by $229M. The composition is unambiguous: product revenue grew 14% to $11,642M while services fell 1% to $3,707M, and within product, Networking alone contributed $8,294M at +21%. Security fell 4%, Observability was flat, and Collaboration, the smallest of the four product groups, added 6%. Geographically the strength was broad, with EMEA up 15%, Americas up 8%, and APJC up 8%.

The order book is the better forward read, and it is stronger than the revenue line. Total product orders grew 18% on top of double-digit growth in the year-ago quarter, and they grew double digits in all three geographies (Americas +23%, EMEA +11%, APJC +15%) and across all three customer markets (service provider and cloud +65%, public sector +11%, enterprise +8%). Excluding hyperscalers, product orders still grew 10%, an acceleration from the 9% ex-hyperscaler figure disclosed at Q1. Networking product orders grew more than 20%, the sixth consecutive quarter of double-digit growth.

Assessment: This is the cleanest demand quarter in the four we have covered. The concern is not whether the revenue is real but whether the mix behind it is the mix shareholders were underwriting. Hardware is now doing all of the work while the recurring franchise decelerates, and that composition is precisely what shows up two lines lower in the income statement.

Margins

Non-GAAP gross margin of 67.5% fell 120 basis points year over year and 60 basis points sequentially, and missed the 68.14% the Street had modeled. All of the compression sits in product: non-GAAP product gross margin of 66.4% fell 130 basis points on what the company describes as negative impacts from mix and higher memory costs, partially offset by productivity improvements. Non-GAAP services gross margin of 70.9% fell 70 basis points, which is a smaller but genuinely new negative given services gross margin expanded 40 basis points in Q1.

Below the gross line, the story flips. Non-GAAP operating expenses of $5,046M grew 6% against 10% revenue growth, taking OpEx to 32.9% of revenue from 34.0%, which is 110 basis points of genuine cost leverage. It was not quite enough. Non-GAAP operating margin of 34.6% is 10 basis points below the year-ago quarter's 34.7%, even though it beat the high end of Cisco's own guide and was, as management noted, the highest in four quarters. Non-GAAP operating income grew 9% while revenue grew 10%.

Assessment: The company's "operating leverage" framing is half right and worth stating precisely. OpEx discipline was real and substantial. It offset 110 of the 120 basis points of gross-margin decline, leaving operating margin down slightly. That is expense control absorbing a cost shock, not leverage compounding into the operating line.

EPS

Non-GAAP EPS of $1.04 grew 11% and beat consensus by two cents. Management framed this as "growing EPS faster than revenue," which is true at the reported rounding but thin underneath. Non-GAAP net income of $4,143M grew 10.2% against revenue growth of 9.7%, a 50-basis-point gap. That gap came from below the operating line and from the share count, not from margin: non-GAAP interest and other expense improved to $196M from $219M, the non-GAAP tax rate was unchanged at 19.0%, and diluted shares fell 0.5% to 3,984M.

GAAP EPS of $0.80 grew 31%, a far larger increase than the non-GAAP figure, driven by two items that flatter the comparison rather than the business: amortization of acquisition-related intangibles fell to $459M from $600M as the Splunk intangibles age, and acquisition and divestiture-related costs fell to $102M from $222M. The GAAP tax rate of 12.9% also compares against 15.9% a year ago.

Assessment: EPS quality is acceptable but not pristine. There is no aggressive add-back and no tax-rate manoeuvre, and the buyback contribution is modest. But the claim that EPS grows faster than revenue is now carried by half a point of below-the-line items, and Cisco's own Q3 guide breaks it outright: at the midpoints, Q3 non-GAAP EPS grows roughly 7.3% against roughly 9.5% revenue growth.

Segment Performance

Networking

Networking is not merely carrying the quarter, it is now larger than the rest of Cisco combined at 54% of revenue and it grew 21%. Within it, management cited double-digit growth in campus switching, data center switching, wireless, service provider routing, enterprise routing and compute. Data center switching has now grown double digits in six of the last eight quarters. The order data is stronger still: Networking product orders grew more than 20%, and Wi-Fi 7 orders grew 80% sequentially.

"AI infrastructure orders taken from hyperscalers totaled $2.1 billion in Q2 compared to $1.3 billion just last quarter and equal to the total orders taken in all of fiscal year '25, marking another significant acceleration in growth across our silicon, systems and optics." — Chuck Robbins, Chair and CEO

The silicon and optics roadmap moved forward on three fronts during the quarter. Cisco shipped its one-millionth Silicon One chip and reiterated the plan to deploy the architecture across its high-performance networking systems by fiscal 2029. At Cisco Live Amsterdam it introduced the 102.4 terabit-per-second G300 chip and four systems built on it (Cisco 8000 and Nexus 9000, in air-cooled and liquid-cooled variants), plus two new pluggable optics, a 1.6 terabit OSFP and an 800-gig LPO. Acacia posted its strongest quarter to date with triple-digit bookings growth, and all major hyperscalers are now deploying its coherent pluggable optics.

Two structural points deserve emphasis. First, none of the newly announced silicon is in the FY26 AI order target, which management stated explicitly. Second, Cisco won three new use cases at major hyperscaler accounts in the quarter, one in optics and two on the systems side, so the order growth is not purely deeper penetration of existing sockets.

