White Paper on the Market Development of Power and Environment Monitoring Systems

Contents

01Research scope: why this industry has no market figure that can be trusted

The hardest step in writing a white paper is not finding figures. It is deleting the figures that cannot be used.

1.1 Fix the boundary first: what power and environment monitoring actually refers to

“Power and environment monitoring” is used in Chinese for at least four different scopes. That is the root of all the statistical confusion:

This white paper discusses the first three layers, with the emphasis on the physical deliverable that is “the on-site acquisition and monitoring unit.” Building automation and integrated security are not counted in. Once the scope is widened to the fourth layer, market size can easily triple, and it has nothing to do with the people who make the hardware.

1.2 The market-size figures in circulation differ from each other by a factor of ten

We searched the summaries of multiple “China PEM / DCIM market” reports on sale in 2026. Market size for the same country and the same year is as follows:

The highest value and the lowest differ by 10.8 times. What deserves more caution is this: two of them come from the same institution and, for the same year, differ by 2.3 times. That shows the problem is not “who researched it more accurately.” It is that no one has written the scope down clearly — whether software is included, whether integration services are included, whether construction is included, whether overseas is counted, and whether building automation is counted in. Any one of these can move the figure by more than 30%.

1.3 Data grades in this white paper

If the reader remembers only one thing, remember this: in this industry, the only figures that cannot be easily overturned are the official base quantities. How many EFLOPS of intelligent computing power, how many kilowatts of energy-storage capacity, how many China Tower sites, what the PUE ceiling has been pressed down to — these are hard. A market of “so many billions” is soft.

02Demand side: five main lines diverged over the past two years

One industry name, five entirely different curves

2.1 Computing side: from “building computer rooms” to “building intelligent computing,” the object of monitoring has been rewritten

This is the line that changed the most over the past two years. Three official figures are enough to show it:

• As of June 2026, national intelligent-computing scale was 2185 EFLOPS (FP16), +177% year on year, with 595 EFLOPS added in the first half (MIIT).
• In 2025, China had completed 42 intelligent-computing clusters of the 10,000-card class, and computing centers consumed 170 billion kWh in total (National Energy Administration, China “AI+” Energy Development Report 2026). In July 2026, the first fully domestic 100,000-card-class cluster was completed in Zhengzhou.
• The 15th Five-Year Plan for the Development of the Information and Communications Industry sets the 2030 intelligent-computing target at 9800 EFLOPS, an increase of more than five times from 1590 EFLOPS in 2025. It also requires the PUE of new large and ultra-large computing facilities to be pressed below 1.2, one notch tighter than the 1.25 at the end of 2025.

For people who do PEM, what these three statements mean in plain terms is this: the object of monitoring has changed from “a room” to “a cabinet,” and even to “a chip.” A 200 m² computer room used to be covered by 20 temperature and humidity points. A 130 kW liquid-cooled cabinet now needs a full set: multi-point temperature, flow, pressure, leakage, water quality, and CDU status. Point density has gone up, and so has the value of a single project — but the technical requirements are an entirely different set.

The supporting standards are being filled in as well. GB 40879, Limits and Grades of Energy Efficiency for Data Centers, and GB/T 44989, Evaluation of Green Data Centers, have already become the basis for acceptance. In July 2026, GB/T 48023—2026, Technical Specification for Cold-Plate Liquid-Cooling Systems in Data Centers, was issued, and liquid-cooling systems moved from “each vendor’s own interface” to “a standard to follow.” Over the same period, MIIT and four other departments ran the 2026 national green computing-facility recommendation, with explicit requirements of ≥3000 standard racks, a rack-loading rate ≥60%, PUE at GB 40879 Grade 2 or above, and a green-power share not below the renewable-consumption responsibility weight of the province where the facility sits.

2.2 Power side: “unattended operation” for 3 million distribution rooms, from pilots into volume

The largest change on this line over the past two years is that its nature changed — from “demonstration project” to “acceptance clause.”

The public data of the Foshan Power Supply Bureau of China Southern Power Grid are the most complete sample available: 4821 intelligent distribution rooms (including pole-mounted units) were built and fully connected to the intelligent distribution-network monitoring system, cutting inspections by 12264 visits, cutting inspection labor by 24075 hours, cutting maintenance outages by 90%, and raising maintenance efficiency by 60%. These figures persuade because they state “how many fewer trips, how much less outage,” not “how many systems were installed.”

