AI Infrastructure Orders Blow Past FY25 Target a Quarter Early, Sovereign Optionality Emerges, CFO Transition Manageable — Upgrading to Outperform, FV $70–82

Rating Action

We are upgrading Cisco Systems, Inc. (CSCO) from Hold to Outperform with a fair-value range of $70–82, anchored on ~18–21x our refreshed FY26 non-GAAP EPS framework of $3.95–4.05 (which we now expect formally guided at Q4 FY25). The upgrade rests on three updates relative to our Q2 initiation: (1) the AI-orders trajectory has been re-rated — not just “is it real” but “how big does it get”; (2) sovereign-AI optionality (HUMAIN, G42, AIP) is a forward leg the Q2 thesis did not have; (3) NVIDIA's explicit Spectrum-X partnership validates the enterprise-AI bridge thesis at the GPU incumbent's reference-architecture layer.

The triggers we set at Q2 are answered cleanly. AI orders YTD >$1B with a quarter to spare? Yes — Q3 alone >$600M with the original FY25 target already crossed. Organic product orders ex-Splunk holding double-digit? Tracking at +9%, in the low-double-digit zone we wanted. Tariff resolution clarifying? Partially — Q3 actuals came in favorable to the conservative Q2 guide, though Q4 still bakes in the full proposed regime with no mitigation. Security ex-Splunk is the one trigger we cannot clean-cleanly verify because Cisco lapped the Splunk acquisition this quarter and stopped breaking out organic. We think this is reasonable given the integration completion but flag it as a transparency loss.

What would change our view at Q4 FY25? An FY26 guide that is materially below the Street's current modeling on the AI-revenue ramp; a sustained step-down in webscale AI orders below the $600M Q3 cadence; a sovereign-deal slip suggesting the Middle East narrative is more political than commercial. None of these is our base case.

Results vs. Consensus

Above the high end on every metric Cisco guided. Composition: order growth (+20% reported / +9% organic) is the leading indicator, AI orders is the durability indicator, and the FY25 guide tightening is management putting the upside back into the model in a measured way (raising the low end by $500M and the EPS low end by $0.09).

Segment Performance

Networking — Back to Growth on the Right Composition

Reported revenue +8% with growth across most of the portfolio led by double-digit growth in switching and enterprise routing, partially offset by a decline in servers. The composition that matters: this is order-driven growth following through, not backlog-flush optics.

• Campus switching: Orders up high-single-digits against a tougher prior-year compare. WiFi 7 portfolio orders triple-digit sequential growth in Q3 — the access-point cycle turning is the leading indicator of the broader campus refresh that is forecast to accelerate into FY26.
• Data-center switching (Nexus): YTD orders up double-digits. Cisco was named market leader in Gartner's Magic Quadrant for data center switching during the quarter. The 51.2T Silicon One G200 began shipping in Q3 — the chip is now “at the heart of any of these systems orders,” in Robbins's words, with two-thirds of the >$600M AI orders systems-based and G200-anchored.
• Industrial IoT (ruggedized Catalyst): YTD orders +35%. Reads as continued benefit from US re-shoring / sovereign infrastructure spend. Meraki for Government also achieved FedRAMP authorization.
• Webscale: SP & Cloud orders +32% with three of the top six webscalers each growing triple-digits; AI orders >$600M.

Security — Reported +54%, the Largest Splunk Deal Ever Booked This Quarter

• Reported revenue: +54% YoY, primarily from Splunk and SASE.
• Splunk synergy proof-point: A large multi-year deal with a major financial-services company for Splunk's security and observability platforms — driven by the combined Cisco-Splunk salesforce, won from an industry competitor, the largest Splunk deal ever. This is exactly the cross-sell evidence the Q2 thesis was waiting on.
• New product traction: Secure Access, XDR, Hypershield collectively added >370 new customers in the quarter. The majority of new Hypershield enterprise customers are bundling with the N9300 Smart Switch — the network-embedded security thesis converting commercially.
• Organic ex-Splunk: Cisco lapped the Splunk acquisition during Q3 (last year's Q3 included six weeks of Splunk contribution); Herren told the Street he no longer plans to break out organic-vs-inorganic going forward. Reads as transparency reduction we'd flag, but the underlying Splunk performance — in line on revenue, slightly ahead on profitability, integration largely complete — carries the thesis.