Assessment: The Networking franchise is executing at a level that would justify a higher multiple in isolation. The problem is that its incremental dollar carries a lower gross margin than the Cisco average, and the segment's growth is therefore mathematically dilutive to the consolidated margin rate even before memory costs are considered. Underwriting Networking now means underwriting gross profit dollars, and those grew 7.7% this quarter against 21% segment revenue growth at the top.

Security

Security fell 4%, a deterioration from the 2% decline in Q1, and it is now three consecutive quarters of negative reported growth for a segment management has repeatedly framed as a mid-teens grower over the long term. The stated causes are unchanged: declines in prior-generation products and the continuing shift of Splunk deals from on-premises to cloud subscriptions, partially offset by growth in new and refreshed products.

The underlying product data is genuinely better than the revenue line. Roughly one third of the security portfolio consists of new or refreshed products (Secure Access, XDR, Hypershield, AI Defense and refreshed firewalls). Excluding the refreshed firewalls, more than 1,000 new customers bought one of these products in Q2, up more than 100% sequentially, bringing the cumulative total since launch to roughly 4,000. Firewall unit orders grew double digits for a third consecutive quarter. Secure Access booked over 2.5 million users, with more than half of added customers being new logos.

"And where that would lead us is as we exit Q4 this year, the organic Cisco security portfolio will be growing revenue close to double digits as we exit. So the teams are a little -- they're a little behind where we -- not the teams, but this portfolio is a little behind where we thought we'd be exiting the year, but it's still performing relatively well." — Chuck Robbins, Chair and CEO

That sentence is the most important thing said about Security on this call, and it is a walk-down on two axes. The exit-rate target moved from the mid-teens framing used in prior quarters to "close to double digits," and the measure narrowed from Security to the "organic Cisco security portfolio," which excludes Splunk. Management also declined to give a security order growth percentage this quarter, having disclosed +4% at Q1.

Assessment: Security is now a two-year rebuild being described one quarter at a time. The new-product traction is real and the customer counts are the right leading indicator, but the segment has missed its own framing three quarters running and the goalposts moved again this quarter. We are treating the mid-teens target as retired until management restates it with a date attached.

Collaboration

Collaboration grew 6% to $1,054M, its first meaningful growth in the period we have covered and a reversal from the 3% decline in Q1. Management attributed it to double-digit growth in devices plus growth in CPaaS, Webex and Cloud Contact Center. At 7% of revenue it remains the smallest product group and it received one sentence on the call.

Assessment: A genuine positive surprise that changes nothing about the investment case. Devices growing double digits after several quarters of decline is more likely a refresh cycle in hybrid-work hardware than a competitive inflection against Zoom or Teams. Do not extrapolate it, but do stop modelling Collaboration as a structural drag.

Observability

Observability was flat at $277M, decelerating from 6% growth in Q1. It was not mentioned once on the call, in prepared remarks or in Q&A. Given the segment is the vehicle for ThousandEyes and the analytics half of the Splunk logic, a stall here is not trivially dismissible.

Assessment: Small enough at 2% of revenue not to move the model, large enough as a signal to note. A flat quarter with zero commentary in a call that ran long on AI is the kind of omission that becomes a question two quarters later.

Services

Services revenue fell 1% to $3,707M and services gross margin fell 70 basis points to 70.9% on a non-GAAP basis. Services RPO grew 2%. This is the quiet negative in the print: services is the highest-margin quarter of Cisco's revenue and it is shrinking while the lowest-margin part grows 14%.

Assessment: Mix, restated. A services line that declines while product grows 14% mechanically drags consolidated gross margin regardless of what memory does. Two of the four quarters we have covered now show services flat to down, which makes this a trend rather than a quarter.

Key KPIs

Key Topics & Management Commentary

Overall Management Tone: Management was confident on demand and conspicuously unspecific on cost. The prepared remarks led with a pre-emptive memory-price section before any analyst raised it, which is the posture of a team that knew where the questions were going, and every subsequent margin answer stayed at the level of strategy (price increases, contract terms, supply-chain scale) rather than quantification. Demand answers, by contrast, came with numbers attached at every turn. The gap between those two registers is the single most informative feature of the call, and it is a change from Q1, when the DRAM commentary was framed as an already-absorbed line item in the guide.

1. The Memory Shock and What the Q3 Guide Actually Says

The Q3 non-GAAP gross margin guide of 65.5% to 66.5% is the number that mattered. It sits 210 to 310 basis points below the 68.6% Cisco printed in Q3 FY25, and 100 to 200 basis points below the 67.5% it just reported. For a business whose gross margin has traded in a roughly 67% to 69% band for years, a guide with a 65 handle at the low end is a regime statement, not a quarter.

"So really, as you look at the Q3 guide, there's 2 primary things at work. One is mix and the other is memory prices, as you mentioned." — Mark Patterson, CFO

Robbins raised the topic unprompted in his opening remarks, before the financials, describing "the recent significant increases in memory prices across the market" and laying out three responses: price increases already announced, revised contractual terms with channel partners and customers, and using Cisco's operating scale to negotiate supply. He closed the section with a competitive claim rather than a number.

"Overall, we feel confident in our ability to manage this industry-wide dynamic better than our peers." — Chuck Robbins, Chair and CEO

Assessment: At no point in the call did management split the Q3 guide between mix and memory, quantify the modeled memory inflation, or put a duration on it. That is the whole problem. A cost shock you can size is a discount-rate question. A cost shock you cannot size is a multiple question, and the market answered it as a multiple question.

