Cisco's $9 Billion AI Order Book Answers Last Quarter's Questions. At 27 Times, the Price Answers Ours.

Results vs. Consensus

Year-over-Year Comparison

Sequential Comparison

Revenue

Total revenue of $15,841M grew 12.0% and beat the $15,560M consensus by $281M, a record quarter and above the high end of Cisco's own guide. The composition is a sharper version of last quarter's: product revenue grew 16.8% to $12,117M while services fell 1.4% to $3,724M, the third consecutive quarterly services decline. Within product, Networking alone contributed $8,815M at 25% growth, accelerating from 21% in Q2 and 15% in Q1. Security was flat at $2,008M, which is an improvement from Q2's 4% decline. Collaboration fell 1% and Observability grew 3%.

The order book is again the better forward read, and this quarter it is extraordinary. Total product orders grew 35%, up from 18% in Q2. Excluding hyperscalers they grew 19%, up from 10%. All three geographies accelerated into double digits (Americas 35%, EMEA 39%, APJC 25%), and all three customer markets grew (service provider and cloud 105%, public sector 27%, enterprise 18%). Networking product orders grew more than 50%, a seventh consecutive quarter of double-digit growth. Hyperscaler AI infrastructure orders of $1.9 billion compare with $0.6 billion a year ago and bring the year-to-date total to $5.3 billion, already past the roughly $5 billion management set as the full-year target ninety days ago.

Assessment: The demand picture is the strongest Cisco has produced in the five quarters we have covered, and the breadth is genuine rather than a hyperscaler artifact. The qualifier is that the ex-hyperscaler acceleration is not what it appears. Management volunteered that 4 to 5 of the 9 points of acceleration (from 10% to 19%) came from list-price increases on identical unit volumes. Backing that out leaves ex-hyperscaler unit demand growing at roughly 14% to 15%, which is still a real acceleration and roughly half of the headline improvement.

Margins

Non-GAAP gross margin of 66.0% fell 260 basis points year over year and 150 basis points sequentially, and landed precisely on the midpoint of the 65.5% to 66.5% range guided in February. All of the compression again sits in product: non-GAAP product gross margin of 64.3% fell 330 basis points on mix and memory costs, partially offset by productivity. Services gross margin of 71.6% rose 30 basis points, reversing the 70 basis point decline that was a new negative in Q2.

The geographic disclosure is where the mix argument stops being an assertion and becomes arithmetic. Americas total gross margin fell to 63.7% from 67.7%, a 400 basis point decline, while EMEA was essentially unchanged at 71.3% versus 71.2% and APJC fell 110 basis points to 66.1%. The Americas is simultaneously the segment carrying the hyperscaler revenue and the segment that grew fastest in mix, from 59.2% of total revenue to 60.4%. A memory-price shock would hit all three regions. This did not.

Below the gross line, the offset was total. Non-GAAP operating expenses of $5,046M grew 4.6% against 12.0% revenue growth, falling to 31.9% of revenue from 34.1%: 220 basis points of leverage against 260 basis points of gross-margin decline. Sequentially, operating expense was flat to the dollar at $5,046M on $492M of incremental revenue. Non-GAAP operating income of $5,412M was a record in dollars and grew 10.9%, with operating margin of 34.2% down 30 basis points from 34.5%.

Assessment: Management's claim that gross margin has stabilized is supportable on the guide but not yet on the drivers. What stabilized is the reported number against a guide set for it. The composition still deteriorated: product margin fell another 210 basis points sequentially, and the pricing actions that are supposed to arrest it contributed zero to Q3 revenue and arrive in Q4 alongside what management describes as a more acute memory cost. The operating-margin defense is real and it is finite. Holding operating expense flat in dollars on 3.2% sequential revenue growth is not a lever that extends indefinitely, particularly with a restructuring underway that management says is about redeployment rather than savings.

EPS

Non-GAAP EPS of $1.06 grew 10.4% and beat the $1.04 consensus by two cents. Compilers split on the consensus figure, with the two most widely syndicated at $1.04 and two others at $1.03; against the lower number the beat is three cents. Either way, the EPS beat is materially narrower than the revenue beat, which is the arithmetic of a 260 basis point gross-margin decline.

The bridge is cleaner than last quarter's. Non-GAAP net income of $4,227M grew 10.3% against revenue growth of 12.0%, and the entire gap is gross margin rather than anything below the line. Non-GAAP interest and other expense worsened to $194M from $236M of income drag (a modest help), the non-GAAP tax rate rose to 19.0% from 17.5% (a real headwind of roughly 150 basis points), and diluted shares fell 0.5% to 3,982M. Net income grew slightly slower than EPS because of the buyback, and that is the whole story.

GAAP EPS of $0.85 grew 37.1%, again far ahead of the non-GAAP figure, and the drivers are the same aging-intangibles and lower-charges effects as last quarter plus one new item. Acquisition-related intangible amortization fell to $449M from $507M, acquisition and divestiture costs to $90M from $214M, and restructuring charges were a $1M credit against a $34M charge. New this quarter: a $273M gain on investments flowed into GAAP results and was excluded from non-GAAP, worth roughly seven cents.

Assessment: EPS quality is good. There is no aggressive add-back, the tax rate moved against the company rather than for it, and the buyback contribution is half a point. The framing worth holding onto is the one management chose in February and reversed here: the Q3 guide had EPS growing slower than revenue, and the actual delivered 10.4% EPS on 12.0% revenue, which is still slower. The Q4 guide finally breaks the pattern the right way, with implied EPS growth of 17% on revenue growth of 14.5%.

Segment Performance

Networking

Networking is now carrying the entire company. Revenue of $8,815M grew 25%, an acceleration from 21% in Q2 and 15% in Q1, and it accounts for 72.7% of product revenue. Orders grew more than 50%, driven by triple-digit growth in service provider routing and compute plus double-digit growth in data center switching, campus switching, wireless, enterprise routing and industrial IoT. Within that, the campus refresh and the hyperscaler AI build are two distinct engines running at once, and management sized both.