Collaboration — Quietly Stable

Reported +4% — growth in Devices, Webex Suite, and CPaaS. Better than the +1% Q2 print; the on-prem Webex secular decline is being absorbed by hardware refresh and CPaaS adoption. Not a thesis driver but no longer a thesis drag.

Observability — +24% Reported

• Reported +24% YoY (including Splunk Observability + ThousandEyes).
• SnapAttack acquisition closed in Q3 — adds capability to Splunk for the SOC of the future.

Key Topics & Management Commentary

The AI-Orders Trajectory: $350M → $600M+ → YTD >$1B One Quarter Early

The original FY25 $1B target is hit with a quarter to spare. The composition shift is meaningful: where Q2 orders were roughly half systems / half optics, Q3 orders were more than two-thirds systems with the balance in optics. The systems mix is anchored on the Silicon One G200 51.2T chip that began shipping during the quarter. Robbins on the demand-vs-capacity reality:

“Those customers are telling us that they — if we could get more capacity out, they would buy more.”
— Chuck Robbins, CEO, on G200-based systems

The webscale customer concentration disclosure: five of the six largest webscalers are deploying Cisco 8K / Silicon One / optics combinations in their back-end training networks. Three grew triple-digits in Q3.

Sovereign AI: HUMAIN, G42, and the AIP — Forward Optionality, Not Current Order Book

Cisco announced engagement with HUMAIN (the new Saudi Arabian AI company), expanded G42 collaboration in the UAE, and joined the AI Infrastructure Partnership (BlackRock GIP, MGX, Microsoft, NVIDIA, xAI, GE Vernova, Nextera Energy). Robbins's framing of the HUMAIN scope:

“It's hundreds of billions of dollars at the end of the day that they will be spending. ... I think there'll be as big as any of the major web scalers in the United States is how I would think about it.”
— Chuck Robbins, CEO, on HUMAIN

Robbins also disclosed the personal relationship: HUMAIN's CEO Tareq Amin was previously CTO at Reliance Jio (where Cisco built that network) and CEO of Rakuten Mobile (where Cisco built the Open RAN network). None of this is in the Q3 order book. The team was due back in the Middle East the week after the call. We treat sovereign as forward optionality with a 2H FY26 / FY27 ramp profile.

NVIDIA Partnership: Silicon One Inside Spectrum-X

The structurally significant disclosure of the call: Cisco Silicon One will become the only third-party silicon included as part of NVIDIA's Spectrum-X Ethernet networking reference architecture. Cisco will also build interoperable systems combining NVIDIA Spectrum silicon with Cisco operating system software. This positions Cisco as NVIDIA's enterprise-AI networking partner of record — the bull-case enterprise-AI bridge crystallizing at the GPU incumbent's blueprint level. Combined with the Cisco Secure AI Factory with NVIDIA (security end-to-end from application to workload to infrastructure), this is the enterprise-AI commercialization story making the leap from concept to architecture.

CFO Transition: Herren Out at FY25 End, Patterson In

Scott Herren retiring at end of FY25 after five years that drove the software-and-subscription transition. Mark Patterson (current CSO, 25-year Cisco veteran with deep finance + sales + strategy experience) elevated to CFO effective Day 1 FY26. Jeetu Patel promoted from EVP/GM Security & Collaboration to President and Chief Product Officer. Kevin Weil (CPO at OpenAI) joins the Board. The optics: a leadership broadening rather than a void; integration is past the people-side completion line; Patterson's portfolio (strategy + 25 years finance + 10+ years sales) fits a company pivoting from product-cycle execution to AI partnerships and sovereign deals. We treat as manageable.

Tariffs: Q3 Came In Favorable to Conservative Estimate; Q4 Bakes In Full Effect

Herren's Q2 GM guide of 67–68% was built on the full proposed China +10% / Mexico-Canada +25% with no mitigation. Q3 actual non-GAAP GM came in at 68.6% — favorable to the estimate. Q4 guide 67.5–68.5%, reflecting a full quarter of current tariffs (China 30%, partially offset by semi/electronics exemptions; Mexico/Canada 25% on non-USMCA-eligible; rest of world at 10% base until July 9 then reverting to country-specific reciprocal tariffs). Cisco's supply-chain mitigation lever (which previously reduced China exposure by ~80% over time) is again the gating variable. Pricing is a lever; not the first one Cisco would pull.