2. AI Infrastructure: A Full Year of Orders in a Single Quarter

Hyperscaler AI infrastructure orders of $2.1 billion in Q2 compare to $1.3 billion in Q1, more than $800 million in Q4 FY25 and more than $600 million in Q3 FY25. The Q2 figure alone equals the entire FY25 total. Management raised the FY26 AI order expectation to "in excess of $5 billion," against the commitment made at Q1 to take at least twice the FY25 order figure, and raised FY26 AI infrastructure revenue from hyperscalers to "over $3 billion" from roughly $3 billion.

"Given the strong demand for our Silicon One systems and optics, we now expect to take AI orders in excess of $5 billion and to recognize over $3 billion in AI infrastructure revenue from hyperscalers in FY '26." — Chuck Robbins, Chair and CEO

The mix within those orders was disclosed as roughly 60% systems and 40% optics, described as reasonably consistent over the last few quarters. Three new use-case wins were added in the quarter, one optics and two systems.

Assessment: This is the strongest single data point in the print and it validates the bull pillar we have carried since the Q3 FY25 upgrade. The order-to-revenue gap is now the interesting question: more than $5 billion of FY26 orders against just over $3 billion of FY26 revenue means roughly $2 billion carries into FY27 before a single new order is taken. That is the FY27 revenue foundation the market did not price today.

3. What Is Deliberately Not in the $5 Billion

Robbins was unusually precise about the perimeter of the AI order target, and the exclusions are material. Neither the P200-based products announced in Q1 nor the newly announced G300 silicon and the two new optics modules are inside the number.

"But just to reiterate, P200 and G300 are not in the $5 billion expected for fiscal '26." — Chuck Robbins, Chair and CEO

On the 51.2 terabit portfolio already shipping, the constraint is stated as supply rather than demand: Cisco is "selling as much as we can build at this point," with demand from a couple of major customers "literally asking for as much as they can get."

Assessment: Two separate upside options sit outside the guide: newly launched silicon that has not been ordered yet, and existing silicon that cannot be built fast enough. Both are real. Both also come with the same margin question, since the constraint that limits how much Cisco can build is the same component market that is compressing the margin on what it does build.

4. Campus Refresh: The Second Leg Is Ramping

The campus story that we identified at Q1 as the underwritten second leg accelerated. All four campus platform families (enterprise switching, enterprise routing, wireless and industrial IoT) are ramping faster than prior transitions. Wi-Fi 7 orders grew 80% sequentially and campus switching orders grew close to double digits. Industrial IoT has now grown double digits for seven consecutive quarters, which management attributes to onshoring of US manufacturing, AI workloads moving to the network edge, and physical AI.

"That being said, we're in the top of the first inning. So this thing is just getting started. So there's a lot of runway." — Chuck Robbins, Chair and CEO

The structural argument is unchanged from Q1 and it is the right one: an installed base "representing tens of billions of dollars across early catalyst generations nearing end of support," combined with security woven into the network fabric and the operational modernization Agentic AI requires.

Assessment: The campus leg is doing exactly what we underwrote at Q1, and it matters more now than it did then. Campus carries a materially higher gross margin than hyperscaler AI systems and optics, so the faster campus ramps relative to AI, the less severe the mix component of the margin compression becomes. Campus is now the margin hedge as well as the growth hedge.

5. The Recurring Revenue Franchise Is Going the Wrong Way

This is the disclosure nobody discussed and it belongs near the top of the analysis. Total ARR ended the quarter at $31.0 billion, up 3%, having ended Q1 at $31.4 billion, up 5%. Product ARR growth slowed to 6% from 7%. Subscription revenue of $7.8 billion represented 51% of total revenue, down from $8.0 billion and 54% one quarter earlier. Total software revenue grew 2%, decelerating from 3%. Total RPO growth slowed to 5% from 7%, and from 16% in the year-ago quarter.

"Total ARR ended the quarter at $31 billion, an increase of 3% with product ARR growth of 6%." — Mark Patterson, CFO

That sentence is the only time ARR was raised on the entire call. It was not returned to in prepared remarks and no analyst asked about it. Part of the decline is arithmetic: subscription share falls when hardware revenue grows 14% and the subscription base grows in the low single digits. But the dollar decline in ARR, from $31.4 billion to $31.0 billion, is not a mix artifact.

Assessment: The bull case for owning Cisco through an AI hardware cycle has always rested on a growing software and subscription base to carry the margin structure through the hardware peak. This quarter that base shrank in dollars and decelerated on every growth measure. Patterson's own remedy for FY27 gross margin is that "software growth will improve," which is a forecast pointing in the opposite direction to the data he just reported. We are elevating this from a footnote to a formal bear point.

6. Splunk: A Retired Metric

The Splunk cloud mix shift continues, and management now expects the revenue drag to persist through the second half of FY26. The framing remains that the shift is strategically positive because cloud subscriptions are stickier and enable faster feature adoption. What changed is the evidence offered.

At Q1, management pointed explicitly to Splunk ARR and product RPO both growing double digits as the honest measure of the underlying business, and specifically nominated those metrics as the right ones to watch. At Q2, neither was mentioned. The substitute disclosure is a new-logo count: 500 new Splunk logos in the first half, on track for 1,000 for the year.