"Networking product orders continue to accelerate, growing more than 50% in Q3, driven by triple-digit growth in service provider routing and compute and double-digit growth in data center switching, campus switching, wireless, enterprise routing and industrial IoT products. This marks the seventh consecutive quarter of double-digit growth for our networking portfolio overall."
— Chuck Robbins, Chair and CEO

Enterprise data center switching orders grew more than 40% and have grown double digits in seven of the past nine quarters, with Nexus orders tagged for AI deployments up almost 50% sequentially. That is the enterprise inference build, and it is separate from both the hyperscaler line and the campus line.

Assessment: A 25% revenue growth rate in a segment this size is the single most important number in the print, more than the AI order headline, because it is revenue rather than backlog and because it spans three independent demand pools. The risk it introduces is the margin one: this is the hardware that carries the lowest gross margin in the portfolio, and it is now 56% of total company revenue.

Hyperscaler AI Infrastructure

Hyperscaler AI infrastructure orders of $1.9 billion compare with $0.6 billion a year ago and $2.1 billion in Q2, the first sequential decline in this line since the cycle began. Year to date the total is $5.3 billion. Management raised the FY26 order forecast to approximately $9 billion from approximately $5 billion, which is 4.5 times the FY25 total of more than $2 billion, and raised expected FY26 AI revenue from hyperscalers to approximately $4 billion from approximately $3 billion. The Q3 product mix was described as balanced between Silicon One systems and optics, against roughly 60/40 in favor of systems in Q2.

Five new hyperscaler design wins landed in Q3: three systems (two Silicon One P200 for scale-across, one G200 for scale-out) and two Acacia optics wins (800G ZR+ and 100G ZR), each with a different hyperscaler. A third P200 win followed in the first weeks of Q4. Five of the top hyperscalers each grew orders in triple digits.

"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. 4.5x our FY '25 total."
— Chuck Robbins, Chair and CEO

Assessment: The raise is real and it is enormous, but the shape deserves attention that the tape did not give it. Reaching $9 billion requires approximately $3.7 billion of orders in Q4, roughly double the $1.9 billion just booked and roughly 75% above the $2.1 billion record set in Q2. Management's defense is that the business is nonlinear and the chart in the deck shows Q4 accelerating, which is a forecast rather than a backlog. The scale-across P200 wins, which are the most strategically important development in the quarter, contribute nothing to that Q4 number by management's own statement.

Acacia and Optics

Acacia booked more than $1 billion of orders in a single quarter, its strongest ever, up triple digits year over year, and is tracking to grow more than 200% for FY26. Cumulative coherent pluggable shipments now exceed 750,000 units at 400G and 40,000 units at 800G, which management believes far exceeds the next largest supplier at both speeds. The deck describes significant traction in 800G ZR with one hyperscaler and frames the scale-across use case as a 14 times bandwidth multiplier over front-end connectivity.

Assessment: Optics has quietly become more than half of the incremental AI order dollar, and it is a better business than the market credits: a component-level franchise with a shipment lead measured in multiples, selling into a transition (coherent pluggables displacing discrete transponders) that has cost and power arguments behind it rather than just capacity. It is also the piece of the AI story with the least visible competitive moat over a five-year horizon, which is why Silicon One matters more to the durability argument than Acacia does.

Campus Networking

Campus networking orders were a record, growing more than 25% and accelerating. Wireless orders were the highest ever, growing more than 40%, with Wi-Fi 7 now half the wireless mix and still growing at high double digits sequentially. Smart Switch orders grew triple digits sequentially off a small base, with the deck noting many customers still in test ahead of expected larger orders. The industrial IoT portfolio had its strongest quarter ever and has grown double digits for eight consecutive quarters.

Management sized the opportunity behind this for the first time in install-base terms, describing the pre-Catalyst 9000 install base as "in the tens of billions of dollars" and nearing end of support.

"Research conducted recently with around 3,500 technology leaders across global enterprises confirms increased urgency to modernize campus and branch networks. With traffic across these networks expected to increase 3x over the next 3 years because of AI, 93% of respondents are accelerating their network modernization plans."
— Chuck Robbins, Chair and CEO

Assessment: This is the leg of the story that justifies a durable multiple rather than a cyclical one, and it strengthened materially. A refresh driven by end-of-support obsolescence in a "tens of billions" install base is the most predictable revenue in networking, it carries the portfolio's best gross margin, and it is the direct offset to hyperscaler mix. It is also the leg that gets no credit in the tape, because the AI number is louder.

Security

Security revenue of $2,008M was flat year over year, an improvement from Q2's 4% decline and Q1's 2% decline, and the nine-month figure remains down 2%. The order picture is better than the revenue picture: the core portfolio excluding Splunk saw double-digit order growth across new and refreshed products, with strong double-digit order growth in firewalls specifically. More than 1,000 new customers bought the new products in Q3, bringing the cumulative total to approximately 5,000 since launch. Splunk's shift from on-premises to cloud subscriptions accelerated by another two to three points of mix and continues to suppress reported revenue.

"I think last call, I said we would exit this fiscal year approaching double-digit revenue growth on the traditional -- on the organic Cisco security portfolio. And I think we're heading in that direction, and I feel good about us actually making that happen."
— Chuck Robbins, Chair and CEO

Assessment: Reaching flat is genuine progress against a bear point that had been confirmed three quarters running, and the firewall order strength is the first leading indicator with real content behind it. The exit-rate commitment survives, narrowed to the organic portfolio, and now has one quarter of supporting evidence. It also requires a very large move in a single quarter to go from flat to approaching double digits, and management again offered no order-growth percentage for the segment as a whole.

Services, and the Recurring Franchise

Services revenue fell 1% for a third consecutive quarter, attributed to the timing of service contract start dates. The recurring metrics beneath it continued to erode. Total ARR of $31.2 billion grew 2%, down from 3% in Q2 and 5% a year ago; product ARR grew 4%, down from 6% and 8%; services ARR fell 1% for a third consecutive quarter. Total RPO of $43.5 billion grew 4%, down from 5% and 7%, with product RPO at 6% and services RPO at 2%. Subscription revenue of $7,791M fell 1.6% in dollars year over year and slipped to 49.2% of total revenue from 51.1% in Q2 and 54% in Q1. Software revenue of $5,660M grew 0.7%.