Capital Return: $3.1B in Q3, $9.6B YTD, Buyback Authorization Healthy

Q3 capital return $3.1B ($1.6B div + $1.5B buyback). YTD $9.6B with $15.4B remaining under the buyback authorization. Operating cash flow $4.1B (+2%), driven by revenue and earnings growth, lapping last year's tax-payment quirk. Buyback pace ahead of the FY25 framework we modeled at Q2.

Q4 FY25 & Full-Year Guidance

Read: The FY25 tightening is measured — raising the low end without raising the high end, which preserves Q4 conservatism. The Q4 GM guide implies modest sequential erosion from Q3's 68.6%, which is consistent with the full-quarter tariff effect Herren laid out and the July 9 reciprocal-tariff revert assumption. We expect formal FY26 guidance at Q4 FY25 alongside the AI-orders re-base — this is the next material catalyst.

Analyst Q&A — Notable Exchanges

Q&A was again AI-orders-heavy, with sovereign and CFO-transition threads layered in. The most thesis-relevant exchanges:

• Meta Marshall (Morgan Stanley) opened on tariff pull-forward (none seen across channel inventory, Meraki activations, linearity, future-ship-date pattern, pipeline) and federal (US federal +double-digits in Q3 despite a big-deal slip; civilian still stressed but DoD/IC is 75% of US federal).
• Tal Liani (BofA) probed whether 2025 is a peak year for cloud capex; Robbins's framing that sovereign cloud strategies are emerging globally and the customer signal is multi-year was the most thesis-relevant push-back.
• Aaron Rakers (Wells Fargo) drew the Saudi/HUMAIN context (zero in the $600M Q3 book; team returning the following week) and confirmed G200 51.2T at the heart of two-thirds-systems Q3 mix.
• Joseph Lima Cardoso (for Samik Chatterjee, JPMorgan) got the HUMAIN sizing framing (“hundreds of billions of dollars”; networking, compute, security, observability scope) and enterprise campus context (triple-digit sequential WiFi 7 orders; strong enterprise routing; agentic AI driving network modernization).
• Michael Ng (Goldman Sachs) drew the organic-revenue-disclosure withdrawal (Herren: integration complete, no longer plans to break out organic post-lap) and confirmed 12–18-month campus refresh cycle.
• Matt Niknam (Deutsche Bank) probed webscale share-shift dynamics; Robbins framed the InfiniBand-to-Ethernet migration as gated on customer comfort with the technology, with silicon diversity as the strategic differentiator and Silicon One as the “key” competitive moat.
• Amit Daryanani (Evercore ISI) got the AI-orders forward-growth framing (non-linear; capacity-gated; if Cisco can build more, customers will buy more) and the Q4 GM math (full-quarter tariff cost, no mitigation, sequentially down ~50bps mid-to-mid).
• Andrew Spinola (for David Vogt, UBS) drew the AI-orders concentration framing (3 of 6 webscalers triple-digit; balanced rather than one-customer-dominated).
• Karl Ackerman (BNP) got the NVIDIA partnership context and CPO (co-packaged optics) framing — Cisco demonstrated CPO in 2023; will deliver when customer demand materializes.
• Adrienne Colby (for Atif Malik, Citi) got the AI-pipeline disclosure withdrawal (no longer breaking out forward AI pipeline) and Oliver Tuszik named new head of go-to-market (former EMEA region lead, former global partner organization head).
• Sebastien Naji (William Blair) drew the white-box-vs-systems framing (no meaningful share shift; systems still predominant; two-thirds of >$600M Q3 was systems).

What They're NOT Saying

• No formal FY26 guide. The FY26 framework will come at Q4 FY25 alongside the FY25 close. The Street's modeling will sharpen between now and then; this is the next material catalyst.
• No organic ex-Splunk Security disclosure post-lap. Herren explicitly said going forward Cisco won't break out organic vs. inorganic given integration completion. Reasonable, but the Q2 thesis pillar of “needs to ramp toward HSD by year-end FY25” is now unverifiable from the outside. Trust-the-management territory.
• No customer-by-customer AI orders dollar split. Three of six webscalers triple-digit, but no “customer A is X%” framing. Concentration risk is still latent.
• No sovereign deal sizing. HUMAIN framed as “hundreds of billions” in Saudi spend with Cisco as a strategic partner; no Cisco-share number, no order-book contribution. Realistic given timing — the team was just headed back to KSA — but will become the durability question for sovereign at Q4.
• Limited commentary on M&A pipeline. SnapAttack closed in Q3 (Splunk SOC capability); no further deals signaled.