Assessment: A company that nominates a metric one quarter and stops reporting it the next has told you something. New-logo counts are a fine leading indicator and a poor substitute for ARR. Combined with consolidated ARR growth decelerating to 3%, the most parsimonious reading is that Splunk ARR growth also slowed. We will treat that as the working assumption until management restates the figure.

7. Pricing Power Into a Component Shock

Cisco is doing the thing that distinguishes a franchise from a commodity assembler: raising prices into a cost shock and renegotiating terms, and reporting that customers are absorbing it. Patterson said Cisco saw no evidence of pull-forward buying in the quarter, having checked linearity, pipeline, software activations and channel data.

"And while they may not like it, they understand that it's a dynamic that we're all dealing with." — Chuck Robbins, Chair and CEO

Robbins added a competitive point that matters for how much of the memory bill Cisco actually eats: core networking platforms carry less memory content than compute platforms, so Cisco's price increases are "more nominal" than those its server-exposed competitors are discussing, and UCS is a smaller share of Cisco's mix than servers are for its peers.

Assessment: This is the strongest argument against reading the Q3 guide as permanent. Cisco's memory exposure is genuinely lower per revenue dollar than Dell's or HPE's, and it has pricing authority those two do not. The catch is timing. Price increases and contract amendments recover margin with a lag of quarters, while component cost lands immediately, and the Q3 guide is what that mismatch looks like.

8. Advance Purchase Commitments: $1.8 Billion of Capital at Risk

Advance purchase commitments rose $1.8 billion in the last 90 days and are up roughly 73% year over year, with a large portion in memory. Inventories rose to $3,920M from $3,164M at the fiscal 2025 year end, an increase of $756M.

"If you look at it, it's evidenced by our advanced purchase commitments that just in the last 90 days are up $1.8 billion. If you look at it on a year-over-year basis, they're up about 73%. A big chunk of that is around memory." — Mark Patterson, CFO

Assessment: Read this two ways at once. It is the most concrete evidence that Cisco believes its own order book, since a company does not commit that much capital to secure supply for demand it doubts. It is also a locked-in cost basis: if memory prices roll over in late FY27, Cisco will be carrying committed inventory bought at the top. Both readings are correct, and the first one dominates over our 12-month horizon.

9. Sovereign and Neocloud: Now Formally an FY27 Story

Non-hyperscaler AI orders (neocloud, sovereign and enterprise) totaled $350 million in Q2, up from $200 million in Q1, with the pipeline now above $2.5 billion, from above $2.0 billion. Cisco announced plans to form a joint venture with AMD and HUMAIN to deliver up to one gigawatt of AI infrastructure by 2030, with operations expected to begin this calendar year and a 100-megawatt Phase 1 in Saudi Arabia. Management also flagged European interest in air-gapped sovereign infrastructure.

"the first thing I would say is that the sovereign side, there's really no real need nor expectation for meaningful impact in FY '26. And so we don't need that to actually accelerate for the guide that we've provided. It's purely upside. And then the neoclouds, I'd say, are generally the same. We expect the ramp to begin in the second half, but really be FY '27." — Chuck Robbins, Chair and CEO

Assessment: The pipeline grew and a named gigawatt-scale joint venture now exists, but the materiality moved. At Q1 the framing was that the opportunity across sovereign, neocloud and enterprise would ramp in the second half of FY26. This quarter the ramp still begins in the second half and is characterised as "really be FY '27." We are moving this pillar from confirmed to tracking. It remains genuine unmodeled upside and it is no longer a FY26 catalyst.

10. The NVIDIA Relationship Started Producing

Cisco tracks the number of active field engagements with NVIDIA, and that count grew 70% sequentially. Jensen Huang appeared at Cisco's AI Summit the week before the print. In the enterprise, Robbins characterized Cisco as the network standard where the two companies are partnered, with the AI Defense security portfolio bundled into AI factory solutions.

"We track the number of engagements we have in the field with NVIDIA. And while it was not a massive number leading up, we increased it by 70% from -- sequentially." — Chuck Robbins, Chair and CEO

Assessment: A 70% sequential increase on an admittedly small base is the correct shape for an early partnership but not yet a modelable revenue line. The strategic value is that it positions Cisco's networking and security layer as GPU-agnostic while riding the dominant GPU vendor's enterprise motion. Watch for the first quarter in which management attaches a dollar figure to it.

11. Capital Return and the Cash Cycle

Cisco returned $2,968M in Q2: a $0.41 dividend totaling $1,617M and approximately 18 million shares repurchased at an average price of $76.29 for an aggregate $1,351M. The buyback pace fell from $2,001M in Q1. The remaining authorization is $10.8 billion. The quarterly dividend was raised one cent to $0.42, an increase of 2%, alongside a restated commitment to return a minimum of 50% of free cash flow annually.

Operating cash flow of $1,822M fell 19%, which management attributed to the final $2.3 billion transition tax payment under the 2017 Tax Cuts and Jobs Act plus continued investment to meet demand. Capital expenditure was $283M, putting free cash flow near $1,539M for the quarter against $2,968M returned. Cash and investments ended at $15.8 billion, down from $16.1 billion at the fiscal 2025 year end.