Assessment: This is the metric set that has now decelerated for three straight quarters on every line, and this quarter subscription revenue joined ARR in declining in absolute dollars. Part of it is arithmetic (subscription share falls when hardware grows 17%), part of it is the Splunk recognition transition, and part of it is that services ARR has been negative for three quarters with no transition to blame. Management raised it in one prepared-remarks sentence, and one analyst touched it as a follow-on to a Splunk question. It is the quiet counterweight to everything the AI line is doing, and it is the reason the multiple question is a real one.

Key KPIs

Key Topics & Management Commentary

Overall Management Tone: The most confident posture of the five quarters we have covered, and the first in which management volunteered arithmetic against its own headline rather than defending it. The pull-forward question, the price-versus-units split, and the gross-margin trough call were all answered with numbers rather than frameworks, which is the direct reversal of February, where the opening question about a margin trough drew a four-part framework and no figure. The one place management stayed vague was the forward AI order shape, where the answer to a specific question about a required doubling was a restatement that the business is nonlinear.

1. The AI Order Forecast Nearly Doubles

The number that moved the stock is the FY26 hyperscaler AI order forecast going to approximately $9 billion from approximately $5 billion, set only ninety days earlier, alongside AI revenue going to approximately $4 billion from approximately $3 billion. Cisco has now raised this target three times in four quarters: from roughly $4 billion implied at Q1, to above $5 billion at Q2, to $9 billion here. Against FY25's more than $2 billion of orders, the new figure is 4.5 times.

Asked directly why the earlier guide was so conservative, management gave an answer about the mechanics of hyperscaler procurement rather than about forecasting discipline.

"I think we've had just probably more design wins and more success than we expected at the beginning of the year is one thing. But the other thing is that sometimes these customers they make decisions. And when they decide to go, sometimes you don't know about it 3 months ahead. You find out about it and they're like, we're ready to move."
— Chuck Robbins, Chair and CEO

Assessment: The candor is welcome and it cuts both ways. If a $4 billion upward revision can appear inside one quarter because customers move without notice, the same visibility gap works in reverse. That is the concentration risk restated as a forecasting property, and it is the reason the order book, however large, deserves a lower multiple than an annuity of the same size.

2. The Q4 Order Requirement Nobody Priced

Hitting approximately $9 billion of FY26 hyperscaler orders against $5.3 billion booked through three quarters requires roughly $3.7 billion in Q4 alone. That is 95% above the $1.9 billion just printed and roughly 75% above the $2.1 billion record set in Q2. One analyst did the math on the call and asked what drives it.

"I just want to remind you, we've said that this business is nonlinear. And if you look at the chart that we put in with the prepared remarks, it really does indicate that Q3 was 1.9, a little lower than Q2, but then Q4 accelerates. So there's a lot of timing dependencies here... We would expect of those 5 design wins that we talked about, we'll begin to see some early orders in Q4 and -- but not at scale until we move into fiscal year '27."
— Chuck Robbins, Chair and CEO

Assessment: This is the softest disclosure in an otherwise hard-numbered call. A forecast that requires a near-doubling in one quarter, supported by a chart and a statement that the new design wins will not contribute at scale until next year, is a pipeline conviction rather than a booked position. It is also the number the market extrapolated into a 13% move. If Q4 orders land at $3.0 billion rather than $3.7 billion the fiscal-year revenue is unaffected, but the FY27 setup and the credibility of the forecasting cadence both take damage.

3. Gross Margin Stabilizes, and Management Says So

In February the opening question of the call asked whether Q3 would be the gross-margin trough, and the answer never used the word. This quarter management said it unprompted, twice, and put the Q4 guide midpoint at the same 66.0%.

"We saw gross margins in Q3 at 66%, which is right where we expected, right where we planned for and right at the midpoint of our guide. And we do believe that gross margins have stabilized. So if you look at that, it's reflected in our Q4 guide as well with the midpoint, again, right at 66%."
— Mark Patterson, CFO

The supporting actions are specific: more than twenty active programs to reduce memory utilization across the portfolio, including wireless products becoming orderable in Q4 that require 50% less memory; DDR4 to DDR5 conversion projects; hardware-only price increases; tightened quote-honoring terms; and a strategic investment in a supplier paired with a three-year supply agreement.

Assessment: The trough call is the second most important thing management said this quarter and it was largely ignored in the coverage. It converts the defining bear point of February from an open-ended reset into a bounded one. The caveat is that stabilizing at 66.0% is stabilizing 260 basis points below where the business ran a year ago, and the Q4 guide range is unchanged rather than raised, so what has been called is a floor, not a recovery.

4. Where the Margin Actually Went

The single most useful number in the deck is not in the press release headline. Americas gross margin fell 400 basis points to 63.7% while EMEA was flat at 71.3% and APJC fell 110 basis points. A component-cost shock is geographically indifferent. A customer-mix shift is not, and the Americas is where the hyperscalers are.

Management confirmed the ranking when pressed to split memory from mix.

"In terms of gross margins, the 2 biggest factors there, David, are certainly mix and memory. The bigger factor actually is mix, certainly. As you look at the growth of the business, you've got hardware really accelerating, which we're very pleased with, approximately 30% growth where you've got software at 1%."
— Mark Patterson, CFO

Assessment: Mix is the bigger factor, on management's own statement, for the second consecutive quarter. Mix does not reverse when memory prices normalize; it reverses when campus and software grow faster than hyperscaler hardware, which is the opposite of what FY27 is guided to look like. That is the case for a permanently lower gross-margin structure and, correspondingly, for a lower multiple than the software-adjacent one Cisco carried in the Splunk era.

5. Four to Five Points of the Order Acceleration Is Price

The ex-hyperscaler order growth going from 10% to 19% was the datapoint that made the quarter look broad-based rather than hyperscaler-dependent. Management disassembled it without being forced to.