How large is the installed stock: about 3 million sites distribution rooms above designated size nationwide, including high-voltage customer substations (industry scope, Grade B). After DL/T 1649, T/CEC 5022, and related standards came into force one after another, equipment-status thresholds, warning grades, platform architecture, and the diagnostic closed loop all have explicit rules. The situation in which every scheme went its own way has started to converge.

Typical retrofit content (the State Grid Qingyang Dongli switching station as the example): temperature and humidity, water leakage, SF6 gas, UHF partial discharge, online battery monitoring, intelligent access control, and AI cameras, on a “cloud–network–edge–endpoint” architecture, with four modules for environment / equipment / security / video. Alarms are pushed synchronously to the power-supply service command platform, and interlocked with air conditioning, fans, and lighting. This is a complete station auxiliary-control system, not four sensors.

2.3 Telecom side: a replacement market in the installed base, but one centralized procurement can decide survival

Telecom used to be the largest PEM market. It is now “the largest installed base + the hardest centralized procurement.”

As of June 2026, China Tower’s site count exceeded 2.17 million, with nearly one million equipment rooms. Cumulative base stations it has built exceed 6.2 million (5G exceeding 3.28 million). 260,000–270,000 “communications towers” have already been upgraded to “digital towers,” loaded with cameras, radar, and sensors.

An installed base means a steady replacement demand for FSUs (field supervision units), but the allocation mechanism of this market is centralized procurement. The most direct evidence is Yaoao Technology: because it did not win China Mobile’s centralized procurement of power and environment monitoring for 2024–2026, PEM monitoring-system revenue in the first half of 2026 was RMB 21.5037 million, down 46.12% year on year; net profit attributable to the parent was RMB -8.7 million, a year-on-year decline of 6578%; gross margin fell from 19.51% to 16.29% (source: the company’s 2026 interim report).

2.4 Energy-storage side: the fastest growth, but the threshold has changed from “able to monitor” to “able to prove safety”

By the end of June 2026, new-type energy storage built and in operation nationwide reached 153 million kW / 396 million kWh, up 61% year on year. Energy-storage battery output in the first half of 2026 exceeded 420 GWh, +110% year on year (MIIT). The 15th Five-Year Plan for Building a New-Type Power System proposes about 160 million kW of new-type energy storage added during the 15th Five-Year Plan period, reaching 300 million kW in 2030.

For a PEM vendor, an energy-storage station is a new form of “computer room”: cabin temperature and humidity, combustible gas (H2/CO), smoke, fire-protection interlock, air-conditioning / liquid-cooling status, access control, and video, plus voltage / temperature / internal resistance at battery-cluster level. The difference is this — the first demand in this scenario is not “monitoring.” It is “safety compliance.” Industry conferences have listed “safety” as the first problem for years running. The standards system is extending from cell-level safety to system architecture, intelligent O&M, and evaluation specifications.

That means the entry threshold is not “whether you can collect the data.” It is “whether your detection scheme can pass fire-protection acceptance, whether it can interlock with the fire-alarm control panel, and whether the corresponding test reports exist.”

2.5 The other side: the long tail is getting thicker

Another change over the past two years that is easy to miss is fragmented growth of scenarios: archive storerooms, medical logistics (drug stores, blood banks, operating rooms, PCR laboratories), rail-transit depots, underground utility tunnels, charging stations, grain storage, and museum collection stores. The common features of these scenarios are: small per site, detailed in requirements, the owner does not understand the technology, and extremely dependent on the integrator.

They will not appear in the top three lines of any “market size” report, but they are the real granary of small and mid-sized vendors — because a large vendor will not do non-standard protocol integration for a single RMB 50,000 order.

03Supply side: two years of volume up and profit down

Use financial reports in place of opinions

3.1 The closest comparable sample: data-center temperature-control and environment vendors

Pure-play PEM monitoring companies are mostly unlisted or small. The most complete public financials belong to the “data-center temperature control and precision environment” group. They share the same customers as PEM (computer rooms, data centers, carriers) and the same sales channels (integrators, general contractors). They are the best observation window at this stage.

Put these four rows together and the shape is very consistent: revenue is rising, profit is falling, and cash flow is bleeding.