Assessment: The 2% dividend increase is the smallest signal management could send while still sending one, and cutting the buyback by $650 million sequentially says more about the cash cycle than the dividend does. Neither is alarming: the transition tax payment is now permanently behind and free cash flow should normalize sharply in the second half. But a company paying out roughly twice its quarterly free cash flow while building inventory and advance commitments has less balance-sheet slack than the headline cash position suggests.

Guidance & Outlook

The guide is a study in contrast. On revenue it is aggressive: the FY26 midpoint of $61.45 billion sits $710 million above where the Street sat, implies 8.5% growth on FY25's $56,654M, and represents the second consecutive raise (the original FY26 frame given at Q4 FY25 was $59.0 billion to $60.0 billion). The Q3 revenue midpoint of $15.5 billion is $320 million above consensus. On margin it is a capitulation: 65.5% to 66.5% for Q3 non-GAAP gross margin against 68.6% in the year-ago quarter.

Both guidance assumptions on tariffs are unchanged from the prior quarter, with the Q3 and FY26 guides assuming current tariffs and exemptions hold through the end of fiscal 2026.

Implied H2 and Q4 ramp: H1 revenue was $30,232M. At the FY26 midpoint of $61.45B, H2 must deliver $31.22B. Subtracting the Q3 midpoint of $15.5B leaves roughly $15.72B implied for Q4, which is sequential growth of only 1.4% into what is normally Cisco's seasonally strongest quarter. On EPS, H1 delivered $2.05 non-GAAP; the FY26 midpoint of $4.15 leaves $2.10 for H2, and the Q3 midpoint of $1.03 implies roughly $1.07 for Q4, which would be growth of about 8% against Q4 FY25's $0.99.

Street at: Pre-print consensus sat at $15.18B and $1.03 for Q3, and $60.74B and $4.12 for FY26. The revenue guide clears all three revenue marks comfortably; the Q3 EPS guide brackets consensus rather than clearing it, and the FY26 EPS midpoint exceeds it by three cents.

Guidance style: Less conservative on revenue than Cisco's historical pattern, and unusually explicit on the margin problem. Two things are notable. First, management raised full-year EPS despite the gross-margin reset, which means the operating-expense and pricing offsets are being underwritten inside the guide rather than held as cushion. Second, the Q4 implied sequential of 1.4% is the softest exit-quarter shape in this cycle, and management explained it not as conservatism but as a consequence of hyperscaler order lumpiness. That is an honest answer and an uncomfortable one, because it means the guide's back half depends on order timing management cannot control.

Analyst Q&A Highlights

Whether Q3 Marks the Trough for Gross Margin

The first question of the call went straight to the number that would define the session, and it asked for two things: attribution and a trough call. Management delivered a four-part framework and neither. The response covered price increases already announced, contract terms under revision with partners and customers, the supply-chain scale argument evidenced by advance purchase commitments, and a redirect to earnings-per-share growth as the metric to judge them on. The word "trough" was not used in the answer.

Q: "And then maybe on a follow-up, Mark, you just touched on the gross margin decline in April. I assume it's all memory related, but would love to just understand what's happening there. And if you feel like that's a trough on the gross margin and memory issues or not."
— Amit Daryanani, Evercore ISI

A: "So really, as you look at the Q3 guide, there's 2 primary things at work. One is mix and the other is memory prices, as you mentioned. First off, I just want to acknowledge the significant growth in hardware that we're seeing across our existing and new platforms that have been introduced and the fact that they're accelerating much faster than previous launches. Secondly, in terms of memory, we're going to control what we can control."
— Mark Patterson, CFO

Assessment: A framework where a number was needed. Acknowledging mix first, before memory, was the most substantive part of the answer, because mix is the part that does not reverse when component prices normalize. Declining to call a trough on the opening question set the tone for the next hour and, we would argue, for the next session's tape.

The 1.4% Implied Q4 Sequential Ramp

The sharpest arithmetic of the call came from backing the fourth quarter out of the full-year guide. Cisco's seasonally strongest quarter is implied to grow only 1.4% sequentially against a normal 5% to 6%. The response worked the seasonality from the other end, pointing out that product revenue is typically down mid single digits sequentially in the reported quarter and was instead up 5%, so a low-single-digit Q3 is normal seasonality applied to an abnormally strong base. The second half of the answer was the more revealing one.

Q: "The second one is I can back out your 4Q guidance. And when I look at the sequential growth from Q3 to Q4, it's only 1.4%. So normally, it's 5%, 6%. This is a seasonally strong quarter. So why is it so low? Is it just conservatism? Or is it -- is there a growth concentration in the next 2 quarters?"
— Tal Liani, Bank of America Securities

A: "Yes. One other comment, Tal, I'd make on that is that I think with the nonlinear nature of the hyperscaler business, it creates a little bit of uncertainty relative to our historical numbers that you're so used to seeing."
— Chuck Robbins, Chair and CEO

Assessment: The most honest sentence on the call and the one with the longest half-life. Cisco is telling shareholders that as hyperscaler revenue grows as a share of the mix, the seasonal patterns that made this business forecastable stop applying. That is a permanent increase in the variance of the earnings stream, and it argues for a lower multiple independent of anything memory does.

Whether the Implied Q4 EPS Requires Margin Recovery

A second route into the same question came through the implied fourth-quarter earnings line, which sits a couple of cents above consensus and therefore requires something to improve. The answer conceded timing on the pricing recovery and then pointed to fiscal 2027 software growth as the second lever, which is notable given software revenue grew 2% in the quarter and total ARR grew 3%. The exchange was also the only point in the call where a follow-up question was deferred rather than answered.