"If we just -- what we did was we just ran basically those SKUs in Q3 and essentially 5 -- about 4 to 5 points worth of acceleration was just purely based on price, not incremental units, but just price."
— Mark Patterson, CFO

The price increases apply to hardware only, with nothing on software, and were targeted at products where Cisco is most competitive and memory utilization is highest. Separately, quote-honoring terms were cut from roughly 30 days notice plus 30 to 45 days of quote protection to 15 days plus 15.

Assessment: Voluntarily halving your own headline is unusual and it earns credibility. It also means the underlying ex-hyperscaler unit demand accelerated to roughly 14% to 15%, not 19%. That is still a strong number and it is a different number, and it matters for FY27 because price increases annualize into a comparison rather than compounding.

6. Silicon One P200 and the Scale-Across Opening

Cisco won its first two Silicon One P200 design wins for scale-across use cases in Q3 with two different hyperscalers, and a third in the opening weeks of Q4. Scale-across connects GPU clusters across data centers, and Cisco's deck frames it as a 14 times bandwidth multiplier relative to the front-end connectivity that coherent optics traditionally served. The deck also states an internal target of comprehensive Cisco-wide adoption of the Silicon One architecture for high-performance networking systems by FY29.

"I've said repeatedly on these calls over the last couple of years that as we move to the future, that if you don't have silicon, you're going to struggle to be relevant to the hyperscalers... It also gives us a lot more control over the supply chain and allows us to be a little more confident in our ability to deliver to our customers."
— Chuck Robbins, Chair and CEO

Assessment: Scale-across is the most consequential development in the quarter and contributes nothing to any current number. It opens a third addressable layer (after scale-out and front-end) at the exact moment multi-datacenter training is becoming the constraint, and Cisco arrives with both the silicon and the coherent optics. This is the part of the story that could justify the current multiple in three years. It is also entirely prospective, and prospective is what a 27 times multiple is already paying for.

7. The Supply-Chain Bet Now Sits on the Balance Sheet

Inventory rose to $4,708M from $2,832M a year ago, and inventory purchase commitments rose to $16,033M from $6,345M. Combined, Cisco has $20,741M of inventory and forward commitments against $9,177M a year ago, an increase of $11,564M, of which $6,766M came in the last ninety days alone. Silicon supply is secured through calendar 2026 with calendar 2027 negotiations underway. Management reported no supplier decommits in the quarter, against a competitor's public warning that decommits were possible.

"Overall, inventory and advanced purchase commitments, we're really able to lean in there given our financial strength. So those in totality, $6.7 billion increase just in the last 90 days, so up 48%. When you look at it year-over-year, again, inventory and advanced purchase commitments are up $11.6 billion year-over-year."
— Mark Patterson, CFO

Assessment: Using the balance sheet to buy supply certainty is the right call in this market and it is not free. Cisco has committed roughly $11.6 billion of incremental purchasing against a demand forecast that management describes as nonlinear and low-visibility. If the AI order pace holds, this converts to revenue and the commitments look prescient. If Q4 orders come in at $3.0 billion instead of $3.7 billion, this is where it shows up first, and it shows up as gross margin rather than as a writedown.

8. A $1 Billion Restructuring That Management Says Is Not About Savings

Cisco announced a restructuring plan on the same day as the print: up to $1 billion of pre-tax charges consisting of severance and other one-time termination benefits, with approximately $450 million in Q4 FY26 and the remainder in FY27, substantially complete during FY27. The stated purpose is realigning resources toward silicon, optics, security and AI.

"This was really not a savings-driven restructure. It's really -- things are moving incredibly fast right now. And this is more realigning from an already strong base as you're seeing in our financials, but really realigning resources around silicon, optics, security and AI. And so being able to move fast, we don't always have the exact resources that we need going forward in the right places. And so that's really what this is about versus savings."
— Mark Patterson, CFO

Assessment: Announcing a billion dollars of severance charges in the same release as record revenue and a doubled AI forecast is a confident move, and the framing is honest as far as it goes. But a plan described as "not a savings-driven restructure" that nonetheless carries up to $1 billion of severance charges is in tension with itself, and the tension matters for the model: if the charges are pure redeployment then FY27 operating expense does not fall, and the operating leverage that carried this quarter's operating margin has to come from revenue scale alone. Management declined to quantify the FY27 leverage effect when asked directly.

9. The First Fiscal 2027 Datapoint

Asked about the durability of double-digit growth, the CFO volunteered the first FY27 framing Cisco has given, ahead of the formal guide in ninety days.

"I think on the AI hyperscale side, it's probably reasonable to expect that we would recognize at least $6 billion of revenue in FY '27. Again, we'll have a more formal guide for you in 90 days. The rest of the portfolio, I think it's reasonable to assume it grows in line with our long-term model."
— Mark Patterson, CFO

The long-term model was subsequently clarified as 4% to 6% in total, with 2% to 5% applying to networking alone and higher rates assumed for security and observability. Applied mechanically, roughly $58.9 billion of non-hyperscaler-AI FY26 revenue growing 4% to 6%, plus at least $6 billion of AI revenue, implies FY27 revenue of roughly $67 billion to $68 billion, or 7% to 9% growth.

Assessment: This is the number that should anchor valuation and it is being under-discussed. Management put a floor of $6 billion on FY27 AI revenue against $4 billion in FY26, which is a $2 billion increment, not the $5 billion gap between FY26 orders and FY26 revenue. That is management's own read on how much of the order book converts next year, and it decelerates total company growth from roughly 11% in FY26 to roughly 8% in FY27. A floor can be beaten. It is still the only company-supplied FY27 datapoint, and it is below what a 27 times multiple implies.

10. Security's Recovery Reaches Flat

The organic security portfolio produced double-digit order growth across new and refreshed products with strong double-digit firewall order growth, and the legacy drag continues but "to a lesser extent than in the first half of the year." More than 1,000 new customers bought the new products in Q3, taking the cumulative total to roughly 5,000 since launch. Splunk's cloud mix shifted another two to three points and remains the reported-revenue suppressant. Management also announced intent to acquire Galileo and Astrix for agentic identity, access management and behavior monitoring.