• Gross margin fell across the board by 2.5–4.8 percentage points. In hardware manufacturing, a drop of this size means the price war has already cut to the bone. “Competition has intensified” does not cover it.
• The fall in profit is far larger than the fall in revenue. Shenling’s revenue was still +4.39%, and net profit was -64.07%. Fixed costs (capacity, headcount, R&D) sit there, and the moment revenue slows they crush profit.
• Operating cash flow was negative across the group. Yimikang was RMB -160 million, Shenling RMB -554 million. The companies themselves stated the reason clearly: higher inventory + a mismatch in upstream and downstream payment terms. The cash-flow risk of a project business has been amplified to the extreme in these two years.
• The only profit growth that looks good has to be discounted. Yimikang’s net profit was +40.74%, but that includes deferred income-tax adjustments and government subsidies, while cash flow flowed out sharply. The growth is real. Whether it is healthy is the problem.

3.2 Why volume rose and profit fell: three structural reasons

3.3 One lost centralized procurement equals three years of bleeding

The Yaoao Technology case deserves to be looked at on its own, because it exposes the most brutal mechanism in this industry. Procurement of telecom PEM is “winner-take-all + cycle lock-in”: the winner receives steady batch orders for a 2–3 year centralized-procurement cycle, and the loser’s demand goes to zero for the same length of time.

Under this mechanism, a company’s performance curve does not “fluctuate with the industry.” It “jumps with the bid result.” The secondary consequence is this: winners are forced to trade a low price for share (gross margin is locked), and losers are forced to an even lower price to grab the fringe market (which pushes the whole industry’s prices down further). The price war is not a result of competition. It is a product of the procurement mechanism.

04Price and channel: unit-price collapse and the reverse selection of channels

Only by splitting “how much a set costs” can you see who takes the money

4.1 The price band: from “selling a system” back down to “selling a host”

Below are the ranges from public channels in August–October 2026 (B2B platform list prices and service providers’ public quotes), for range reference only:

A structural problem sits inside the price structure: sensors and collectors are about 40% of total cost, servers and software about 30%, and installation and commissioning about 20–30%. The gross-margin direction of the three blocks is completely opposite — sensors get thinner and thinner (semiconductor and copper price swings pass straight through), software has a high nominal margin but cannot be sold at a price, and installation and commissioning are pure labor: the farther the site, the greater the loss.

The conclusion is blunt. If PEM is treated as “selling a box,” it is already a hardware business with gross margin around 20%. Only by making “protocol integration + on-site joint commissioning + delivery certainty” part of the product is it possible to pull gross margin back above 35%. That is also why the same system is quoted at RMB 20,000 by one vendor and RMB 80,000 by another. The gap is not in the box. It is in who comes to site when something goes wrong.

4.2 Reverse selection in the channel: integrators have started choosing vendors

It used to be vendors choosing channels (the agency went to whoever could hit the quota). In the past two years it has reversed. There are three reasons:

• Integrators are having orders taken by vendors. Vendors sell direct to the end owner, and the integrator is bypassed after finishing the scheme. This conflict broke out in concentration in 2024–2025.
• Integrators need qualification backing. A Xinchuang project requires adaptation certificates, test reports, and software copyrights. A trader-type vendor cannot produce them, and the integrator is blocked at bid.
• Integrators need a technical backstop. On-site commissioning staff are short, and when something goes wrong someone must take it remotely. If the vendor does not, the entire cost lands on the integrator.

A new form of supply has therefore appeared: source vendors who declare that they “do channels only and do not deal with end owners,” trading regional protection and project registration for integrator loyalty. This is a signal — the focus of competition in the industry has moved from “whose product is better” to “whose channel policy someone dares to bet on.”

05Technology: four lines that are certain, and three lines that are overrated

Telling real demand from what looks good on an exhibition stand

5.1 Four technology lines that will certainly happen

5.2 Three technology lines that are overrated

5.3 An underestimated technical point: alarm quality

On the sites we have seen, the most common problem is not “no monitoring.” It is “too many alarms, and no one looks.” A temperature and humidity sensor mounted directly under an air-conditioner outlet, a leak-detection rope laid under a condensate pipe, a battery internal-resistance threshold set too aggressively — these three installation and parameter errors can produce ten times as many false alarms as real faults.