Q: "Backing into the EPS guidance, it seems like EPS is a little higher than where the Street is in the fourth quarter, maybe a couple of pennies. And that would imply that the gross margin is expected to trough and get better? Or is that just due to operating margin?"
— Benjamin Reitzes, Melius Research

A: "So on gross margin, certainly, a lot of what we talked about relative to what we can control and some of the memory price increases that we've seen, some of that is just timing. So we think that, that will improve as we move forward. And then certainly, we're not here to talk about FY '27. But as you get into FY '27, we think that software growth will improve as well, and that will certainly help us."
— Mark Patterson, CFO

Assessment: The fiscal 2027 software argument is the weakest link in the margin defense, because the software and subscription metrics reported in this very quarter all decelerated. Management is asking shareholders to underwrite a reacceleration in the exact line that just slowed. The deferred follow-up, the only one of the call, came immediately after this answer.

Whether the AI Order Target Is Deliberately Conservative

Given the hyperscaler capital-expenditure announcements of recent weeks, a target of "in excess of $5 billion" invited the charge of sandbagging. The answer declined the compliment and instead surfaced the concentration disclosure that we regard as the most important new risk datapoint in the print.

Q: "But it certainly sounds like the over $5 billion of order numbers sounds a bit conservative given the momentum that you're seeing in sort of the CapEx programs that have been announced over the last couple of weeks. So maybe if you could talk to why that number is only moving up to in excess of $5 billion."
— David Vogt, UBS

A: "I think the thing to really remember about these customers is they're nonlinear. So -- and you just have to -- it's quite lumpy. There's only -- there's less than a handful of these major customers that are placing these orders. So we're giving you that number based on the pipeline we see."
— Chuck Robbins, Chair and CEO

Assessment: "Less than a handful" is the most specific concentration language management has offered, and it reframes the $2.1 billion order figure. Fewer than five customers generated a number equal to all of last fiscal year. The order book is real and it is narrow, and the guide's refusal to extrapolate it is credibility rather than conservatism.

Customer Response to Price Increases and Pull-Forward Risk

Raising prices into a demand upcycle invites two questions: does volume hold, and does it pull orders forward and hollow out later quarters. Management addressed both, and the pull-forward answer was unusually methodical about how the conclusion was reached, citing in-quarter linearity, pipeline timing, software activations and channel checks.

Q: "And then maybe just a second question, just in terms of the price increases, any demonstrable kind of change to demand or forward ordering that you saw as a result of that?"
— Meta Marshall, Morgan Stanley

A: "we really didn't see anything in the quarter that would point us to evidence that there were any significant pull forwards there at all. We looked at the typical things in terms of linearity in the quarter. We had -- frankly, we had good double-digit growth as we move through the entire quarter. We looked at whether we had any pipeline pull forwards from future quarters, didn't really see that, looked at software activations, as you'd expect, channel checks and whatnot. And we actually saw the pipeline begin to build really in the out quarters rather than being pulled forward."
— Mark Patterson, CFO

Assessment: The most useful answer on the call for judging the durability of the 18% order growth. Management named four independent tests and reported the result of each, which is a materially higher standard of evidence than the qualitative reassurance most of the margin answers offered. If the order growth were artificial, at least one of those four checks should have flagged it.

The Path Back to Growth in Security

Security has now declined for three consecutive quarters and the question asked for the leading indicators that would support a recovery in the balance of the fiscal year. The answer was structured in three parts (the Splunk accounting drag, new-product customer traction, and refreshed firewall units) and closed with an exit-rate commitment that was lower and narrower than the framing used in prior quarters.

Q: "However, could you speak to the order rates and new product demand traction of your security portfolio that could augur well for recovery in your security offerings for the balance of '26?"
— Karl Ackerman, BNP Paribas

A: "And where that would lead us is as we exit Q4 this year, the organic Cisco security portfolio will be growing revenue close to double digits as we exit. So the teams are a little -- they're a little behind where we -- not the teams, but this portfolio is a little behind where we thought we'd be exiting the year, but it's still performing relatively well."
— Chuck Robbins, Chair and CEO

Assessment: A concession delivered in the middle of a positive answer, and easy to miss. The exit-rate ambition dropped from the mid-teens framing to "close to double digits," and the scope narrowed to the organic portfolio, which excludes Splunk. Note also that no security order growth percentage was disclosed this quarter, having been given as +4% last quarter.

Whether Enterprises Will Pre-Buy Ahead of Announced Price Increases

The most commercially specific exchange of the call asked why sophisticated enterprise buyers would not simply accelerate orders ahead of publicly announced price increases, drawing the analogy to the supply-chain-driven order pull-in of several years ago. The answer conceded some pre-buying will occur and then made a component-content argument that materially bounds Cisco's exposure relative to its server-heavy peers.

