After the publication of the previous article, an entrepreneur raised a question that strikes at the core:
"The vast majority of Industrial IoT companies today can still only earn one-time or fixed service fees for hardware, network deployment, and O&M. AI is merely an add-on feature, treated as a bonus point during bidding, or even reduced to a tool for price cuts to secure orders. The cost reductions, loss mitigations, and production increases brought by AI all benefit the customers, leaving vendors without any excess profits. Customers can switch algorithm suppliers at any time, and once hardware and networks are standardized, a price war will soon ensue."
"How exactly can AI capabilities be packaged into quantifiable revenue-sharing rights, tied to real operational metrics on the industrial shop floor, transforming IoT companies from equipment suppliers into co-partners sharing the risks and rewards of customers' production efficiency?"
This question strikes a painful chord with most IoT companies today.
In industrial reality, AI has not only failed to bring a premium but may even be reduced to a bargaining chip for customers to negotiate lower prices.
Because no matter how advanced the technology or how excellent the product, customers are unwilling to pay for the "process." The immense wealth created by industrial upgrading will only be paid to those who can truly "deliver results."
This means that the closer an AIoT company is to the customer's "ultimate business outcomes," the larger a slice of the cake it can get in this industrial chain.
When a company's business model is stuck at the stage of "selling tools," the distance to the actual results is astronomical, and AI might just be a free giveaway. Only when a company dares to and has the capability to underwrite the customer's ultimate returns, becoming the one who "delivers results," will the AIoT system—composed of hardware, networks, AI algorithms, and industry know-how—transform into a "cash flow engine" that continuously creates compound interest for the enterprise.
But here is the problem: since "profit-sharing based on results" sounds so appealing, why is it that across the entire IoT and industrial AI track, everyone is still desperately selling equipment and engaging in price wars? Why is it so easy to talk about "delivering results" but so hard to execute?
To measure exactly how far AIoT companies are from "business outcomes," I reviewed the financial reports of four industry benchmarks: Yiyuan Tongxin, Tuya Zhineng, Zhongkong Jishu, and Hanwei Keji.
How Far Are We from "Delivering Results"?
To clearly see how far the entire industry is from "business outcomes," we need to establish a "two-dimensional coordinate system" to conduct an in-depth diagnosis of current business models:
The vertical axis represents "irreplaceability."
The core measurement metric is the gross profit margin. The higher the gross profit margin, the stronger the company's bargaining power in front of customers and the deeper its moat.
The horizontal axis represents the "distance to results."
The core measurement metric is the revenue recognition method. Is revenue recognized upon shipment, upon system go-live, or continuously based on the "incremental value ultimately created for the customer"? The further to the right, the closer to the customer's ultimate results.
In this coordinate system, the ultimate position for all To-B enterprises is in the "upper right corner": possessing extremely high irreplaceability while continuously sharing profits based on the incremental results created.
With this ruler clarified, we will next place the four most representative companies in the AIoT track (Yiyuan Tongxin, Tuya Zhineng, Zhongkong Jishu, Hanwei Keji) into the coordinate system one by one.
Let's first look at Yiyuan Tongxin, which is in the connectivity layer.
In the first half of 2026, Yiyuan Tongxin's operating revenue reached as high as CNY 15.259 billion. By product line, the gross profit margin for communication modules and antennas was 16.28%, and for automotive, smart, and solutions, it was 21.42%. In terms of revenue recognition, it still mainly ships in standardized product forms.
Next, let's look at Tuya Zhineng, which is in the platform layer.
In the first half of 2026, Tuya's revenue was USD174 million. Of this, the PaaS business accounted for 73.0%, with the product part recognized at once upon customer acceptance; for the cloud connectivity services in PaaS, which are recognized period by period after device activation over the product lifecycle, only USD 579,000 was recognized in the first half, accounting for only 0.33% of total revenue.
However, Tuya has a revenue line recognized over time, coming from industry SaaS and cloud value-added services. In the first half, USD 17.99 million was recognized through deferred revenue, accounting for about 10.4% of total revenue. The semi-annual report stated that subscriptions for scenarios such as AI energy saving, pet care, and video understanding have begun to generate initial paid usage and renewals.
Next is Zhongkong Jishu, which is the one among the four that sees the "business endgame" most thoroughly.
Zhongkong is an enterprise that has written "charging by results" into its corporate strategy. In the 2025 annual report and the 2026 semi-annual report, Zhongkong explicitly proposed: to gradually shift from one-time buyout licenses to a value subscription model of Model as a Service (MaaS) and Result as a Service (RaaS).