Assessment: Flat is the right waypoint and the firewall data is the first genuinely leading indicator in three quarters. The exit-rate commitment is intact but the gap between flat in Q3 and approaching double digits in Q4 is a very large single-quarter move, and management again disclosed no segment-level order-growth figure to underwrite it. Grade this as on track and unproven.

11. Tariffs Quietly Eased, and Nobody Mentioned It

The Q3 guidance assumptions describe all countries at 10% with a semiconductor and electronic-component exemption, Mexico and Canada at 10% on goods not eligible for USMCA exemptions, and small copper, steel and aluminum tariffs. The February assumption set was materially heavier. The change was disclosed only in the deck's assumptions slide, was not mentioned in prepared remarks, and drew no question.

Assessment: A meaningful easing of the tariff regime is embedded in the raised guide and no one quantified it. That cuts against the quality of the raise at the margin: some portion of the improved outlook is an external policy change rather than execution. It also cuts the other way as a risk, because a reversal would hit a guide that now assumes the lighter regime holds through July.

Guidance & Outlook

The full-year guide implies a Q4 of roughly $16,827M against the nine-month actual of $46,073M, comfortably inside the stated range. That is a 6.2% sequential increase, against the 1.4% implied Q4 sequential that the February guide produced and that we flagged at the time as the softest exit-quarter shape of the cycle. The exit shape has been rebuilt on a base that is itself higher.

Implied Q-over-Q ramp: Q4 needs $16.7B to $16.9B, up 5.4% to 6.7% sequentially, with non-GAAP gross margin held at the Q3 level and operating margin improving. The pricing benefit finally reaches the revenue line in Q4 while the memory cost impact becomes, in the CFO's word, "much more acute." The two are designed to offset; the guide holds the margin band unchanged rather than raising it, which is management declining to underwrite a net positive.

Street at: The Q4 revenue midpoint of $16.8B was described across coverage as roughly $1 billion above where consensus sat, and the full-year raise took FY26 revenue growth from roughly 8.5% under the February guide to 11.0% here. FY26 non-GAAP EPS growth goes to 12.3% from FY25's $3.81.

Guidance style: Substantially less conservative than the February posture and, on the AI line, less conservative than Cisco's own history. Raising an order target by 80% one quarter after setting it is not a company managing expectations downward. The tell is that the margin band was left unchanged while every growth figure moved up, which is the shape of a company confident about demand and unwilling to promise anything on cost.

Analyst Q&A Highlights

Durability of Double-Digit Growth and the First Fiscal 2027 Frame

The opening question asked whether the accelerating back-half growth rate is durable or a pull-forward, and whether enterprises are genuinely upgrading their networking architecture. The answer went further than the question, offering the first FY27 revenue framing Cisco has put on record, three months ahead of the formal guide.

Q: "If I think about the back half guide, specifically the July quarter guide, it really implies that revenue growth is accelerating into the low teens, which is a nice bump up from the longer-term framework you guys have historically had. Can you just talk about the durability of this double-digit growth? Are you starting to see enterprises starting to upgrade their networking architecture and that's really what's helping it? Or is this a pull forward?"
— Amit Daryanani, Evercore ISI

A: "I think on the AI hyperscale side, it's probably reasonable to expect that we would recognize at least $6 billion of revenue in FY '27. Again, we'll have a more formal guide for you in 90 days. The rest of the portfolio, I think it's reasonable to assume it grows in line with our long-term model. And remember, the long-term model that we gave you initially, that included all the AI hyperscale opportunity as well."
— Mark Patterson, CFO

Assessment: The most valuable answer of the call, and the one that should discipline the valuation. Volunteering an FY27 floor ninety days early is a confidence signal. The floor itself is modest: $6 billion against $4 billion this year is a $2 billion increment against an order book carrying $9 billion, which tells you management's own view of the conversion rate is far below what the order headline implies.

Whether the Ex-Hyperscaler Acceleration Is a Pull-Forward

The most rigorous exchange on the call challenged the 19% ex-hyperscaler order growth as customers buying ahead of supply constraints rather than genuine demand. Management answered with three separate data points rather than reassurance, and conceded the premise in part before bounding it.

Q: "Non-AI orders grew 19%. That's excluding hyperscalers. Now the non-AI, the environment is not that better this quarter versus last quarter. And I'm wondering what drives it? And are you concerned that enterprises -- non-AI customers are buying ahead of time because of supply constraints, because of difficulties to get products on time."
— Tal Liani, Bank of America

A: "First off, let me just say that I think that it's reasonable to assume that there is some level of pull ahead into Q3. I do think it's very difficult to speculate exactly how much, but we believe it was a very modest amount that was in Q3. And I'll give you 3 data points as to why we think that... If you look at the Q4 pipeline that we pulled forward into Q3 and closed, we actually did not notice really any difference or any incremental pipeline that was pulled forward from Q3 -- sorry, from Q4 into Q3 than if you looked at the same time period a year ago."
— Mark Patterson, CFO

Assessment: A model answer, and the reason we treat the order acceleration as real. Conceding some pull-forward, quantifying half the acceleration as price, checking the Q4 pipeline for depletion, and reporting it grew through the quarter is four falsifiable claims where one platitude would have sufficed. It is also the exchange that reduces the headline 19% to roughly 14% to 15% of unit demand, which no coverage of the print picked up.

Supply Chain Tightness and the Required Q4 Order Jump

A two-part question asked about industry-wide supply tightness, including a competitor's public warning about possible decommits, and then asked what drives the roughly $3.7 billion of Q4 AI orders implied by the raised full-year forecast. The supply half was answered in detail. The order half was not.

Q: "You've got a very significant increase implied in your AI order intake into fiscal 4Q, up from $1.9 billion. I guess if my math is right, close to $3.7 billion. Just curious, what's driving that? Is that the scale across wins? Are you starting to maybe see scale up build in your pipeline? I'm just curious if you can unpack that pretty notable jump into this next quarter."
— Aaron Rakers, Wells Fargo

A: "I just want to remind you, we've said that this business is nonlinear. And if you look at the chart that we put in with the prepared remarks, it really does indicate that Q3 was 1.9, a little lower than Q2, but then Q4 accelerates. So there's a lot of timing dependencies here that suggest remember that this is nonlinear as we go to the future... So there's actually no scale across in the numbers yet."
— Chuck Robbins, Chair and CEO

Assessment: The sharpest question of the call and the least satisfying answer. A required 95% sequential increase in the single line item that drove a 13% move in the stock was defended with a chart and a restatement of nonlinearity. Management confirmed that the scale-across wins, the most obvious candidate for the jump, contribute nothing to it. That leaves the Q4 forecast resting on pipeline conviction alone, and it is the single largest disconfirmable claim in the quarter.