Q: "And then why wouldn't they start -- why wouldn't customers, especially in the enterprise, start pulling in their orders here given the pricing increases that they know are coming. These are smart buyers."
— James Fish, Piper Sandler

A: "I would say that on the core networking side, the memory content is not quite as high as what you'd see in compute platforms. And so the price increases are more nominal than they are in sort of the compute systems that you see some of our competitors talking about and the things that are a larger percentage of their portfolio. So do I think customers will try to buy ahead in some cases, perhaps, but I don't think it's going to be a big trend in the networking side of our business."
— Chuck Robbins, Chair and CEO

Assessment: The single best argument on the call for why the margin reset should prove shallower than the guide implies, and the market did not weight it. Memory content per revenue dollar is structurally lower in switching and routing than in servers, which is consistent with the relative severity of the peer selloff that followed. It also cuts the other way: if the price increases are "more nominal," so is the recovery they can deliver.

What They're NOT Saying

1. The size of the memory hit: Management identified mix and memory as the two drivers of the Q3 gross-margin guide but never split them, never quantified the modeled memory inflation rate, and never put a duration on it. Mix does not reverse when component prices do, so the split is the difference between a cyclical and a structural reset. It was asked for directly and not provided.
2. Any fiscal 2027 gross margin framework: "We're not here to talk about FY '27" is a defensible position in February. It is less defensible when the same answer then leans on fiscal 2027 software growth as the reason margin improves. Shareholders were given the remedy without the model.
3. ARR, after raising it once: Total ARR growth decelerated to 3% from 5% and the dollar balance fell to $31.0 billion from $31.4 billion. It appeared in a single prepared-remarks sentence and in no answer. For a company that bought Splunk to build a recurring franchise, one sentence is not enough.
4. Splunk ARR and RPO: Last quarter management explicitly nominated Splunk ARR and product RPO as the honest measure of the business beneath the accounting drag, and reported both growing double digits. This quarter neither was mentioned, replaced by a new-logo count. Dropping a metric you nominated one quarter earlier is itself a disclosure.
5. Hyperscaler customer concentration: "Less than a handful" is the closest management has come to sizing it, and no percentage split of the $2.1 billion was offered. Whether two customers or four generated it is the difference between a durable franchise and a two-account exposure.
6. Any free cash flow figure or target: Free cash flow was mentioned once, inside the boilerplate commitment to return a minimum of 50% of it. No FY26 target was given, no quarterly figure was disclosed, and the capital-return math (roughly $3.0 billion returned against operating cash flow of $1.8 billion) went unremarked.
7. Observability: Zero mentions on the call in either prepared remarks or Q&A, in a quarter in which the segment decelerated from 6% growth to flat. It is 2% of revenue, and it is also the analytics thesis Splunk was supposed to accelerate.
8. What the AMD and HUMAIN joint venture is worth to Cisco: Up to one gigawatt by 2030 and a 100-megawatt Phase 1 are capacity figures, not Cisco revenue figures. No ownership split, no capital commitment, and no networking content per megawatt was disclosed.
9. Whether the buyback cut is deliberate: Repurchases fell to $1,351M from $2,001M sequentially while the dividend rose 2%. Neither the pace change nor the shape of second-half capital return was addressed, and no analyst raised it.

Market Reaction

• Pre-print setup: CSCO closed at $85.54 on February 11, up 11.0% year to date against 1.4% for the S&P 500, up 15.6% over the trailing 30 days and up 37.0% over the trailing twelve months. The 52-week closing range entering the print was $53.19 to $86.78, so the stock went into the number within 1.4% of its 52-week closing high.
• After-hours move: Shares fell roughly 7% in extended trading on February 11 and were down a similar amount in Thursday pre-market, with the guidance rather than the quarter cited as the cause.
• Next-day session: The stock gapped down 6.9% to open at $79.61, traded a range of $74.08 to $80.37, and closed at $75.00, down 12.3% or $10.54. The low of the day was set close to the close, so there was no intraday recovery attempt.
• Volume: 68.1 million shares against a 21.1 million 30-day average, or 3.2 times normal.
• Historical context: The 12.3% decline is the worst single session for CSCO since May 19, 2022.
• Peer reaction: Dell fell 9.1%, NetApp 7.3%, HPE 6.8%, Super Micro 5.0% and Arista 3.9%, against a 2.0% decline in the Nasdaq Composite and 1.6% in the S&P 500.

This was a component-cost repricing, not a Cisco repricing. Every hardware name with meaningful memory content in its bill of materials traded down on the day, and the ordering of the moves tracks memory exposure per revenue dollar almost exactly: servers and storage worst (Dell, NetApp, HPE, Super Micro), switching-only least affected (Arista at 3.9%). Cisco fell hardest because it went into the print priced closest to its highs and because it was the one that put a 65-handle gross margin guide on the tape. The read-through the market took from Cisco's guide was industry-wide, and Cisco supplied the datapoint that repriced its own sector.

The de-rating exceeded the earnings revision by a wide margin. The FY26 non-GAAP EPS guide went up, from a $4.11 midpoint to $4.15. At $85.54 the stock traded at 20.6 times the new midpoint; at $75.00 it trades at 18.1 times. That is 2.5 turns of multiple compression on numbers that improved. The market is not disputing FY26. It is disputing the quality and the variance of the earnings stream beyond FY26, and management gave it two reasons to: no fiscal 2027 margin framework, and an explicit statement that hyperscaler lumpiness has broken the seasonal predictability the multiple was partly built on.