Zhongkong admitted in the semi-annual report that the domestic industrial software field is still dominated by traditional licensing. Although its sights are set on the "upper right corner" of the coordinate system, its current contracts are still helplessly signed at the "system commissioning" step.
Finally, there is Hanwei Keji, which is in the perception layer.
Hanwei's smart comprehensive solutions recognize revenue based on the "percentage of completion." It is the most advanced among the four companies in terms of the distance to results. However, financial reports show that Hanwei's attempt to take a step closer to "results" was exchanged for prolonged payment terms and goodwill impairment.
This leads to the biggest pain point: under the old business rules, the closer an enterprise is to the customer's "results," the more dangerous its "cash flow" becomes; the more it wants to dive into the deep water of business, the easier it is to be dragged into the quagmire of payment terms.
Zhongkong and Hanwei attempted breakthrough trials, resulting in prolonged payment terms, while Tuya and Yiyuan, because they only sell standardized tools, actually collect payments faster.
Is this the price that must be paid for "delivering results"?
My judgment is: no, this is the common ailment of "bearing the risks of a partner without getting the rights of a partner." The enterprise bears the heavy responsibility of underwriting the business, but is not matched with the corresponding rule-making power.
Although Zhongkong and Hanwei see the endgame, the underlying color of their business models is still the traditional project-based system. In a business model that truly "delivers results," although payments are also collected in installments, the underlying logic has changed completely. The settlement amount for each period no longer relies on subjective manual acceptance, but is directly determined by objective "metering readings."
Next, let's break down the core secrets mastered by those enterprises that have successfully pulled this off.
Four Core Prerequisites for "Delivering Results"
To sign a "delivering results" profit-sharing contract with customers, technology alone is far from enough; four major business prerequisites must be met simultaneously: controllable, measurable, attributable, and bearable.
Without any one of them, charging by results is just armchair strategy. The pioneers who have achieved results in the industry have, without exception, made extremely arduous breakthroughs in these four dimensions.
1. Controllable: From "Decision Support" to "Closed-Loop Execution," Grasping the Core Initiative of the Business
In this regard, historical industrial giants have already set the example.
Rolls-Royce sells its aero engines on a "power-by-the-hour" basis; Philips Lighting sells "light as a service" at Schiphol Airport. This business model works because the equipment is manufactured or owned by them, and they are responsible for O&M, with everything from tuning to maintenance in their own hands.
Returning to the domestic implementation of industrial AI, there are already people trying in the industrial IoT circle.
Mogu Lianwang's choice is very representative. In my view, they did not touch the yield rate of the customer's core production line, because there are too many variables in the core production line, and the control right is also in the hands of the plant manager. They chose air compressor stations and chiller stations. These public auxiliary workshops have clear boundaries and do not directly interfere with the core process. Most importantly, through IoT and AI control systems, Mogu Lianwang can directly dispatch the start-stop and operating parameters of the equipment, achieving unattended operation and keeping the "steering wheel" in their own hands.
2. Measurable: Establishing Value Anchors and Setting Undisputable Evaluation Benchmarks
Enterprises cannot manage what they cannot measure themselves, let alone pay for what cannot be measured. The value created by AI must be translated into clear, undisputable physical or financial metrics for both parties.
In terms of metering, one must be pragmatic and impartial. Taking Mogu Lianwang's public auxiliary workshop renovation as an example, for process industries with stable output and working conditions, the electricity meter reading is directly used as the standard. Electricity meters have national standards and grid standards, which are the most objective and leave no room for nonsense. For scenarios with variable working conditions and output, "unit energy consumption" is used for metering. For example, historically, 0.3 kWh of electricity was consumed per cubic meter of compressed air. After AI renovation, it dropped to 0.25 kWh, making the energy savings clear at a glance.
3. Attributable: Clarifying Contribution Boundaries and Achieving the Leap from "Sharing Increments" to "Selling Output"
Suppose your AI helped the factory save 20% on electricity bills in a single month. When it's time to share the money at the end of the month, the plant manager says, "The reason we saved electricity this month is because of fewer recent orders and a low operating rate. What does this have to do with your AI?" The increment is calculated, but who gets the credit?
Among Mogu Lianwang's customers, about 30% adopt profit-sharing based on energy-saving benefits, and about 70% adopt buyout plus annual O&M fees. The profit-sharing contracts come in three types: Energy Management Contract (EMC) and energy hosting (BOT/BOO) models.