The solution to this class of problem is not in the algorithm. It is in the delivery specification: point-siting rules, default thresholds, alarm grades (urgent / important / advisory), and suppression and deduplication policy. Done well, this layer can give a project more real operational value than installing twice the sensors. And this is precisely the part hardest to copy, because it is written in the engineering manual, not on the product brochure.

06Market layers: three markets, three completely different sets of rules

PEM is not one market. It is three different businesses sharing one name.

The most common mistake in the industry is to plan PEM as one market — the same product, the same pitch, the same price list, aimed at every customer. In fact, the three customer types — carriers, bank and airport data centers, and small and mid-sized computer rooms — have almost nothing in common on the three dimensions of who pays, how they pay, and who carries the integration risk.

6.1 A side-by-side comparison of the three markets

6.2 Type one: the carrier market — the list decides survival

This is the largest piece, and the one with the hardest rules. As of June 2026, China Tower had more than 2.17 million sites and nearly one million equipment rooms. Add the three major carriers’ own core buildings and transmission nodes, and this is an installed-base market counted in the millions of sites, with a round of equipment replacement required every 5–8 years.

But its allocation mechanism is not “competition.” It is “allocation.” A group centralized procurement locks for 2–3 years. The winner takes most of the share inside the cycle, and the loser’s demand goes straight to zero. The Yaoao Technology example (Section 03) has already quantified the cost: PEM revenue -46.12%, net profit attributable to the parent -6578%.

The participants are mainly specialist vendors that have long been on the list. Industry material shows Liwei Zhilian among the leaders in base-station FSU share. Yaoao Technology is one of the important FSU suppliers to China Tower and the three major carriers. The rest are a number of regional FSU plants. The channel list price is about RMB 8,800 per unit (Grade C). The actual centralized-procurement transaction price is usually significantly below the list price (Grade D estimate, to be checked against the actual bid documents).

6.3 Type two: the high-end industry market — a case is the entry permit

At the tier of banks, airports, rail transit, government, and power headquarters, what is being bought is not “PEM.” It is “auditable operation of critical infrastructure.” Demand has already been upgraded from monitoring to DCIM: asset and capacity management, change process, energy-efficiency PUE, audit trail, and Xinchuang compliance. Not one of them can be missing.

The threshold at this tier is compound:

• Qualifications: SRDI, security design and construction qualifications, MLPS assessment, and proof of process capability.
• Cases: tender documents often state outright “no fewer than X comparable projects in the past three years.” Without them, the bid is rejected.
• Scale: registered capital, size of the implementation team, nationwide service outlets. Missing one item is judged “insufficient capacity to perform.”
• Capital: retention is held for 1–2 years. A RMB 5 million project requires advancing RMB 250,000–500,000 as a deposit, and the implementation cost has to be advanced as well.

The two companies that live best at this tier happen to represent two different moats: Gongji Technology by “a platform that has served only one customer group for thirty years,” and Jitong Intelligent by “industry entry held up by national-level reference projects” (see Section 07).

6.4 Type three: the small and mid-sized computer-room market — the most fragmented, the most tiring, and the one that still listens to reason

This tier has no authoritative statistics, because the projects are too fragmented and the amounts too small to enter any industry report. It is also the tier with the largest count: enterprise-built computer rooms, equipment rooms in schools and hospitals, ELV rooms in chain stores, archive storerooms, and edge nodes on campuses.

Three features of it decide that it is the only viable path for small and mid-sized vendors:

Gross margin is highest at this tier (35–50%), because service and installation are included. But the amount per project is the smallest, and the selling-expense ratio is the highest. It is a business that has to be run out by people, not a business that rolls forward on brand.

6.5 A dividing line that is easy to miss

07Market performance of six vendors: the real ecology, from public data

Not the brochure. Only operating data and product moves that can be checked.

The six below cover three markets, three scales, and three ways of staying alive. The standard for choosing them is not “the best known.” It is that verifiable public information can be obtained (official-site disclosure, periodic reports, business-registry and bidding data). Where there are financial figures, the figures speak. Where there are not, product moves and project structure speak.

7.1 Shenzhen Gongji Technology: the clear leader in headquarters-level financial DCIM

How to read it: this company proves that “platform + the head of an industry” can reach a 60% share, at the cost of doing one thing for thirty years — every product move is organized around “the same customer, the same computer room,” and has never left the radius of data-center infrastructure management.

