Khan Capitals | May 2026
Key Takeaways
- Cisco’s networking revenue jumped 25% year on year to $8.82bn, well ahead of the $8.47bn consensus, with the company raising full-year guidance for the third quarter running. The result delivered the sharpest single-day move in the stock since 2023.
- The print reframes the AI infrastructure trade away from the silicon-only narrative. Switching, optics and Ethernet fabrics are now visibly absorbing hyperscaler capex that has run at a $725bn annualised pace into 2026.
- Order growth of 30% across Cisco’s AI infrastructure book provides a forward read on Q2 hyperscaler capex before Microsoft, Alphabet, Meta and Amazon report their July prints. Component bookings typically lead reported capex by one to two quarters.
- The result lands at an awkward moment for fixed-income markets. Hot April CPI and PPI prints have pushed the 10-year above 4.45% and the 30-year through 5.00%, leaving high-multiple equities to digest a rising real cost of capital even as their earnings backdrop improves.
- The market is treating Cisco’s beat as confirmation of the AI thesis rather than as a signal about its breadth. The wider implication is that incumbent infrastructure names with installed-base economics may be the cleanest expression of the trade in the next twelve months, particularly if chip-export uncertainty around the Beijing summit lingers.
The Number That Mattered
Cisco Systems closed its fiscal third quarter on Wednesday 13 May with revenue of $15.84bn and adjusted earnings per share of $1.06, comfortably ahead of the $15.56bn and $1.00 consensus. Shares rose 13.4% in Thursday’s session, the largest single-day move in the stock for more than two years. The standout line item was networking revenue, which climbed 25% year on year to $8.82bn against a consensus closer to $8.47bn. Guidance for the fourth fiscal quarter was raised, and the company increased its full-year revenue outlook for the third successive print.
For a business that spent the better part of a decade trading as a low-growth value compounder, the rate of change in networking is the more important number than the absolute beat. Cisco’s networking segment had been contracting modestly through the 2022-2024 enterprise digestion cycle. A 25% growth print indicates that the company has now lapped that hangover and entered an order book dynamic that looks substantively different from the pre-pandemic enterprise refresh cadence.
Management attributed the strength to AI-related infrastructure orders, which they pegged at over $2bn for the quarter and more than $5bn cumulatively across the past four reporting periods. The company also called out hyperscaler customers by name, a marked shift from a vendor that has historically positioned itself as an enterprise-led franchise.
Why Networking Is the Tightest AI Bottleneck
Training large language models is now a problem of moving data between GPUs as much as it is a problem of computing on them. A modern training cluster requires high-radix Ethernet or InfiniBand fabrics capable of handling petabit-scale east-west traffic with deterministic low latency. Optical interconnects, programmable switches and the silicon photonics that link them are no longer commodity components: they are the structural bottleneck in cluster scaling. The shift toward inference at the edge of large training runs has only deepened this dependency, since inference workloads benefit disproportionately from low-latency switching at the rack level.
The economic implication is straightforward. As cluster sizes grow, the dollar share of networking inside the AI infrastructure stack rises. Cisco’s commentary implied that the AI-attached portion of its networking business is growing at a multiple of the segment headline, consistent with industry estimates that networking now represents 15-20% of cluster capex against 5-7% in the H100 era. The leverage runs both ways: if hyperscaler capex compounds, switching revenue compounds faster.
The Mag 7 Capex Tide and Who Catches It
The aggregate hyperscaler capex line for 2026 now sits at approximately $725bn on consensus, with the largest four cloud builders accounting for roughly $625bn of that figure. Microsoft, Alphabet, Meta and Amazon have each guided capex higher at least twice across the past three quarters, a pattern Khan Capitals has previously chronicled. The question for investors is not whether the money is being spent, but which vendor income statements absorb the spend with the most operating leverage.
The picture below summarises how the dollars are distributing across the principal infrastructure verticals on current consensus estimates. The figures are indicative and reflect 2026 calendar year orientations.
| Infrastructure layer | 2026 capex absorbed (USDbn) | Share of stack | YoY growth | Principal beneficiaries |
|---|---|---|---|---|
| AI accelerators (GPU/ASIC) | 320 | 44% | +38% | Nvidia, AMD, Broadcom, custom silicon |
| Networking and optics | 130 | 18% | +52% | Cisco, Arista, Marvell, Coherent |
| Memory and storage | 95 | 13% | +44% | SK Hynix, Samsung, Micron, Pure |
| Power, cooling, real estate | 120 | 16% | +30% | Eaton, Schneider, Vertiv, Equinix |
| Software, orchestration, tooling | 60 | 9% | +22% | Hyperscalers in-house, niche ISVs |
The change in networking’s relative share is the structural story. In the H100 era of 2023, switching and optics absorbed perhaps 7-8% of cluster spend. The arrival of clusters built around Nvidia’s Blackwell and successor architectures, with their dramatically higher per-GPU bandwidth requirements, has shifted the fabric ratio decisively. The numbers above understate the durability of the move: optics and switching have higher gross margins and substantially longer replacement cycles than the accelerators themselves, which means the installed base compounds at a slower amortisation rate.
