Industry Insights

Smart water metering in 2026: what actually changed

Tenders quadrupled, water now leads LoRaWAN demand, and AMI endpoints head for 154M. What the 2026 numbers change in a meter-reading design.

A water utility signs a ten-year metering contract, the meters arrive, and eighteen months later the reading success rate sits at 94 percent. Nobody can say which four hundred thousand meters are the missing six percent, because the contract never specified what a "read" is, what medium the frame travels over, or what happens when a cast-iron pit lid sits between the antenna and the street. The 2026 market numbers say this scenario is now the exception rather than the rule — the technology question has largely been settled — and the risk has moved into the arithmetic nobody puts in the tender.

That is the actual change of the past twelve months. Not a new radio, but a shift in where smart water metering projects fail: from "which LPWAN" to "how many collectors, what protocol on the bus, and how many decibels are left at the worst pit in the district".

Three numbers from 2026 worth putting in a tender

On 10 November 2026 the LoRa Alliance published its year-end market position: 150 million LoRaWAN devices deployed worldwide, of which more than 30 million are meters — one device in five. The same release quotes Kurrant's Tender Tracker showing that EU-wide tenders mentioning LoRaWAN more than quadrupled between 2019 and 2025, growing a further 13 percent in 2025 over 2024, with water metering accounting for roughly half of LoRaWAN-related tenders, ahead of building energy metering, network deployments and street lighting.

The second number is a forecast, not a shipment: Berg Insight counted 79.1 million water AMI endpoints installed across Europe and North America at the end of 2025 and projects 154.5 million by 2031, a compound growth rate of 11.8 percent. Its wider figure for all communicating water meters — AMI plus drive-by AMR — goes from 190.1 million to 249.2 million over the same period. The gap between those two series is the interesting part: the growth is concentrated in fixed networks, not in mobile reading.

The third is regulatory. AMP8, the English and Welsh water industry investment cycle that runs from April 2025 to March 2030, has produced a wave of announced metering decisions, and a Netmore analysis quoted in the same LoRa Alliance release found LoRaWAN present — as sole technology or as the primary component of a dual-technology deployment — in 70 percent of them.

Water AMI endpoints in Europe and North America, 2025 to 2031

Evidence, 2026

Dated source

What it changes for a buyer

30M of 150M LoRaWAN devices are meters (one in five)

LoRa Alliance, 10 Nov 2026

Metering is now the anchor application, so module roadmaps follow meter requirements

EU tenders mentioning LoRaWAN >4× 2019 level, +13% in 2025

Kurrant via LoRa Alliance, 10 Nov 2026

Dual-radio bids are normal; a single-technology bid invites comparison

Water AMI 79.1M (2025) → 154.5M (2031), 11.8% CAGR

Berg Insight via IoT Business News, 16 Jun 2026

Collector density, not meter count, becomes the cost driver

LoRaWAN in 70% of announced AMP8 decisions

Netmore via LoRa Alliance, 10 Nov 2026

Interoperability clauses become negotiable, not aspirational

69,000 meters, Aarhus Vand + Kamstrup, rollout to 2030

Netmore, 25 Aug 2026

Municipal rollouts are phased over years, not quarters

The scale is not hypothetical. In August 2026 Netmore announced a 69,000-meter deployment with Aarhus Vand and Kamstrup, starting from a 1,000-household pilot in Tranbjerg and running through 2030. That phasing matters more than the headline: a municipality installs in tranches, so the network has to accept new meter generations for a decade.

Read the table as a purchasing document rather than a market summary. Four of the five rows imply the same requirement: the network has to accept meters from more than one vendor and more than one radio, because the utility will still be adding meters when the first batch is obsolete.

The interface is not the protocol

Most specification mistakes happen one layer too high. A meter data sheet says "RS485" and a collector data sheet says "RS485", and the tender assumes they will talk. They will not, necessarily: RS485 defines a balanced electrical interface, not a byte stream. The frame on top of it is a separate decision, and in water metering there are at least four families in active use.

Layer

What it actually fixes

What it leaves to the integrator

RS485 (ANSI/TIA-485)

Differential voltage levels, common-mode range, multidrop wiring

Baud rate, parity, addressing, timing

CJ/T 188

Chinese residential meter data frame: address, data identifier, sequence

Which data identifiers a given meter implements

DL/T 645

Chinese electricity meter frame and data identifiers

Mapping to billing systems

EN 13757-2/-3

Wired M-Bus physical and application layers

Gateway translation to IP

EN 13757-4

Wireless M-Bus radio frame (drive-by and fixed network)

Receiver placement, overlap with other radios

LoRaWAN

LPWAN link layer on regional ISM parameters

Regional parameters, join procedure, payload codec

The practical consequence: a collector that speaks "RS485" can be electrically perfect and still return nothing usable, because it is polling with the wrong data identifier or the wrong frame length. This is why the TP6306 collector ships as distinct variants — the -WL for cold water meters on CJ/T 188, the -EM for electricity meters on DL/T 645, the -SE for Modbus RTU sensors — rather than one firmware claiming to cover everything. Each variant handles up to 32 slaves on the bus, 6 on the -SE models.