Splitting the Gross-Margin Headwind Between Memory and Mix

The question that went unanswered in February was asked again: split the product gross-margin headwind between memory cost and product mix. This time it drew both an answer and an unprompted trough call, plus a preview of the operating-margin defense management intends to run if gross margin deteriorates further.

Q: "Just on a second question, just any way to kind of split the gross margin headwind on product just between memory and product mix?"
— Meta Marshall, Morgan Stanley

A: "We saw gross margins in Q3 at 66%, which is right where we expected, right where we planned for and right at the midpoint of our guide. And we do believe that gross margins have stabilized... I'll just give you an example that we saw in Q3, while we had gross margins declining 2.6%, if you look at OpEx actually, it also declined more than 2% as a percentage of revenue. A year ago, it was 34.1%, and it was 31.9% this quarter. So I think there's going to be businesses we get into that have different scale. And so there's a lot we can do to protect that 34% op inc."
— Mark Patterson, CFO

Assessment: The stabilization claim is the substantive concession and it resolves February's defining ambiguity. The split still was not given in basis points. What was given instead is more revealing: an explicit statement that the company will defend the 34% operating margin through operating expense if gross margin keeps falling, and an acknowledgment that Cisco is entering businesses "that have different scale." That is management telling shareholders the gross-margin structure is permanently different and the operating line is the one to hold them to.

Whether Memory or Mix Dominates as AI Revenue Grows 50%

A follow-on line of questioning pressed on what offsets the mix headwind if AI revenue grows from $4 billion to $6 billion in FY27 while software grows 1%. The answer ranked the two drivers explicitly and, in doing so, sized the hardware growth rate against the software growth rate in a single sentence.

Q: "How much of that margin dynamic is largely driven by the mix to hardware? ... if we expect $6 billion of AI revenue next year, up from $4 billion, like what are the offsets that you're thinking about to help mitigate sort of that product mix versus effectively software security mix, if that's going to grow 50% next year from a baseline of $4 billion this year?"
— David Vogt, UBS

A: "The bigger factor actually is mix, certainly. As you look at the growth of the business, you've got hardware really accelerating, which we're very pleased with, approximately 30% growth where you've got software at 1%. So it tells you that our hardware margins are actually really good. And it also tells you that the supply chain team that we have is world-class."
— Mark Patterson, CFO

Assessment: Hardware at roughly 30% growth against software at 1% is the entire margin thesis in one comparison, and management offered it as evidence of hardware margin quality rather than as a warning. Both readings are correct. The hardware margins are good for hardware margins. They are structurally below software margins, and the gap between the two growth rates is widening, not narrowing, which is why the FY27 mix headwind is larger than the FY26 one.

The Restructuring and Fiscal 2027 Operating Leverage

Given a billion dollars of announced charges, a question asked whether FY27 operating leverage should be modeled better than FY26 and how much of the savings gets reinvested. The answer rejected the premise of the question rather than answering it.

Q: "With the restructuring that you're announcing today, should we think about the operating leverage in the business to be marginally better next year compared to what we saw this year? How should we think about sort of how much of those savings get reinvested in the business and how to think about operating leverage next year?"
— Samik Chatterjee, JPMorgan

A: "Just with regard to the restructuring, this was really not a savings-driven restructure. It's really -- things are moving incredibly fast right now. And this is more realigning from an already strong base as you're seeing in our financials, but really realigning resources around silicon, optics, security and AI."
— Mark Patterson, CFO

Assessment: A clean non-answer on the modelling question, and it should be read as guidance in itself. If the restructuring produces no net expense reduction, then FY27 operating expense grows with the business and the 220 basis points of leverage that carried this quarter does not repeat. Combined with a gross margin management has only floored, that puts FY27 operating margin expansion in question at exactly the moment the multiple assumes it.

Pricing Mechanics and the Timing of the Revenue Benefit

Two consecutive questions worked through the pricing strategy: what was raised, how quote-honoring terms changed, and critically when the benefit reaches revenue rather than orders. The timing answer is the one that matters for the Q4 model.

Q: "As you look forward, is it fair that in the July quarter, you'll get a more significant impact from pricing being fully baked into the full quarter? Is that a dynamic here that is part of your guidance?"
— George Notter, Wolfe Research

A: "On the revenue side, you really didn't see any impact in Q3 per se that was due to the price increases just based on timing of when you take the order and when it goes into the build plan, et cetera. You'll start to see some of that benefit in Q4, which is good because that's really -- those are the actions that are helping us stabilize gross margins. So the impact of higher memory prices is actually much more acute in Q4, but you also have the price increases helping us in terms of what we'll ship out in Q4."
— Mark Patterson, CFO

Assessment: This exchange reframes the stabilization claim into something testable. Q3's 66.0% was achieved before any pricing benefit and before the worst of the memory cost. Q4 gets both at once, and management guided the same band rather than a higher one. The implication is that pricing is expected to roughly offset the incremental memory cost, no more. If it under-delivers, Q4 lands at the low end and the trough call is retracted within a quarter.

Why the Original AI Order Target Was So Conservative

A direct challenge asked whether the 80% upward revision reflected a market that turned out larger than expected or a forecasting posture that was deliberately low. The answer was neither, and the mechanism it described is the concentration risk stated from the inside.