What this means for Tongwei: this path cannot be copied (it takes an R&D team of 150 and thirty years), but it draws a line — a place on the platform layer is occupied by time, not by features. We should not compete with it here. We should become a supply unit inside the projects of this company and of companies of its kind.

7.2 Shenzhen Jitong Intelligent: building the threshold out of national-level projects

How to read it: note the piece it was filling in during 2026 — the intelligent-monitoring edge gateway, officially described as “solving inconsistent interfaces and fragmented protocols on legacy equipment.” A company with 14000 cases, able to enter Daxing Airport and the Hong Kong–Zhuhai–Macao Bridge, still has to build a separate product for “protocol fragmentation.” That shows the matter has not been solved in the industry. It has not “ceased to be a problem.”

What this means for Tongwei: two points. First, cases are not for the customer to look at. They are for the tender evaluation panel — the same capability, with or without “a project that can be written into the track-record table,” is a full notch apart in bargaining power. Second, the giants are also troubled by protocols. That means our accumulation on “awkward protocols” has value. It is not self-consolation.

7.3 Shenzhen Mgrid (mgrid): the product school that turned small and mid-sized computer rooms into “models”

How to read it: this is the most productized of the six. The method is not “one scheme per project.” It is “four models covering four kinds of computer room”: small and mid-sized rooms, embedded acquisition, networked centralization, and micro-modules. When a customer selects, they are “choosing a model,” not “negotiating a scheme.”

What this means for Tongwei: the one most worth benchmarking, because it is fighting us for the same market — small and mid-sized computer rooms, the same price band. The answer it gives is: make delivery into a selection table, not into a scheme document. This is also the factual basis for “Trend Five” in Section 10: someone has already made it work.

7.4 Xiamen Shangwei (838070): a small company in transition, with its household accounts written in the announcements

This is the only PEM name among the six with a complete public financial report, so it is the one most worth reading closely. An NEEQ base-tier company, founded in 2004, 2025 revenue RMB 19.0679 million, about 40 employees.

More important is the reason it wrote in its own announcement: “revenue from the traditional PEM business fell by a large margin compared with the same period of the prior year,” and it has issued an announcement that “unrecovered losses exceed one-third of total paid-in capital.” Its response is a full turn toward digital and intelligent power O&M (distribution-station rooms, substation communications rooms). It led the drafting of the group standard Functional Specification for Intelligent Monitoring Systems of Distribution Rooms. Its products have passed type testing by the State Grid automation-equipment EMC laboratory, safety and quality testing by the Ministry of Public Security, and military security technology testing.

How to read it: this company lays out, in full, the real situation of a small or mid-sized PEM vendor —

• Gross margin can reach 66%, and net profit is negative. A “technology-type small company” does not lack pricing power. What it lacks is scale to spread fixed costs.
• Net assets are only RMB 320,000, and the asset-liability ratio is 97%. This is a sword overhead: one delayed collection, one project loss, and liabilities can exceed assets. Profit and cash are two different things.
• The direction of the transition is “change the customer,” not “change the product.” It moves from computer rooms to distribution-station rooms. What it sells is still acquisition + platform + sensors. The customer changes from the IT manager to the grid system. This is the least-effort form of transition, and also one of the kinds that requires the most patience (type testing has to be passed, and a group standard has to be written).

What this means for Tongwei: the path it is walking — keep the technical accumulation, change only the customer group, and trade standards and type testing for entry — is almost the only upgrade path open to Tongwei. It is also a warning: R&D spending during the transition will eat all of the profit, and a company with thin net assets will not hold out until the transition is finished.

7.5 Shenzhen Jinpengzheng: the limiting form of a 6-person company

How to read it: this is the most numerous kind of company in the industry, and the least mentioned — a decent technical base (42 software copyrights, a self-developed platform), an extremely small scale (6 people), and a revenue structure that is mainly maintenance of the installed base. That it can stay alive shows maintenance is a real business. That it cannot grow shows maintenance alone produces no growth.

What this means for Tongwei: two conclusions that can be used directly. First, maintenance is a cash-flow floor — annual maintenance of existing customers takes almost no selling expense, and Tongwei’s existing installed-base customers should be screened at once. Second, maintenance cannot be the whole — six people is the ceiling of a maintenance model, because one person’s time can be sold only once.

