Author: Peng Zhao (Founder of Zhicifang, Co-Founding Partner of Yunhe Capital) Original
In the first three quarters of 2026, we have witnessed AI (Artificial Intelligence) integrating into the physical world at a breakneck pace. Whether it is embodied AI in factories or autonomous driving in cities, for large models to perceive reality and execute actions, every bit of data and control flow they ingest and emit must pass through the modules and sensors of IoT (Internet of Things) companies.
However, when I opened the financial statements of these IoT companies, a harsh reality surfaced: In the AI era, the substantial premiums and profits have not stayed on the books of IoT companies.
In today's business landscape, a value "scissors gap" has taken shape: Connectivity at the bottom layer is irreversibly depreciating, while intelligence at the top layer is appreciating without limits. Caught in between, IoT companies find themselves in the awkward position of missing out on value, struggling to share in the dividends of the intelligence leap.
This scene is all too familiar. It closely mirrors the fate of telecom operators in the mobile internet era: they spent massive amounts of capital and overcame countless obstacles to build the underlying high-speed information networks, only for OTT companies like WeChat and Douyin, which grew on top of these pipes, to reap the massive profits.
Today, the metaphor of history seems to be repeating itself. In the wave of AI sweeping across everything, if IoT companies fail to find a new positioning in time, the greatest survival crisis they face could be an irreversible collective fate of being reduced to "dumb pipes."
So, is there a solution to this seemingly unsolvable dead end?
Five Paths to Self-Rescue and Three Cold "Iron Laws of Value"
We first need to see through a harsh business truth: in the AI era, the value-added profits in the industrial chain will most likely flow to places occupying the following three ecological niches.
The first is scarcity: possessing irreplaceable technological or resource barriers;
The second is the lock-in effect: building extremely high customer migration costs and replacement thresholds;
The third is a closed business loop: holding the decision-making power and value distribution rights over the final delivered results.
Measured against these three iron laws, today's IoT companies awkwardly find that they do not occupy advantageous terrain:
Connectivity lacks scarcity and lock-in: Connectivity has been reduced to a highly standardized commodity, which can be replaced at low cost at any time;
Data is handled but not operated: IoT companies are merely "porters" of data rather than "holders";
The business closed loop is not in their hands: the vast majority of IoT hardware is only responsible for perception, not for decision-making and execution.
Faced with this systematic value stripping, the industry is not unaware. Over the past two years, countless practitioners have explored in anxiety and gradually summarized five "paths to self-rescue":
Path 1: Connectivity moves from "coverage" to "determinism." This is the self-rescue of the connectivity layer. Broad-coverage connectivity will continue to depreciate, but highly deterministic connectivity can be repriced. Examples include the ultra-low latency and high-reliability guarantees of 5G-A, the cost-performance tiers of RedCap, and "integrated sensing and communication" that turns the network itself into a sensor.
Path 2: Embedding intelligence into connectivity (Edge AI). When AI models run on the edge, the value physically lands in modules and gateways. Today, the price difference between smart modules with NPUs and ordinary Cat.1 modules is empirical evidence of AI value flowing into IoT companies.
Path 3: From "transmitting data" to "holding data." The "Data Twenty Articles" splits data rights into holding rights, processing and usage rights, and product operation rights. IoT companies are closest to the data source and are in a position to obtain "processing and usage rights" through edge-side cleaning and processing, and "operation rights" by providing high-quality datasets.
Path 4: From "selling equipment" to "selling results." Charging by the number of devices means the value added by AI has nothing to do with IoT companies; but if charging by results (such as energy-saving revenue sharing, reducing downtime, and improving yield rates), the value added by AI automatically flows into IoT contracts.
Path 5: From components to entities, from products to standards. Vertical integration is the "hard path," such as sensor companies directly making dexterous hands, or module companies stepping in to make complete machines; acting as a hub horizontally is the "clever path," such as the Matter protocol for the CSA (Connectivity Standards Alliance).
These five paths may look like they all lead to Rome, but the reality is quite harsh.
When we put these five paths back into the "three iron laws of value" mentioned earlier for testing, a business law surfaces: The difficulty of transformation is directly proportional to the ability to share profits; the easier the path, the harder it is to get a share of the profits.
Why does this reversal occur? To understand this, we must see through the underlying logic of wealth flow in the AI era. All value added by AI is essentially "result-level" value addition. One less hour of downtime in a factory, a one percentage point increase in production line yield, one less kilowatt-hour of power consumed by equipment... these incremental profits created by AI will only flow into contracts that are "signed based on results."
And what are most IoT companies selling today? They are selling "technical specifications" and "equipment parameters." How much latency, how high the precision, how low the power consumption. When responsibility stops at "specification delivery" or "equipment acceptance," revenue is locked into the pattern of "hardware BOM cost + meager manufacturing profit."
The first two paths (deterministic connectivity, edge AI) are the easiest to take, but they are merely upgrades in technical specifications. The premium brought by specification upgrades will soon be wiped out in hyper-competition.
The latter three paths (holding data, selling results, setting standards/entities) can indeed pass the test of the iron laws, but the cost is extremely high: it requires an IoT manufacturing company to undergo a complete "metamorphosis" and become a company with a completely different DNA.
Therefore, this article may not provide an easy "self-rescue checklist," because there is no smooth road ahead. We can only propose an analytical business mechanism and verify it with the real financial statements of four listed companies.