If you are building the meter rather than the collector, the radio side of the same problem looks different: the module has to be small, low-power and predictable, which is the design point of the TP1107-MP36 — 18-pin LCC, 40.5 × 25 × 4.4 mm, 17 dBm transmit, −120 dBm sensitivity at 19.2 kbps FSK, 5.5 V supply with 1.9 A transmit current. Note the current figure: the airtime is short, but the instantaneous draw is not, so the battery and the bulk capacitor are a pair, not a single choice.

Where the "one collector, 32 meters" assumption breaks

The arithmetic of a fixed network is dominated by one ratio: meters per collector. At 32 slaves per bus, one million meters needs 31,250 collectors. That number is the project's real unit cost, because each collector needs a mounting point, a 220 V AC feed, an antenna position and a maintenance visit.

The read cycle itself is not the constraint. With a 32-byte request and an 84-byte response — a deliberately pessimistic assumption, since CJ/T 188 frames are usually shorter — and 2400 bps, one meter takes 0.48 s of bus time plus 0.15 s of turnaround, so 0.63 s. Thirty-two meters is 20.3 s per full cycle. Read four times a day and the bus is busy for 81 seconds out of 86,400 — 0.094 percent of the day. Move to 9600 bps and the same cycle takes 8.7 s, which buys nothing in a system that is idle 99.9 percent of the time.

Read cycle time against the number of meters on one RS485 bus

Quantity, per collector

Value

Basis

Bus time per meter

0.63 s

116 B at 2400 bps + 150 ms turnaround

Full cycle, 32 meters

20.3 s

computed

Bus time per day, 4 cycles

81 s

0.094% duty

Uplink time per cycle

2.13 s

32 records × 16 B at 2400 bps

Uplink time per day

8.5 s

4 cycles

Collector load

under 2 W

TP6306 specification

Estate load, 31,250 collectors

62.5 kW, ~548 MWh/yr

computed at the 2 W ceiling

That last row is the one utilities under-budget. Sixty-two kilowatts is a small building's continuous draw, and 548 MWh a year is a line item on an energy contract, not a rounding error. A battery-powered meter radio does not have this problem because there is no collector; a mains-powered RS485 collector does, and the number scales linearly with the meter count divided by 32.

The data volume, by contrast, is trivial: 16 bytes per record, four records a day, 365 days, is 23.4 kB per meter per year — 23.4 GB for a million meters. Storage and backhaul are not the design constraint. Read success rate is.

Reading through a cast-iron lid: the 131 dB budget

Link budget first, radio theory second. The TP6306 transmits up to +20 dBm and receives at −111 dBm on its 2.4 kbps uplink, so the budget between two such units is 131 dB. Free-space loss at 470 MHz is 32.44 + 20log10(470) + 20log10(d), which is 85.9 dB at one kilometre, 91.9 dB at two, 97.9 dB at four.

Subtract the clutter and the picture changes quickly. A meter in a basement adds roughly 12 dB; a meter under a cast-iron pit lid plus one wall adds about 27 dB, which is the number most field surveys land on. At one kilometre, the pit case leaves 18.1 dB of margin. At two kilometres, 12.1 dB. At four, 6.1 dB — enough on paper, and not enough once a wet lid, a re-radiating fence and a 3 dB body-loss from a hand on the antenna are added.

Link margin at 470 MHz for open air, basement and pit installations

Distance

Open air

Indoor, +12 dB

Pit + wall, +27 dB

1 km

45.1 dB

33.1 dB

18.1 dB

2 km

39.1 dB

27.1 dB

12.1 dB

4 km

33.1 dB

21.1 dB

6.1 dB

Zero-margin distance

180 km

45 km

8.1 km

Those zero-margin distances are arithmetic, not coverage. At 470 MHz with both antennas at 1.5 m, the radio horizon is about 10 km — 4.12 × (√h₁ + √h₂) with heights in metres — so the open-air and indoor columns are horizon-limited long before they are budget-limited. The pit column is the only one where the budget runs out first, and it runs out at 8 km, which is why pit meters are the design case and everything else is margin.

Four failure modes the business case hides

The silent one percent. A 99 percent read rate sounds like success until you multiply it by a million meters: 10,000 meters report nothing on any given cycle. At four reads a day that is 10,000 gaps per cycle to triage, and a utility with no per-meter diagnostic ends up sending a van. The diagnostic that matters is not RSSI but the frame-level error counter on the bus plus the last-successful-read timestamp, both of which a collector can log locally.