7.6 Jianda Renke (Shandong Renke Measurement and Control): the scale champion of the component layer

How to read it: it has the highest revenue of the six (RMB 162 million), but what it sells is not a system. It is sensors. It proves that, in this industry, “components + e-commerce channels + scaled manufacturing” can produce a company at the hundred-million-yuan level. It also proves the ceiling of that path: it cannot do integration, because integration does not scale. Its customers are precisely the integrators and engineering contractors.

What this means for Tongwei: the most important point — it is our supplier, not our competitor. The sensors, gateways, and monitoring hosts it sells are exactly the components inside our schemes. And our customers are exactly the people it wants to reach but cannot serve: the ones who need “a whole set commissioned.” Once this positional relationship is clear, the commercial strategy is clear: buy components from it, and sell systems to its customers.

7.7 A side-by-side comparison of the six

08Competitive landscape: a three-layer industry chain, and four regularities

Section 06 cuts by “customer and money.” This section cuts by “position on the chain” — two projections of the same thing.

8.1 Three layers by position on the industry chain

Of the three layers, L1 competes on relationships and L3 competes on scale. Only L2 competes on “technical density.” And the features of L2 are exactly these: every site needs a set, every project has differences, the owner cannot do it, and large vendors will not put labor into a non-standard order.

The correspondence of the two projections is clear: the “high-end industry market” of Section 06 is taken mainly by L1 players, the “small and mid-sized computer-room market” is taken by L2 players, and the “carrier market” is split between L1 and L2 by centralized-procurement share. L3 players do not face these three kinds of customer directly. They supply L2.

8.2 Three ways of staying alive inside L2

Not everyone inside L2 lives the same way. Three typical ways observed:

• Industry-specialist type: do only financial computer rooms, or only archive storerooms, or only base stations. The advantage is a mature scheme and fast replication. The disadvantage is that once that industry’s budget contracts, there is no second curve.
• Regional-engineering type: win on a local service radius, earning the money of “fast response, able to be on site.” Gross margin is high, but scale cannot be reached, and it depends heavily on the owner’s personal relationships.
• White-label supporting type: do ODM for branded equipment plants and integrators, and do not put out one’s own brand. The amount per order is stable and repeat purchase is clear, but low gross margin and tight payment-term constraints have to be accepted.

Of the three ways, the white-label supporting type has the most predictable cash flow, and it is also the easiest for the upstream party to press on price. Its moat is not in the product. It is in the switching cost of “changing supplier means doing the certification again and commissioning the protocols again.”

8.3 Four regularities read from the six

Regularity one: in this industry, scale and gross margin run in opposite directions

Jianda Renke, which makes components, has revenue of RMB 162 million and gross margin in the 10–20% range. Shangwei, which makes units, has revenue of RMB 19 million and a gross margin of 66%. Gongji, which makes the platform, has gross margin above 50%. This is not a difference in management. It is decided by position — the closer to the customer, and the more integration risk carried, the higher the gross margin; the closer to manufacturing, and the more standardized, the lower the gross margin.

Corollary: to raise gross margin, the move has to be toward the customer end, not toward the product end. This sentence is critical for Tongwei — the direction for raising gross margin is not “making a better box.” It is “carrying more of the delivery responsibility.”

Regularity two: those who have lasted all held the same customers; those who fell behind all changed tracks more than once

Gongji has done only data-center infrastructure management for 30 years. Jitong has done only critical-facility monitoring for 28 years. Renke has done only environmental sensors for 10 years. In the other direction, Shangwei states the problem bluntly in its announcement: the old business is declining, the new business has not yet grown up, and the stretch in between is the hardest. The transition it is making now (computer room → distribution-station room) changes the customer group. The product line has not been torn down and rebuilt. This is a relatively safe way to change.

Corollary: a transition should change customers on top of one’s own technical accumulation. It should not mean learning a new trade on someone else’s ground.

Regularity three: in this industry, a case is not marketing material. It is an entry qualification.

Jitong puts “Daxing Airport, the Hong Kong–Zhuhai–Macao Bridge” on its official site not for appearance — it is because, at the next bid, the “comparable project record” column on the evaluation form has to be filled with them. Gongji writes “head-office level of the major state-owned banks” in detail. The logic is the same.

