It's time for Cisco to wake up.

On August 12 local time, Cisco released its financial results for the fourth quarter and the full fiscal year 2026, with all core indicators comprehensively exceeding Wall Street's expectations. Revenue for the quarter reached 17.3 billion U.S. dollars, up 18% year on year, higher than the widely expected market forecast of 16.8 billion U.S. dollars; non-GAAP earnings per share stood at 1.22 U.S. dollars, also surpassing analysts' estimate of 1.17 U.S. dollars. Full-year revenue hit 63.3 billion U.S. dollars, representing a 12% year-on-year increase, with net profit of approximately 13.4 billion U.S. dollars.

Driven by exceptionally strong demand for AI infrastructure, Cisco secured 4 billion U.S. dollars in related orders in this single quarter, bringing the full-year total to 9.3 billion U.S. dollars, exceeding the company's recently raised target of 9 billion U.S. dollars.

Cisco's management describes the current situation as a "network supercycle", believing that it is driven by two concurrent forces. On the one hand, hyperscale data centers are carrying out large-scale GPU cluster construction, which has led to a surge in demand for 800G and 1.6T high-speed Ethernet architectures. On the other hand, a large number of enterprise campus network devices have entered the replacement window, which, combined with the popularization of Wi-Fi 7 and network security upgrades, has formed rigid demand independent of the AI boom.

After excluding ultra-large customers, Cisco's product orders still grew by 25%, and double-digit growth was achieved in all regions around the world, confirming the broad resilience of enterprise-side demand.

However, there are hidden concerns beneath the impressive figures.

Non-GAAP product gross margin decreased by 270 basis points year on year to 64.8%, which is mainly squeezed by the strong bargaining power of hyperscale manufacturers and the high costs of chips and optical components for high-speed AI switches.

The management's guidance for gross margin in the next quarter is 65% to 66%, indicating that the pressure will continue.

Rising orders, thinner profits?

At least on the surface, Cisco's financial figures are almost impeccable. The only pity is that its stock price trend seems to directly expose the market's true attitude.

In after-hours trading, Cisco's stock price once rose to 132 U.S. dollars and then quickly pulled back. This trend of rising first and then falling shows that after digesting the details of the financial report, the market began to re-evaluate the quality of growth.

Why did the stock price fall even when the results beat expectations? The key lies in the expectation gap.

The market has priced in quite high expectations for Cisco's AI story. Although the financial report exceeded expectations, it did not provide sufficient incremental evidence to support a higher valuation.

More importantly, this financial report reveals a contradiction: Cisco's revenue growth is increasingly dependent on low-margin AI orders, and the rising proportion of low-margin orders is diluting the overall profitability.

It is reported that in the fourth quarter, Cisco's non-GAAP product gross margin decreased by 270 basis points year on year to 64.8%, and the total non-GAAP gross margin dropped from 68.4% to 66.3%. At the same time, the non-GAAP operating profit margin reached 35.9%, with operating profit increasing by 23%. While the gross margin declined, the operating profit margin rose instead, which is driven by operating leverage.

Revenue growth dilutes fixed R&D, sales and administrative expenses, boosting profit margins in the short term, but this improvement masks a fragile structure: the rising proportion of low-margin products means that the overall gross margin center is moving downward.

Once revenue growth slows down, expense stickiness will amplify the profit decline in the opposite direction.

The market sees this very clearly.

Operating leverage can amplify in both directions, and gross margin is the thermometer of pricing power. The decline in Cisco's gross margin indicates that the added value of its AI products is being squeezed by both upstream and downstream parties. Upstream chip and optical component suppliers have technical barriers, while downstream hyperscale manufacturers have the bargaining power from centralized procurement. Caught in the middle, Cisco retains less profit from every dollar of revenue.

This involves the core logic of the so-called economic rent. Cisco once had considerable rent in the traditional enterprise network market. The number of enterprise customers is large but scattered, and individual customers have limited bargaining power. Cisco has built high customer stickiness and pricing power through proprietary protocols, certification systems and long-term service relationships.

This market structure allows Cisco to maintain a gross margin above 65% for a long time while maintaining stable cash flow. The AI backend network market is a completely different ecosystem.

The number of customers is very small, but the procurement scale is extremely large. Every contract has gone through repeated bidding, and the procurement teams of hyperscale manufacturers have a clearer understanding of the cost structure than most suppliers.

Under this buyer monopoly pattern, the pricing power of suppliers is systematically weakened.

Cisco's entry into this market means that it has voluntarily left its familiar territory with rich rents and stepped into a battlefield with thin rents.