The four companies I have selected are Hanwei Technology, Quectel, Tuya Smart, and Supcon Technology:
1. Hanwei Technology (Representative of the Perception Layer): As a leading domestic sensor company, it represents the front line of physical world data collection;
2. Quectel (Representative of the Communication Layer): As a global giant in cellular IoT modules, it is synonymous with "connectivity";
3. Tuya Smart (Representative of the Platform Layer): As a global leading IoT PaaS platform, it is the hub for devices going to the cloud;
4. Supcon Technology (Representative of the Application Layer): As a leader in industrial automation control, it is deeply rooted in vertical industries.
Three Costs to Cross the Chasm: Enduring with Time, Exchanging with Capital, or Even Surrendering Control
Since moving from "selling specifications" to "selling results" is a thorough genetic reorganization, companies must pay a high price to complete this transformation. Observing the current practices of IoT companies, we have found three "hard currencies" for paying this price in the market: time, capital, and even the company itself.
The first cost is "time," which means relying on endogenous strength and doing it personally.
Quectel disclosed in its semi-annual report its self-developed wheeled-arm embodied robot whole-machine platform and lawn-mowing robot solutions; Supcon Technology listed quadruped, wheeled, and rail-mounted inspection robots, as well as standard humanoid robot solutions for education and exhibition halls; Hanwei Technology, through its controlling subsidiary Suzhou Nanosensor, has laid out flexible tactile sensors and currently formed an embodied AI product matrix covering touch, smell, balance, force control, and vision.
This path is the slowest and most arduous, but its advantage lies in steady and sustainable progress, without burying high goodwill on the balance sheet.
The second cost is "cash and equity," which means buying directly through external mergers and acquisitions.
Quectel invested CNY 60 million to acquire an 11.54% stake in AndarTech, a design company focusing on CMOS process mmWave radar chips, which can cut into the intelligent driving and robot perception markets. This is the "lightweight version" of the capital path, buying only upside returns without taking on integration risks. The "heavyweight version" is Hanwei Technology. Over the years, it has acquired more than a dozen companies, including Jiayuan Environmental Protection, Shenyang Jinjian, and Intersense, expanding its business map from sensors all the way to environmental protection engineering, urban software, and heating operations.
The third cost is the most counterintuitive: "handing over the business itself," that is, taking the initiative to be acquired.
Take a recent overseas case: Identiv has been deeply engaged in the RFID tag field for many years. In fiscal year 2025, the revenue from this business reached USD 21.5 million, but the gross profit margin was only 6.1%. In September this year, Trackonomy announced the completion of the acquisition of Identiv's IoT tag business. The merged Trackonomy integrates three major segments: semiconductors, smart tags, and end-to-end solutions.
Putting these three "payment methods" under scrutiny, we will find that these companies share an extremely clear consensus: They spare no expense, all for the purpose of seizing the "core position" in the ecological niche.
Whether it is Quectel taking a stake in underlying radar chips, or Supcon and Hanwei personally stepping in to make robots and tactile sensors, none of them are blindly following the trend of acquiring large model or pure embodied AI software companies.
However, buying the core position does not mean everything is fine. There are two hidden limiting conditions here:
First, buying an asset position does not equal reshaping the business contract.
Business contracts are divided into two types: one is being responsible for "stock results," which belongs to traditional operations, such as ensuring heating supply and network connectivity; the other is being responsible for "incremental results," which belongs to value sharing, such as how much less coal is burned through AI and how much the downtime rate is reduced. Most of the excess profits in the AI era exist in the second type of contract.
Looking back at the numerous positions Hanwei has acquired, the one closest to the result is heating operations. It indeed bears the responsibility of delivering the final result, but the problem is that heating prices are government-led, and the contract is locked in the "stock." Bearing responsibility but lacking pricing power led to a gross profit margin of only 10.85% for this business in 2025. Ultimately, in October 2025, Hanwei announced its intention to sell a 65% equity stake in its heating business subsidiary, Hanwei Zhiyuan, to the state-owned asset platform of Zhengzhou High-tech Zone for CNY 440 million, citing the reason as "coordinating with Zhengzhou's city-wide heating network and the integration of heating in the central urban area."
It is evident that a position is an asset, which can be bought with capital; but a contract is a relationship, which must be reconstructed with strength.
Second, regarding the role positioning in mergers and acquisitions and being acquired.
The Identiv transaction mentioned earlier leads to a counterintuitive inference. In this transaction, the acquirer is a Physical AI company, and the acquiree is a traditional hardware company. The essence of this transaction is that the party with a higher valuation (the AI company) used its own equity to absorb the asset with a lower valuation (the hardware business). For the seller, Identiv, it successfully obtained a "call option" on the future growth of the AI company, while shaking off the quagmire of low-margin hardware.
But what if it is done the other way around? If an IoT company uses its own stock, valued under "hardware logic," to acquire assets valued under "AI logic" at a high premium, then every transaction made will dilute the equity of its own shareholders.
Therefore, for IoT companies, being "acquired" at the right time may be a better solution to maximize value.
This inference will be the key clue for us to deduce the endgame of the industry.
Final Thoughts
In this winner-takes-all era, IoT companies stand at a critical crossroads. Stepping forward requires a thorough, life-or-death genetic reorganization; stepping back may lead to the irreversible fate of being reduced to "dumb pipes in the intelligent era."
Whether enduring with time, exchanging with capital, or simply handing over control, they are all paying for a ticket to "value distribution."
At this juncture in 2026, when AI is accelerating its implementation, among the leading IoT companies at the eye of the storm, who can truly touch the threshold of excess profits in the AI era? We will continue to dissect this in the next article.