Corollary: projects have to be chosen, and they cannot be chosen only for revenue. One reference project that does not make money but carries the right to be named may be worth more than three profitable miscellaneous orders. This does not conflict with the “volume up, profit down” of Section 03 — taking a reference project at a low price is an investment. Taking miscellaneous orders at a low price is consumption.

Regularity four: 6 people can produce 42 software copyrights, but 6 people cannot produce growth

The Jinpengzheng example shows that the technical threshold of this industry is far lower than imagined. A small team can fully produce a usable product. What actually blocks people is not technology. It is “time can be sold only once” — one person’s on-site hours, once sold to project A, cannot be sold to project B.

Corollary: all growth has to come from “things that can be reused”: reusable product models, reusable protocol drivers, reusable delivery manuals, reusable channels. Any revenue that requires putting labor in again is not growth. It is only busyness.

09The objectivity section: the parts that did not change over the past two years

Talking only about change easily turns into selling anxiety. This section talks about what did not change.

An honest industry review must answer both “what changed” and “what did not change.” The following five points barely changed in substance over 2024–2026:

10Outlook 2027–2029: six trends we treat as certain

From this section on, these are this company’s views. They are no longer statements of fact.

The six below are judgments we are willing to sign publicly and to have tested three years from now. Each one is attached with “what signal would appear first if this judgment is wrong.”

Trend one: the billing unit of monitoring will change from “the point” to “the site”

The industry custom now is to quote by monitoring point (so much per temperature and humidity point, so much per electrical point). This way of pricing was reasonable in the era of “one computer room.” It is failing in the multi-site era — because what the customer is actually buying is not 40 points. It is “I no longer have to send people to these 200 sites.”

We judge that, by 2028, the mainstream way of quoting will turn to “a packaged price per site + an annual platform fee,” and the point will recede into an internal accounting unit. This favors vendors who can build a complete site scheme at a low BOM cost.

Falsification signal If, three years from now, tender documents still use “unit price per point” as the main basis of bid evaluation, this judgment is void.

Trend two: protocol capability will become the only barrier a price war cannot punch through

Hardware can be copied. Interfaces can be copied. There will always be someone cheaper. But “this site has a 2008 UPS, it speaks only the vendor’s private protocol, and the vendor no longer does this business” — no second company is willing to take that job.

Our judgment is: over the next three years, profit at the L2 layer will concentrate toward “people who can take the awkward jobs,” not toward “people who can make standard products.” Standard products will be flattened completely by e-commerce and reference-design schemes.

Falsification signal If a universal protocol-conversion module appears that covers the great majority of installed protocols and is priced below RMB 500, this judgment is void.

Trend three: “unattended operation” will change from a selling point into an acceptance clause

The practice of 4821 distribution rooms in Foshan produced quantifiable gains: 12264 fewer inspections, 24075 fewer labor hours, and maintenance outages reduced by 90%. When figures of this kind are cited repeatedly inside the grid system, they become assessment indicators. Once they become assessment indicators, they become acceptance clauses.

We judge that, in 2027–2028, intelligent retrofits of distribution rooms, pump stations, and base-station equipment rooms will generally require an evidence chain that “remote means can replace on-site inspection” (online rate, alarm closed-loop rate, remote-handling rate), and not merely “monitoring was installed.”

Falsification signal If retrofit projects still take “a complete equipment list” as the sole acceptance standard, this judgment is void.

Trend four: Xinchuang will change from a “special project” into the “default”

Xinchuang is still a retrofit that can be set up as a project of its own. But as domestic CPUs and operating systems mature on edge devices, it will become the default requirement of procurement. “Domestic by default” is more thorough than “domestic when required,” because it no longer comes with a budget of its own.

The meaning for vendors is direct: the cost of Xinchuang adaptation has to be spread into the standard product in advance, not quoted separately on every project. Any vendor that charges for adaptation as a project will be pushed off the mainstream list around 2028.

Falsification signal If mainstream tenders still allow an x86 + Windows scheme to win, this judgment is void.

Trend five: the form of delivery will shift from “the project” to “standard units + white label”

In the past, every order was custom: drawings, a quote, programming, commissioning, one person watching one project. This model cannot be sustained under the double squeeze of rising labor cost and falling project amount.