Positioning mattered. A stock up 37% over twelve months and 15.6% in the preceding thirty days, sitting 1.4% below its 52-week closing high, has no cushion for a guidance line nobody modelled. The 3.2 times volume and the absence of any intraday bounce say this was position liquidation rather than a valuation debate resolving in real time. That distinction matters for what happens next: forced selling into a print does not require the fundamental case to have changed, and it typically overshoots.

Street Perspective

Debate: Is the Gross Margin Reset Cyclical or Structural?

Bull view: The bull case being made on the Street is that memory pricing is a classic cycle and Cisco is doing all the right things into it: prices already raised, contract terms being renegotiated, $1.8 billion of incremental advance purchase commitments securing supply, and a component-content profile that is structurally lighter on memory than any server vendor's. The Q3 guide is the trough quarter of a lagged pass-through, and gross profit dollars keep compounding regardless.

Bear view: The bear camp contends that management named mix before memory as the driver, and mix does not reverse. Hyperscaler AI systems and optics carry structurally lower gross margins than campus networking, and that revenue is growing fastest. Layer on services revenue declining, services gross margin falling 70 basis points, and a software and subscription base that decelerated on every measure this quarter, and the 67% to 69% gross margin band that supported the historical multiple is simply gone.

Our take: Both halves are right and the bears have the better near-term case. Memory is cyclical and Cisco's pass-through will work with a lag of two to three quarters. Mix is structural and gets worse before it gets better, because AI revenue is scheduled to grow faster than campus through fiscal 2027. The reasonable landing zone is a gross margin that settles a point or so below the old band rather than at the Q3 guide's low end, which is a real de-rating and a smaller one than 2.5 turns.

Debate: Does the Record Order Book Justify Looking Through the Margin?

Bull view: The optimistic camp argues that $2.1 billion of hyperscaler AI orders in one quarter, equal to all of fiscal 2025, plus 18% total product orders, plus 10% growth excluding hyperscalers, plus double-digit order growth accelerating in all three geographies, is the most complete demand picture Cisco has produced in a decade. More than $5 billion of fiscal 2026 AI orders against just over $3 billion of fiscal 2026 AI revenue means roughly $2 billion is already contracted into fiscal 2027, before the newly launched P200 and G300 silicon contributes anything.

Bear view: The skeptics answer that fewer than five customers generated that order figure, that management itself declined to extrapolate it, and that the implied fourth-quarter sequential of 1.4% shows the order strength is not translating into linear revenue. Orders that arrive lumpy from a handful of buyers deserve a lower multiple than an annuity, whatever their absolute size.

Our take: The order book is the strongest asset in the story and the concentration disclosure is the strongest argument against paying a premium multiple for it. Both should be reflected: raise the earnings estimates, lower the multiple you apply to them. That is roughly what the guide did and roughly what the market did, except the market did more of the second.

Debate: Is the Software and Subscription Deceleration a Signal or Noise?

Bull view: A growing consensus view is that the ARR and subscription-mix decline is arithmetic rather than deterioration. Subscription share falls automatically when hardware revenue grows 14%, and the Splunk transition from on-premises to cloud recognises revenue ratably, which suppresses reported figures during the transition without touching underlying demand. Long-term product RPO still grew 11%.

Bear view: The bear camp points out that total ARR fell in absolute dollars, from $31.4 billion to $31.0 billion, which no mix argument explains. Total RPO growth has fallen from 16% a year ago to 5%. Splunk ARR and product RPO, nominated by management last quarter as the honest measure, were not reported this quarter. And management's own fiscal 2027 gross margin remedy depends on software growth accelerating from the 2% just printed.

Our take: This is where we diverge most from the consensus framing, and we side with the bears. The recurring franchise was the reason a hardware-cycle company deserved a software-adjacent multiple, and it went backwards in dollars this quarter with no explanation offered and no question asked. We are adding it as a formal bear point and it is the metric we will lead with next quarter.

Model Update Needed

Valuation impact: At the $75.00 close, CSCO trades at 18.1 times the FY26 non-GAAP EPS guide midpoint of $4.15, against 20.6 times the same figure at the pre-print close of $85.54. We apply 19 to 20 times FY26 earnings, which is a full turn below where the stock traded into the print and reflects both the lower gross margin structure and the higher earnings variance management acknowledged. That gives a 12-month fair value of $80 to $86 (prior: $80 to $88), a midpoint of $83, which implies roughly 11% upside from $75.00. Adding the 2.2% dividend yield on the raised $0.42 quarterly rate gives an expected total return in the low teens. The range narrows rather than falls because the earnings estimate went up while the multiple we are willing to pay went down.

Thesis Scorecard Post-Earnings

Overall: The thesis is unchanged in direction and weaker in quality. Two of four bull pillars confirmed emphatically, one moved to a timing risk, one was challenged outright. Three of five carried bear points confirmed, one resolved by the selloff itself, and one new bear point added on the recurring-revenue franchise. The growth case is stronger than it was ninety days ago; the margin and mix case is materially worse. That is not a thesis break, it is a thesis with a lower ceiling and a wider distribution.

Action: Maintaining Outperform with reduced conviction. The order book, the second consecutive full-year raise, and the removal of a 20-times multiple justify staying long at $75.00 against a fair value of $80 to $86. We would not add on weakness below $70 without either a gross margin datapoint that confirms Q3 as the trough or a return to ARR growth, and we would revisit the rating outright if the Q3 print shows gross margin at the low end of the guide with no improvement in the recurring metrics.