A 35% increase in orders is certainly spectacular, but the profit margin corresponding to these orders is far lower than the historical level of the enterprise network business.

The market responded to this reality with a fall in stock prices.

Investors see the revenue growth as well as the decline in gross margin; they see the explosion of orders and also the reduction of profit retained from every dollar of revenue.

Wall Street's pricing model is very ruthless. If the marginal revenue of a company can only contribute very low marginal profit, the faster the revenue grows, the more pressure the valuation may face.

This is a natural extension of the valuation logic.

Cisco's after-hours plunge is essentially the market's structural concern over "revenue growth without profit growth".

Who is taking the real money in the AI network industry?

Cisco's product gross margin dropped by 270 basis points, a figure that reflects the brutal distribution of the AI hardware value chain. In AI clusters, doubling the number of GPUs does not mean doubling the effective computing power, unless these GPUs can communicate without congestion.

As a result, the network content in each AI cluster is rising.

However, the value does not fall evenly on all network devices, but is concentrated in the scarcest and most IP-intensive layers, such as SerDes, signal conditioning, optical engines, and retimers.

These links have extremely high technical barriers, switching costs and bargaining power, thus earning typical economic rent.

Cross-company data demonstrates this point.

Cisco's product gross margin was compressed to 64.8% while its revenue grew by 24%. In contrast, component manufacturers supplying AI cluster interconnection reported completely different economic performance: Credo achieved a non-GAAP gross margin of 68.3% with 157% revenue growth; Astera Labs reached 73.7% when revenue doubled; Lumentum hit 50.4% with a 109% surge in revenue. These companies are far smaller than Cisco, but have higher profit margins, indicating that they are in a more favorable position in the value chain.

Why is this the case? Because the value distribution of AI networks follows the rule of "scarcity determines rent".

SerDes and signal conditioning chips are the core bottleneck of high-speed interconnection. The number of suppliers that can provide stable, low-bit-error-rate signals is limited, the technology iteration speed is fast, and the customer verification cycle is long. Once they are included in the design, it is extremely difficult to replace them. This scarcity gives upstream chip manufacturers strong pricing power.

The same is true for optical engines and retimers, which determine the integrity and energy efficiency of data in long-distance transmission, and customers are far more sensitive to performance than to price.

The system integration link is different.

Assembling chips, optical components, software and chassis into a switch, although the technical threshold of this work is not low, it is more substitutable.

Arista, NVIDIA, Celestica and white-box manufacturers are all doing similar things, and customers can switch between multiple suppliers.

When competition is sufficient, the profit margin of the system integration link is naturally suppressed.

Cisco, as a system integrator, has historically earned premiums through proprietary protocols and enterprise customers.

In the AI backend network market, customers are a small number of hyperscale manufacturers with strong buyer monopoly characteristics; at the same time, the Ethernet white-box trend is obvious, and competitors are everywhere.

Cisco's entry into this market is to exchange profits for market share and scale for ecological positioning.

The question is, why should the market pay for this kind of market share that brings revenue growth but no profit growth? Economic rent is migrating from system integration to IP-intensive links such as SerDes, optical engines, and retimers.

This is a redistribution of power within the value chain, and Cisco is on the weakened side.

Therefore, Cisco's Silicon One strategy is precisely an attempt to extend from a system integrator to an upstream chip vendor, to recapture the squeezed rent. The G300 chip reaches 102.4 Tbps, supports 1.6T networks, and has been deployed by five of the six major hyperscale manufacturers, competing directly with Broadcom's Tomahawk 6 at the same level.

From a technical point of view, the indicators of G300 are sufficiently advanced, and the resources Cisco has invested in chip design are generating returns. But the business logic is more complex. Selling chips and selling systems are two completely different businesses: the former requires an open ecosystem, high R&D investment, low gross margin but high turnover; the latter requires proprietary integration, customer relationships, high gross margin but low turnover.

Cisco doing both at the same time may face customer conflicts and channel conflicts.

Will external chip customers worry about Cisco's competition at the system level?

Will system customers question that Cisco reserves the best silicon for itself?

If these contradictions are not handled properly, Silicon One may turn into a consumption battle where the left hand fights the right hand.

A more realistic constraint is that chip external sales require an independent customer base and ecological support.

When hyperscale manufacturers procure chips, they usually adopt a second-source supplier strategy to reduce the risk of relying on a single supplier.

If Cisco only uses G300 in its own switches, it is an internal supporting chip; if it wants to sell it externally, it must gain the trust of customers of Broadcom, NVIDIA or other independent chip suppliers. Building this trust takes time, and requires Cisco to make commitments completely different from its system business in terms of chip roadmap, technical support, and supply delivery commitments.