Why Cisco’s Beat Differs from the H100 Era
Through 2023 and most of 2024, the AI infrastructure trade was effectively a single-name expression: Nvidia. Its monopoly on the accelerator layer captured the bulk of the value chain narrative, and adjacent infrastructure suppliers spent the period trading as cyclical laggards. The April 2025 cracks in the soft-landing consensus, followed by the broadening of capex commitments to in-house silicon at Alphabet and Meta, started to change that picture. Cisco’s print provides the cleanest confirmation to date that the AI infrastructure budget has now diversified beyond the accelerator layer into a recognisable supply chain.
Three features of the Cisco print distinguish it from earlier optical and networking beats. First, the order growth at 30% exceeds revenue growth, meaning backlog is building. Second, gross margin held at the high end of the guidance band, suggesting that pricing power is not being eroded as volumes scale. Third, hyperscaler concentration in the order book is rising, which historically would have been a discount factor but in the current environment effectively underwrites two to three years of visible demand.
By way of contrast, the equipment-and-services beats of the 2017-2019 enterprise cycle were characterised by single-digit margin compression and lengthening enterprise sales cycles. The current beat looks very different in shape: it is closer in profile to the 2010-2012 period when Cisco rode the smartphone backhaul wave, with the important difference that the present cycle has both higher visibility and higher attached software content.
Hyperscaler Disclosure and the Q2 Capex Read
Cisco’s fiscal Q3 covers February to April, which means the order book has a useful read-through to the calendar Q2 capex prints that hyperscalers will deliver in their July reporting. Networking equipment orders typically lead reported capex by one to two quarters, since the bookings precede shipments and the shipments precede the capitalised investment line. On that basis, the 30% order growth implies that aggregate hyperscaler capex is unlikely to moderate in the second half of 2026, regardless of the consolidating commentary from sell-side analysts who have been calling for a deceleration since late 2025.
It is worth noting where the order growth is concentrated. Cisco called out three of the four major US hyperscalers by direct reference, plus a sovereign customer category that almost certainly maps to Gulf and Asian state-aligned AI infrastructure programmes. The Gulf programmes in particular have accelerated since the H1 2025 announcements around the UAE and Saudi national AI strategies, both of which sit downstream of the regional reorientation set in motion by the Iran conflict. State-aligned infrastructure orders carry different sensitivity to interest rate cycles and to mark-to-market discipline than private-sector capex, which adds a margin of stability to the order book.
The Bond Market’s Awkward Counterpoint
The Cisco print arrived in the same week that the 10-year Treasury yield touched 4.49% and the 30-year crossed 5.00% for the first time since the late summer of 2025. April CPI printed at 3.8% year on year, the highest reading since May 2023, and producer prices delivered a 1.4% monthly gain, the sharpest move since the immediate aftermath of the 2022 supply shock. The energy pass-through from the Iran war has flattened the disinflation glide path and arguably reversed it.
The arithmetic implication for high-multiple equities is uncomfortable. A 30-year real yield around 2.50%, a level not seen consistently since 2008, raises the bar for discounted future cash flows. The Nasdaq Composite and the S&P 500 closed at fresh records on Thursday all the same, with the S&P 500 logging its seventh consecutive weekly gain. The dissonance between yields at 2026 highs and equities at all-time highs is the defining feature of the week.
Cisco’s profile within this backdrop is notable. Its forward price-earnings multiple even after the rally trades in the high teens, well below the Mag 7 cohort and a long way below the pure-play accelerator names. That valuation cushion is precisely what makes the AI infrastructure trade tractable in a rising-yield environment: investors can express the thesis through compounders that do not require multiple expansion to deliver.
What the Market Is Underappreciating
The initial reaction to Cisco’s print treated it as confirmation of the AI thesis. The more useful reading is about the breadth of the thesis. A networking-led beat of this magnitude indicates that the AI capex tide is broadening, not just rising. That has two implications that the immediate price action did not fully capture.