Shen Guohui, the founder of Mogu Lianwang, told me that EMC calculates energy savings based on historical baselines. Party B tends to make the baseline larger, while Party A tends to make it smaller, creating a zero-sum game; after two years, they might even dig out the historical baseline to re-verify it. By then, the working conditions have long changed, making it easy to argue. Customers' financial departments also prefer buyouts, as licenses, hardware, and annual O&M fees are fixed numbers, whereas the shared amount fluctuates with energy savings, making the financial department less reassured.
How to break the deadlock? Shen Guohui's judgment is that heavy-asset energy hosting (BOT and BOO) will become the mainstream next, while EMC will instead be gradually weakened.
This leap seems to make assets heavier, but in reality, it uses the reconstruction of the transaction structure to greatly reduce the trust cost between Party A and Party B.
It turns "sharing increments" into "selling output." The object of metering changes from a historical baseline that will be repeatedly debated to a flow meter reading visible to both parties. When the technology company directly takes over the business results, the attribution problem only needs to be negotiated clearly once—agreeing on the unit price of energy after energy saving (such as the unit price per cubic meter of compressed air).
4. Bearable: Utilizing Financial and Insurance Leverage to Resolve Heavy-Asset and Probabilistic Risks
The "bearable" aspect here is divided into two layers, and the pioneers in the industry have given their respective responses:
The first layer is the bearability of "technical probabilistic risk." AI judgments naturally carry probabilities; a single missed report could cause the customer's critical equipment to shut down. Augury, an equipment predictive maintenance company founded by an Israeli team, has launched a diagnostic commitment backed by insurance: if its AI misses a fault and the equipment is damaged as a result, Augury will compensate the customer as agreed, up to USD 100,000 per equipment. Augury's own liability for this compensation is underwritten by HSB, a subsidiary of Munich Re.
The second layer is the bearability of "heavy-asset investment." Moving towards BOT/BOO energy hosting means huge upfront capital occupation. Mogu Lianwang's solution is to introduce financial leverage. At the end of 2023, they, together with Zhongguancun Keji Zulin under Beijing Guozi, promoted Guangdong Mogu Zhongnuo Shuzi Nengyuan Yunying Youxian Gongsi, specializing in energy hosting. Zhongguancun Keji Zulin and Zhongnuo Jijin made strategic investments in this company. Shen Guohui said that over the past two years, the promotion has been generally positive, although they have encountered difficulties such as owners' misunderstanding and disagreements on baseline verification. The introduction of financial leasing has alleviated the financial pressure under the heavy-asset model.
Conclusion
Understanding the "four prerequisites," the next step is the business reconstruction that touches the essence most. How to define the energy-saving baseline, how to calculate the settlement unit price, who underwrites the default risk, and where does the leveraged funding come from? This is by no means a purely technical issue, but requires bringing business, finance, and capital parties to the same table for underlying mechanism design. I have always looked forward to resonating with the pioneers determined to break the deadlock and jointly deduce this complex strategic topic.
However, the business reality that must be faced is that when an enterprise completes the journey from "selling tools" to "delivering results," it will find that it has become a completely different species on its financial statements:
The proportion of revenue recognized over time will significantly increase, while revenue recognized at once upon shipment or acceptance will correspondingly decrease;
Equipment is no longer sold to customers but remains on its own balance sheet, and assets will become heavier accordingly;
Upfront investments need to be slowly recovered through years of operation, and the book profit margin in the first few years may even drop instead of rising;
The enterprise is using heavy assets and a long time to exchange for stable cash flow in the next decade.
In the AI era, hardware can be replaced, data can be exported, algorithms can be retrained, and the generalization ability of large models is getting stronger and stronger. So, what exactly can build a long-term moat? It is the underlying transaction structure that deeply binds business results with commercial settlements.
The arrival of AI is accelerating the answering of this multiple-choice business question.
Shen Guohui, founder of Mogu Lianwang, has observed an important trend. After the popularization of AI, customers are increasingly inclined to cooperate using BOT and BOO. Because in the past, energy saving mainly relied on replacing hardware, and customers could just find an EPC (Engineering, Procurement, and Construction) unit to do it themselves, keeping all the benefits; but with AI, the threshold is higher. It is not cost-effective for customers to build their own teams to do physical AI, and instead, their acceptance of "paying by results" is increasing.