The market has not yet seen sufficient evidence of external sales, and the rent recovery story of Silicon One is still in its early stage.

Can Cisco hold the bottom line of rent?

Inventory data has exacerbated market concerns. It is reported that Cisco's inventory reached 5.69 billion U.S. dollars, an increase of about 80% compared with 3.16 billion U.S. dollars in the same period last year, while annual revenue only increased by 12%.

From a financial perspective, inventory can be understood as a kind of real option. If AI orders continue to be converted, stocking up in advance is a call option to avoid the loss of out-of-stock; but if hyperscale manufacturers' capital expenditures pause for digestion for any quarter, inventory will change from an asset to an impairment risk.

What is more subtle is the bullwhip effect.

Small fluctuations in downstream demand will be amplified upstream in the supply chain.

As a system integrator, Cisco's inventory growth rate far exceeds its revenue growth rate, which may mean that channel inventory is accumulating, or the management is too optimistic about order visibility.

It is not unreasonable for the market to regard high inventory growth as another negative signal for growth quality.

In the hardware industry, inventory management is a lifeline. Excessive inventory occupies capital, increases storage costs, and faces the risk of devaluation caused by technology iteration; too low inventory may lead to missed order delivery.

Cisco's inventory growth of 80% far exceeds the 12% revenue growth, and this divergence deserves vigilance.

If AI orders are as strong as the management says, high inventory growth may be explained as stocking up in advance; but if the pace of capital expenditure of hyperscale manufacturers changes, inventory will quickly turn into impairment pressure.

It is a rational choice for the market to remain cautious about this.

In the AI hardware cycle, technology iteration is extremely fast. The shelf life of a switch or chip may only be two to three years, and the inventory holding cost is much higher than that of traditional enterprise network equipment.

Cisco's management needs to provide a clearer inventory destocking path in the next quarter, otherwise the market will continue to reflect this uncertainty with a discount.

In the next four quarters, the indicators that really determine the trend of stock prices need to shift from the total order volume to more sharp dimensions.

The first indicator is whether the incremental gross margin of AI revenue has bottomed out and rebounded. If the growth of AI revenue continues to pull down the overall gross margin, revenue growth will only dilute shareholder value. Only when the gross margin stabilizes can it prove that Cisco has pricing power in AI networks, not just scale. The management needs to give a clear path to explain how the cost structure of AI products will improve and whether the bargaining pressure from hyperscale manufacturers will ease.

The second indicator is the proportion of revenue from externally sold Silicon One chips. This is direct evidence that Cisco is recapturing rent from the upstream. If the proportion of external sales revenue in AI revenue increases significantly and the customer base is diversified, it indicates that Cisco is migrating from an integrator to the IP layer; on the contrary, Silicon One is only internal supporting, which cannot change the pattern of rent outflow. This indicator is more important than chip technical indicators, because it directly measures the effectiveness of the business model transformation.

The third indicator is the sustainability of enterprise network orders after excluding hyperscale AI orders. After excluding concentrated AI orders, product orders grew by 25%, with double-digit growth in all regions. This is the stickiness of Cisco's fundamental business. If the enterprise network replacement cycle continues and the gross margin remains at a high level, Cisco will have a safety cushion to buffer the fluctuations of the AI cycle; if enterprise demand cools down and the proportion of low-margin AI revenue rises, the profit structure will deteriorate at an accelerated pace.

These three indicators correspond to pricing power, business model transformation and the resilience of the fundamental business.

They reveal Cisco's true value better than any single order figure. The market no longer cheers for order numbers, but begins to reprice with the logic of economic rent. In the next four quarters, Cisco needs to prove that it can not only sell more devices, but also retain enough profit from every dollar of AI revenue.

Growth is the beginning of the story, and rent is the end of the story.

What Cisco's financial report reveals is the universal distribution structure in the AI hardware boom: revenue growth is concentrated in the system integration link, but profit growth is concentrated in the upstream scarce IP layer.

Economic rent is migrating from system integration to IP-intensive links such as SerDes, optical engines, and retimers.

Cisco can participate in this value migration through Silicon One, optics and software, or watch economic benefits pass through its own devices and flow to more upstream suppliers.

This choice will determine its final position in the AI era.

The so-called after-hours plunge mentioned earlier is actually just a signal: in the AI wave, scale can bring attention, but only rent can bring value.

This article is from the WeChat official account "Dongzhen Strategic Consulting", author: Dongzhen Strategic Consulting, published with authorization from 36Kr.