We judge that small and mid-sized vendors who can survive will converge their product form to 3–5 “standard units” (for example: a distribution-room auxiliary-control unit, an energy-storage-station safety-monitoring unit, a computer-room PEM unit, an in-cabinet display unit). Each unit has a fixed interface definition, a fixed configuration list, and a fixed commissioning manual, with only 20% of the space left for customization.

The share of customization is the best indicator of whether a company has finished the transition. A company whose customization exceeds 50% is still, in essence, a contracting crew. A company below 20% is a product company.

Falsification signal If, three years from now, the industry still generally delivers by “one project, one scheme” and gross margin holds steadily above 35%, this judgment is void.

Trend six: the real competitor is not a peer. It is “no need to buy.”

4G/5G module prices are already low enough for a sensor to connect straight to the cloud. Cameras come with their own AI and alarms. PLCs come with their own remote modules. For many small scenarios, “install a PEM set” is becoming an option that can be replaced by “buy a few smart sensors + one cloud account.”

This is not a threat. It is a contraction of the boundary. We judge that PEM monitoring will keep being eroded by “light schemes” in low-end scenarios, and will become still harder to replace in mid- and high-end scenarios that need multiple protocols, multiple sites, interlock control, and passage through acceptance.

Falsification signal If a general-purpose unit appears in the RMB-100 class, supporting 104/61850 and capable of local interlock, this judgment is void.

11Opportunity map: a bottom-up estimate by site and by point

The estimation process is set out in full. Overturning it is welcome.

11.1 Estimation method

We abandon the top-down algorithm of “market size × share,” and use instead:

A figure calculated this way will be an order of magnitude smaller than a “hundred-billion-yuan track,” but it is a figure that can be taken as a sales target.

11.2 The order of magnitude of five segment scenarios

11.3 Scenario priority: ranked by “accessibility,” not by “scale”

12Risk list: seven premises that would void the judgments

Managing expectations matters more than forecasting

13Tongwei Smart’s position

This section is the company’s position. It represents judgments we are willing to put resources into and to bear the consequences of.

In the twelve sections above, we tried to do only subtraction: delete figures that cannot be trusted, delete technologies that are overrated, delete markets that are not ours. This section says what we intend to do.

13.1 How we position ourselves

This positioning sounds conservative. It was bought with eight years of field experience. Across the projects we have done, what actually brings a customer back has never been whether the big screen looks good. It is “that old piece of equipment that would not connect last time — you commissioned it in two days.”

13.2 What we judge this industry lacks most

13.3 Three things we will do

1. Converge the standard units to 3–5. Each unit has a fixed interface, a fixed configuration list, and a fixed commissioning manual, with customization held within 20%. The aim is to change the delivery model from “one project, one scheme” to “selection + a small amount of configuration.”
2. Maintain the protocol library as a core asset. Every private protocol, piece of old equipment, and abnormal condition met on site is settled into a reusable driver and a troubleshooting manual. This is the only thing of ours a price war cannot punch through.
3. Make supporting supply the main business. Do white-label support and small-batch ODM for equipment plants and integrators that have a brand and a channel and lack R&D capability. Do not put out our own brand. Sell delivery certainty to them, and leave the channel to them.

13.4 What we explicitly will not do

13.5 Three sentences we promise customers

13.6 One sentence in summary

This industry will not lack demand over the next three years. It will lack people willing to do one small thing until it does not go wrong.

The essence of power and environment monitoring has never been “intelligence.” It is nothing goes wrong. Every peer who makes this sound mysterious ends up back at the same question: that alarm at three in the morning yesterday — was it real?

We only want to answer this one question well.

AppendixData sources, grades, and the list of standards

A.1 Principal data sources

A.2 Scopes we do not use (and why)

• “The size of China’s PEM market is RMB XXX hundred million” — public figures for the same country and the same year differ by 10.8 times, and no one states the scope. Not used.
• “A hundred-billion-yuan installed-base retrofit market” — that figure counts in civil works, primary-equipment retrofit, and the general contractor’s management fee in full, and it is not on the same order of magnitude as the procurement amount of the monitoring unit. Not used.
• “CR5 reaches XX%” — different sources range from 41% to 58% and contradict one another, which shows the pattern is not settled. Not used.
• A “growth rate” with no year and no subject marked — this white paper uses only data that carries a subject and a period.

A.3 List of key standards