Q: "Can you comment on why the fiscal '26 AI orders were so conservative? I know you mentioned nonlinearity in customer orders, but is there also a function of market being larger than you anticipated?"
— Sam Feldman (for Karl Ackerman), BNP Paribas

A: "In some cases, these massive opportunities will arise and you won't know about them and then all of a sudden, you have a big number in the pipeline 30 days later. So hopefully, it continues, but that's sort of what I would attribute it to."
— Chuck Robbins, Chair and CEO

Assessment: Honest, and more consequential than it sounded on the call. A revenue stream where a multi-billion-dollar swing can appear inside thirty days from a handful of buyers has, by construction, a wide distribution of outcomes in both directions. The market priced the upper tail on Thursday. The same disclosure is the reason a lower multiple, not a higher one, belongs on this order book.

What They're NOT Saying

1. The headcount in a $1 billion severance charge: The filing describes "severance and other one-time termination benefits" totaling up to $1 billion and never states how many people. Press coverage supplied a figure of fewer than 4,000 and under 5% of headcount; the company did not, and no analyst asked. For a plan management describes as redeployment rather than savings, the size of the redeployment is the whole disclosure.
2. The memory-versus-mix split, again: Asked for a second consecutive quarter and answered with a ranking rather than basis points. Mix is named as the bigger factor both times. The split is the difference between a cyclical trough and a permanent structure, and it remains unquantified while management asserts stabilization.
3. Any split of the $5.3 billion of hyperscaler orders: Five of the top hyperscalers each grew in triple digits, which is broader than the "less than a handful" framing of February, and still no percentage or dollar concentration was offered. The FY27 revenue floor of $6 billion depends on which customers those are.
4. Splunk ARR and product RPO, for a second consecutive quarter: Management nominated both two quarters ago as the honest measure of the business beneath the accounting drag. Neither has been disclosed since. Replaced again by a new-logo count, now framed as being on track to exceed 1,000 for FY26.
5. Why total ARR growth halved in a year: Total ARR growth went from 5% to 3% to 2% across three quarters and services ARR has been negative for three. The metric appeared in one prepared-remarks sentence and in no answer. The only ARR question asked was a Splunk question.
6. Whether the $9 billion order figure includes any multi-year commitments: The gap between $9 billion of FY26 orders and a $6 billion FY27 revenue floor is either duration, cancellation assumption, or conservatism. Management never said which, and the distinction determines whether the order book is backlog or pipeline.
7. Any free cash flow figure or target: The deck states capital return was 87% of free cash flow in Q3 without disclosing the free cash flow figure itself. Year to date, dividends of $4,894M plus buybacks of $4,605M total $9,499M against operating cash flow of $8,791M less capital expenditure of $1,020M. That is 122% of free cash flow, with total debt up $3,210M since the fiscal year end. None of it was raised on the call.
8. The tariff assumption change: The guidance-assumption slide now reads all countries at 10% with a semiconductor exemption and Mexico and Canada at 10% on non-USMCA goods, materially lighter than the February set. It was disclosed only in a slide footnote, mentioned in neither prepared remarks nor Q&A, and its contribution to the raised guide is unquantified.
9. What the G300 is doing: The 102.4T G300 appears in the deck alongside the P200 and was named in February as being outside the original AI order target. It produced no disclosed design wins this quarter and was not mentioned on the call.
10. Observability, for a second consecutive quarter: Zero mentions in prepared remarks and Q&A, in a segment that is 1.7% of revenue and is the analytics thesis the Splunk acquisition was supposed to accelerate.
11. Whether the buyback pace is a decision: Repurchases have declined for three consecutive quarters, from $2,001M to $1,351M to $1,252M, while the average purchase price rose from $68.28 to $76.29 to $80.28. Buying less as the price rises is defensible capital allocation and nobody asked management to confirm it is deliberate.

Market Reaction

• Pre-print setup: CSCO closed at $101.87 on May 13, up 32.2% year to date against 8.7% for the S&P 500, up 23.7% over the trailing 30 days and up 64.9% over the trailing twelve months. The 52-week closing range entering the print was $61.29 to $101.87, so the stock went into the number at its 52-week closing high.
• After-hours move: Shares rose sharply in extended trading on May 13, with reported moves ranging from roughly 15% to 20% across outlets quoting at different points in the session. The dispersion reflects thin extended-hours quoting rather than a disputed fact.
• Next-day session: The stock gapped up 15.4% to open at $117.55, traded a range of $113.97 to $119.36, and closed at $115.53, up 13.4% or $13.66. The $119.36 high is an all-time intraday high.
• Volume: 70.9 million shares against an 18.8 million 30-day average, or 3.8 times normal.
• Historical context: The 13.4% gain is Cisco's best single session since 2011.
• Index comparison: The S&P 500 rose 0.8% on the session.

The mechanism was a single number. Raising the FY26 hyperscaler AI order forecast to approximately $9 billion from approximately $5 billion, one quarter after setting it, converted Cisco from a networking company with an AI attachment into an AI infrastructure supplier in the market's classification scheme. The revenue beat was 1.8% and the EPS beat was two cents; neither moves a $400 billion company 13%. The re-rating did.

Positioning amplified it in the opposite direction from February. Three months ago the stock went into the print 1.4% below its 52-week closing high with 37% of trailing gains behind it, and a guidance line nobody modelled produced forced selling. This time the stock went in at its 52-week closing high with 65% of trailing gains behind it, and the print validated the position rather than breaking it. The 3.8 times volume with a close 3.2% below the intraday high is the signature of buyers chasing an opening gap and then meeting supply, not of accumulation.

The de-rating of February has been fully reversed and then some. At the $75.00 close on February 12 the stock traded at 18.1 times the then-current FY26 guide midpoint of $4.15. At $115.53 it trades at 27.0 times the raised midpoint of $4.28. The earnings estimate rose 3.1%; the price rose 54.0% in three months. Roughly a tenth of the move is the earnings revision. The rest is multiple.

Street Perspective

Debate: Does the $9 Billion Order Forecast Justify an AI-Infrastructure Multiple?

Bull view: The bull case being made on the Street is that Cisco has crossed a classification threshold. A company taking $9 billion of hyperscaler AI orders in a single fiscal year, with its own merchant silicon, the leading coherent pluggable optics franchise, and the first scale-across design wins in an emerging multi-datacenter architecture, is not a 15 times networking incumbent. It is an infrastructure supplier in the same cycle as the accelerator and optical names, and those trade at multiples far above 27 times. Price targets moved into the $130 to $150 range on that logic.