The protocol that half-works. A collector polling 32 meters will usually read 32 meters — until one meter is replaced by a different brand with a different data identifier set. The failure appears as a time-out on one address, and it is invisible in aggregate statistics. Specify the data identifiers, not just the protocol name.

The collector that cannot be powered. 220 V AC at a pit-side cabinet is a civil works item. Where mains is unavailable, the collector becomes a solar-and-battery design, and the 2 W continuous draw becomes the sizing case rather than an afterthought.

The latency assumption. Four reads a day means a leak can run for six hours before it is visible; at 1 m³/h that is 6 m³ of water, and at a larger burst far more. If the business case assumes hourly leak detection, the read schedule has to be hourly, and the bus time per collector goes from 81 s to 1,215 s a day — still under 1.5 percent of the day, but no longer negligible for a 32-slave bus shared with other polling.

One million meters, worked end to end

Take a municipal rollout of one million cold-water meters, four reads a day, 32 meters per collector, 2400 bps on the bus.

The estate needs 31,250 collectors. Each reads its bus in 20.3 s, four times a day, for 81 s of bus time and 8.5 s of uplink time — under 90 seconds of radio and wire activity per collector per day. At the 2 W ceiling, the collector estate draws 62.5 kW and consumes roughly 548 MWh a year. The data generated is 23.4 GB a year, which one modest server absorbs without noticing.

Now the same estate at a 96 percent read rate: 40,000 meters silent per cycle, 160,000 silent reads a day, and a support queue that has to be staffed for it. At 99.5 percent, 5,000 silent meters per cycle — still a field-team problem, but a tractable one. The difference between those two numbers is not the radio. It is the pit installation, the bus termination and whether the collector's antenna was placed where the survey said it should be.

Verifying a pilot in one afternoon

You do not need a month to find out whether a district will read. You need four measurements at the worst case, not the average case.

  1. Pick three pits with cast-iron lids and one basement meter room. Log the per-frame error rate over 24 hours at those four points only.
  2. Compare the collector's reported RSSI against a handheld reading at the same pit. A 10 dB disagreement means the antenna is not where the survey assumed.
  3. Read the bus error counters before and after adding one meter to the end of the chain. A jump in CRC errors is a termination or stub problem, and it will not appear until the bus is nearly full.
  4. Force one meter to the edge of its timeout — a slower baud rate, one extra retry — and watch whether the collector drops the whole cycle or just that address. A collector that fails the cycle is the reason 99 percent read rates become 94 percent.

If you want the margin arithmetic in your own tender documents, the formula is four lines of Python:

import math
fspl = lambda d_km: 32.44 + 20*math.log10(470) + 20*math.log10(d_km)
budget = 20 - (-111)                     # +20 dBm TX, -111 dBm RX
for d in (0.5, 1, 2, 4):
    print(d, round(budget - fspl(d) - 27, 1), "dB margin, pit case")

Run it with your own clutter allowance and you get a coverage radius you can put in a contract, rather than a vendor's optimistic one. The same calculation applies to point-to-point links built on the TP5603-D01L10, whose −116 dBm sensitivity and 17 dBm output give a 133 dB budget — two decibels more than the collector, which is worth remembering when a single hop has to cross a road.

FAQ

Is LoRaWAN now the default for water metering?

In Europe it is the most common single answer, not a universal one. Berg Insight still finds proprietary and EN 13757-based RF leading the installed base, with LoRaWAN and 3GPP LPWA the fastest-growing categories for new deployments, and LTE-M the fastest-growing in North America. Treat the 70 percent AMP8 figure as evidence that dual-technology bids are acceptable, not that a single radio is mandatory.

Why specify 32 slaves per collector instead of 64?

Because bus time and failure blast radius both scale with the count. At 32 slaves a full cycle is 20 s and one bad meter costs one address. At 64 the cycle doubles and a stub or termination fault takes out twice as many meters. The TP6306 stops at 32 for the same reason, and the Modbus variant at 6.

Does a higher baud rate solve read-rate problems?

Rarely. At 9600 bps the 32-meter cycle drops from 20.3 s to 8.7 s, but the bus is already idle 99.9 percent of the day. Read failures come from RF margin, protocol mismatches and bus wiring, none of which a faster baud rate touches.

How much margin should a pit installation keep?

Twelve decibels at the design distance is a workable number: it covers a wet lid, a hand on the antenna and a few decibels of ageing. Below 6 dB the link works in summer and fails in the wet season, which is the most expensive failure pattern because it looks like a firmware bug for the first month.

Browse the full collector and module range on the products page if you are mapping a district before the tender closes.

More in Industry Insights