First, the cohort of infrastructure equities that benefits has widened beyond the obvious silicon names. Arista, Marvell and Coherent on the networking side, plus the cooling and power names (Vertiv, Schneider, Eaton) on the physical infrastructure side, all derive incremental order flow from the same hyperscaler capex line. The valuation gap between these compounders and the pure-play accelerators creates a relative-value opportunity that the market is only beginning to acknowledge.
Second, the breadth of the cycle reduces the binary risk associated with single-name disappointments. If the AI infrastructure thesis depended entirely on a single vendor’s quarterly print, any soft datapoint could trigger a 5-7% correction in the broader cohort. With more visible distribution of demand across the supply chain, isolated misses are likely to be absorbed at the sub-segment level rather than rerating the whole space. That is a useful change in the cross-sectional volatility profile of AI-exposed portfolios.
Third, and perhaps most overlooked, the geopolitical layer that surfaced this week through the Trump-Xi summit and the conditional licensing of H200 exports to ten named Chinese firms cuts more cleanly through the accelerator layer than through networking. Cisco’s exposure to Chinese hyperscalers is structurally lower than Nvidia’s, which means its order book is more insulated from the export-licence overhang. That asymmetry is a feature of the current cycle, not a bug.
Investor Implications
For equity portfolios, the Cisco print supports a more balanced expression of the AI infrastructure thesis than the single-name accelerator trade. Positioning that has been concentrated in Nvidia and a small number of custom-silicon adjacencies may benefit from broadening into the networking and physical-infrastructure layers, where valuation multiples remain meaningfully lower and where the cyclical leverage to capex is now demonstrable. Investors may wish to consider the trade-off between higher beta in the accelerators and lower-multiple compounders with comparable revenue growth profiles in 2026.
For fixed-income portfolios, the combination of accelerating capex and rising real yields argues against an aggressive duration extension. The 30-year yield above 5% reflects both higher inflation expectations and an emerging term premium that may not compress quickly even if growth softens. Positioning at the belly of the curve has historically offered the cleanest risk-reward when the long end is repricing for term premium rather than for growth.
For cross-asset positioning, the relevant observation is that the dollar has weakened on the week even as Treasury yields have risen. That combination is unusual and reflects the structural rotation away from US duration as a safe haven, a theme covered in earlier Khan Capitals analysis of the Q1 2026 correction. Gold’s 43% year-to-date gain and the dollar index sitting just under 99 are consistent with this rotation. Portfolios with structurally underweight allocations to non-US infrastructure equities may consider whether the relative-value gap has widened sufficiently to warrant rebalancing.
The framing here is observational rather than prescriptive: the data is consistent with a regime in which AI infrastructure capex compounds through the second half of 2026, real yields stay structurally elevated, and the dollar’s reserve premium continues its slow erosion. Investor allocations should reflect rather than fight that pattern.
Conclusion
Cisco’s fiscal Q3 print did three things at once. It confirmed that the AI infrastructure cycle has broadened materially beyond the accelerator layer. It provided a useful forward read on hyperscaler capex through the second half of 2026. And it demonstrated that incumbent infrastructure compounders with installed-base economics can deliver meaningful earnings surprises without requiring multiple expansion to drive the share price.
The print does not resolve the harder question of how high-multiple equities digest a 5% long bond and an inflation impulse that has now flattened the disinflation glide path. But it does suggest that within the AI infrastructure space, the cleanest expression of the trade in the next twelve months may sit one layer below the accelerators, where the order books are visibly accelerating and the valuations have yet to fully reflect the cycle. That is the institutional read of the week’s most consequential earnings event.
Sources: CNBC: Cisco CSCO Q3 earnings report 2026; Bloomberg: 10-year yield hits highest since July after PPI; CNBC: Treasury yields after CPI; Cisco Systems Investor Relations: Quarterly Results; JP Morgan Asset Management: Weekly Market Recap; Charles Schwab: Q1 Tech Earnings Preview; FRED: 10-year Treasury constant maturity yield; Federal Reserve H.15: Selected interest rates.
Related Reading: The broader hyperscaler capex thesis was explored in AI Capex Hits $725bn: Wall Street Splits on the Hyperscaler Trade, with the longer-run version chronicled in AI Capex Boom: $600 Billion and Counting. The accelerator-layer single-name exposure question is treated in Nvidia’s AI Supercycle, while Tesla’s $5 Billion Capex Surprise tracked the cross-Mag 7 capex acceleration. This week’s parallel macro story is captured in PPI Shock 2026: Records Above 7,400. For the next leg of this story, see our analysis of Dell’s $51.3 billion AI server backlog.
Leave a Reply