Bear view: The bear camp contends the classification is being applied to the wrong denominator. Hyperscaler AI revenue is approximately $4 billion of an approximately $63 billion company, or 6%, growing to a company-floored $6 billion of roughly $67 billion, or 9%. The other 91% grows at 4% to 6% by management's own long-term model. Paying 27 times for the blended entity prices the 6% segment as though it were the whole company, and it does so on an order line that just declined sequentially and requires a 95% sequential jump next quarter to hit its target.

Our take: The bears have the arithmetic and the bulls have the option. Both are right about different time horizons, and the horizon is what the price is choosing. Management's own FY27 framework, the only company-supplied one that exists, produces roughly 8% revenue growth and roughly 9% EPS growth. A 27 times multiple on that is not a valuation, it is a bet that the FY27 floor is far too low. It might be. Our job is to note that the bet is now the entire investment case, which it was not at $75.

Debate: Has Gross Margin Actually Stabilized?

Bull view: The optimistic camp argues the trough is in and management has now said so on the record. Q3 landed on the guide midpoint, Q4 is guided to the same band, twenty-plus memory reduction programs are in flight with hardware coming to market requiring half the memory, silicon supply is locked through calendar 2026, DDR4 to DDR5 conversions are underway, price increases hit revenue in Q4, and services gross margin already turned positive. Gross profit dollars grew 7.8% regardless.

Bear view: The skeptics answer that what has been called is a floor, not a recovery, and it sits 260 basis points below where the business ran a year ago. Product gross margin fell another 210 basis points sequentially. Management named mix ahead of memory for the second straight quarter, and mix worsens in FY27 by design as AI revenue grows 50% while software grows 1%. Q3's 66.0% was achieved before any pricing benefit and before the memory impact turned "much more acute," and the Q4 band was left unchanged rather than raised.

Our take: We move from the bear side of this debate to the middle, which is a change from February. The trough call is credible on the memory half and the actions behind it are concrete rather than aspirational. The mix half is not a trough at all and management has stopped pretending otherwise, which is why the operating margin has become the metric they ask to be judged on. The reasonable landing zone is a gross margin in the mid-60s rather than the old 67% to 69% band, permanently, with the operating margin defended at 34% through expense discipline that has a floor of its own.

Debate: Is the Decelerating Recurring Franchise a Signal or an Artifact?

Bull view: A growing consensus view is that the ARR and subscription figures are pure arithmetic in a quarter when hardware grew 17% and total revenue grew 12%. Subscription share falls mechanically when the hardware denominator explodes. The Splunk transition from on-premises to cloud recognises revenue ratably and suppresses reported figures during the shift without touching underlying demand, and Splunk is on track to exceed 1,000 new logos this year. Product ARR still grew 4% and product RPO still grew 6%.

Bear view: The bear camp points out that subscription revenue fell 1.6% in absolute dollars year over year this quarter, joining total ARR which fell in dollars last quarter, and that no denominator argument explains a numerator going backwards. Services ARR has now declined for three consecutive quarters with no transition to blame. Total ARR growth has gone 5%, 3%, 2% and product ARR 8%, 6%, 4% across the same three quarters. Splunk ARR and product RPO, which management itself nominated as the honest measure, have gone undisclosed for two quarters running.

Our take: We keep the bear position we took in February and we note it has gotten worse, not better, in the quarter that everything else improved. This is the pillar that entitled a hardware-cycle company to a software-adjacent multiple, and it has now deteriorated on every published line for three consecutive quarters while the multiple went up nine turns. That combination is the clearest single argument for the downgrade, and it is the one the market spent Thursday not discussing.

Model Update Needed

Valuation impact: At the $115.53 close CSCO trades at 27.0 times the FY26 non-GAAP EPS guide midpoint of $4.28 and roughly 24.6 times our FY27 estimate of $4.70. In February we said we would apply 19 to 20 times FY26 earnings. We raise that materially, to 22 to 25 times FY27 earnings, because three things genuinely changed: the gross-margin reset has been floored rather than left open, the order book roughly doubled, and the campus refresh gained an install-base sizing that supports a multi-year rather than cyclical framing. Applied to $4.70, that gives a 12-month fair value of $105 to $118 (prior: $80 to $86), a midpoint of $111.50, which is roughly 3% below the close. Adding the 1.5% dividend yield on the $0.42 quarterly rate gives an expected total return of roughly negative 2%. The range moves up 35% at the midpoint and still brackets the price rather than sitting above it, which is precisely the difference between the fundamentals improving and the stock being cheap.

Thesis Scorecard Post-Earnings

Overall: The thesis strengthened materially and the stock strengthened more. All four bull pillars are now confirmed or on track, including Bull #4 which was at risk ninety days ago and is the clearest operational improvement in the quarter. Two of the six carried bear points weakened (margin bounded, security reaching flat), one broadened favorably without resolving (concentration), one worsened decisively (the recurring franchise), one re-emerged (free cash flow and the debt-funded working-capital build), and one was reinstated in full (valuation). A seventh has been added on FY27 operating leverage. This is a better business than we underwrote in February by every operational measure that matters, and it is priced at 27 times against 18 times.

Action: Downgrading to Hold from Outperform. This is a valuation call, and we want to be precise about that because the temptation after a print like this is to read a downgrade as a fundamental doubt. It is not. Management answered five of the eight commitments we carried into this quarter and answered them well. The rating changes because the price moved 54% in three months against a 3.1% increase in the fiscal-year earnings estimate, and because our own FY27 framework, built from the only forward numbers the company has supplied, produces a fair value of $105 to $118 against a $115.53 close. We would return to Outperform on either of two developments: Q4 hyperscaler AI orders landing at or above the roughly $3.7 billion the guide requires, which would validate the forecasting cadence and reset FY27 well above management's floor; or a pullback into the low $100s with the recurring metrics stabilizing. We would move to Underperform if Q4 gross margin lands at the low end of the guide while total ARR growth stays at or below 2%.