Market Reaction

The print landed after-hours on May 14. Initial after-hours reaction was constructively positive on the AI-orders blow-through and the FY25 tightening; the Herren retirement disclosure and CFO transition introduced a brief ambiguity but the Patterson elevation (a known internal candidate) absorbed it cleanly. The trading session on May 15 confirmed the buyside reception — AI-orders trajectory, Saudi sovereign optionality, and the NVIDIA Spectrum-X partnership were the three planks the buyside underwrote. Volume was elevated; CSCO outperformed enterprise-tech peers on the day.

The reaction is consistent with a market beginning to re-rate the franchise on the AI-and-sovereign narrative. We read this as the multiple-expansion leg starting to commence — though not yet fully priced. This is what underpins the Outperform call.

Street Perspective

The bull case being made on the Street post-print has hardened: (1) AI infrastructure orders are now a compounding, not one-shot, business — YTD >$1B with a quarter to spare and the demand-vs-capacity dynamic confirms backlog-driven durability; (2) sovereign AI engagement (HUMAIN/G42/AIP) opens a multi-year ramp on top of the webscale base, with Cisco positioned as a systems integrator across networking, security, observability; (3) the NVIDIA Spectrum-X partnership validates the enterprise-AI bridge thesis at the GPU incumbent's blueprint level; (4) the FY25 raise plus disciplined capital return (94%+ of FCF) supports cash-flow-driven EPS accretion through the cycle.

The bear case being articulated centers on: (1) the Q3 organic-disclosure withdrawal removes a key trust mechanism on Splunk-cross-sell verification; (2) Q4 GM guide of 67.5–68.5% bakes in the worst-case tariff regime and reveals Cisco is still vulnerable to fluid trade policy; (3) the CFO transition mid-cycle adds execution risk — even with a strong internal successor, FY26 guide-day will be Patterson's first — the Street prices that as added uncertainty; (4) services revenue +3% (vs. +6% in Q2) suggests the orders-to-services ratable pattern is decelerating, which would compound any product-side slowdown in FY26.

Our read: bull pillar (1) has clearly become the dominant thesis driver; pillars (2)/(3) add forward optionality but are not yet in the order book; the bear pillars are real but each is bounded. The favorable Q3 tariff outcome plus the >$1B-a-quarter-early on AI orders is the empirical case for the upgrade.

Model Implications

• FY25 revenue: We now model $56.65B (the high end of the tightened company guide of $56.5–56.7B), reflecting Q3 outperformance and Q4 momentum.
• FY25 non-GAAP EPS: $3.78 (midpoint of $3.77–3.79). Our Q2 conservatism on 50–75 bps tariff erosion has played out softer than feared; Q3's 68.6% GM gives confidence in the FY25 trajectory.
• FY26 framework (preliminary, refined post-Q4): Revenue $58.5–60.0B; non-GAAP EPS $3.95–4.10. AI infrastructure orders FY26 at $1.5–2.0B+ (above current implied trajectory); $3B+ if sovereign engagement converts to firm orders by mid-FY26.
• AI infrastructure revenue conversion: Q3-Q4 should show the first material AI-revenue contribution; Herren confirmed conversion in 2H FY25 with a bigger leg in FY26. We model ~$300–500M FY25 AI revenue and ~$1.5–2.0B FY26.
• Splunk: Integration complete on the people side. Cross-sell evidence (largest Splunk deal ever, Q3) supports the cross-sell thesis. We model Splunk in line with internal expectations; the disclosure withdrawal is a transparency cost we accept.
• Capital return: FY25 on pace for ~$13B (vs. our $11B Q2 model). $15.4B remaining authorization runway is healthy.

Thesis Scorecard

Overall: Thesis pillars are converging in our favor. Three of five bull pillars confirmed; two more tracking. Two of the four bear pillars upgraded to active but bounded; webscale concentration latent. The empirical case from Q3 is strong enough to upgrade.

Action: Upgrading from Hold to Outperform; fair value $70–82 (~18–21x our refreshed FY26 non-GAAP EPS framework of $3.95–4.05). The AI-orders trajectory has crystallized; sovereign-AI optionality is the new leg; the NVIDIA partnership validates the enterprise bridge. The next material catalyst is the formal FY26 guide at Q4 FY25 — if it lands at or above the Street's current modeling, the multiple expansion has further to run.

Net: The two clean prints we wanted before paying for the upgrade are done. Q2 was the “is it real” print; Q3 is the “how big does it get” print. We move to Outperform.