BMS obsolescence: your system is working fine, and that’s the problem
There is no alarm for obsolescence.
A building management system will tell you when a chiller trips or a sensor fails. It will not tell you that the controller running your data hall went out of production four years ago. It will not tell you that the server hosting the front end stopped receiving security patches last October. Every point on the screen can be green on a system that nobody can actually repair.
That is what makes BMS obsolescence difficult to act on. It’s a slow drift. Nobody ever decides to run an unsupported control system; they arrive at one a year at a time, because each individual year it worked.
Three layers, three different clocks
A control system is three layers, and each layer doesn’t expire together.
Field devices are the sensors, valves and actuators out in the plant. They age slowly, often fifteen years or more, and they tend to drift out of accuracy long before they fail outright.
Controllers last twelve to eighteen years in service. The catch is that manufacturer support usually ends years before the hardware does.
The head-end is the supervisory software your operators look at, plus the operating system underneath it. This ages fastest of all, because its life is set by Microsoft rather than by anything in your building.
An EPMS follows the same pattern. Its meters and protection relays sit in the switchgear and usually outlast everything above them. Gateways carry their data onto the network, and the head-end that records power events and alarms runs on Windows servers, so it is on the same clock as the BMS. And because the EPMS often runs as a separate system from the BMS, a review that only looks at the BMS will miss it.
Most estates buy all three at once, then think about them as one system for the next fifteen years, and that’s where the trouble starts.
The dates are already in the diary
Windows 10 reached end of support in October 2025.
On 13 October 2026, Windows 10 Enterprise LTSB 2016 and Windows 10 IoT Enterprise LTSB 2016 reached end of support. The IoT edition matters here, because it is common on BMS panel PCs and operator workstations, chosen precisely because it was a long-term release.
The same day, the third and final year of paid security updates for Windows Server 2012 and 2012 R2 runs out. After that there are no further patches to buy at any price.
On 12 January 2027, Windows Server 2016 reaches end of support. It remains a very common host for head-end and historian servers. It remains a very common host for BMS and EPMS head-end and historian servers.
Controls platforms run their own clocks alongside. The Niagara AX platform, which sits underneath a wide range of branded building automation products, was declared end of life in July 2021, with licensing support ending two years later. Moving to Niagara 4 is not a software update. The legacy JACE 3, 6 and 7 controllers cannot run the current version, so the migration replaces hardware as well. That is usually a surprise to whoever budgeted for new software.
Five warning signs, and hardware is the least dangerous
Obsolescence gets discussed as a hardware problem because hardware is the part you can see. In practice it arrives in five forms.
- Hardware. Parts go out of production and the only source becomes the second-hand market.
- Software. The head-end version, the operating system beneath it, and whether the engineering tools to change anything still exist.
- Spares. Usually assumed rather than documented. Ask for a written list matched to your asset register and see what comes back.
- Skills. How many named people can write to your control logic?
- Security. Older controllers often have no authentication, no encryption and no way to receive a firmware update.
A lot of people we talk to worry about the first and ignore the fourth. That is the wrong way round. A discontinued controller is a procurement problem with a known price. A system that one person understands is an operational risk with no price attached until the day it has a very large one.
Why this is becoming a boardroom question
UK data centres were designated critical national infrastructure in 2024. The Cyber Security and Resilience Bill, expected to receive Royal Assent in late 2026 or spring 2027, makes data centre services essential services under the NIS framework, with Ofcom as regulator and incident reporting measured in hours.
Two details matter for anyone running controls. The threshold is 1MW of rated IT load for third-party facilities, so this is not a hyperscale-only regime. And the Bill’s definition of a data centre service explicitly covers power supply, cooling and environmental controls, which puts BMS and EPMS inside the scope rather than alongside it.
That changes the question from when to upgrade into who signs this off.
You are already paying for it
Repairs take longer every year as fewer engineers know the platform. Spares cost whatever the market will bear. Sequences drift under a decade of workarounds nobody removed. Alarm lists grow until operators learn which categories to ignore. And any analytics platform you buy sits on top of point data whose quality was fixed years ago.
None of that appears as a line on a budget, which is exactly why no one has approved a project to fix it.
Score subsystems, not buildings
Here is the most useful thing you can do about it, and it takes an afternoon.
Stop asking how old the building is. Score each layer separately: BMS head-end, BMS field devices, EPMS head-end, EPMS field devices, and the network that connects them. Rate each against the five signals above.
The result is almost always more encouraging than people expect. A typical estate has a head-end that expired years ago sitting above field controllers with a decade of life left. A building-level view says replace everything. A subsystem view says replace the supervisory layers and fix the network, which is a materially smaller project, retires most of the risk, and can be phased across two budget years rather than fought for in one.
Three questions to ask this week
- What operating systems are your BMS and EPMS head-ends running, and are they still receiving security updates?
- Could you obtain a like-for-like replacement for your three most critical controllers, new, within 48 hours?
- How many named individuals can write to your control logic?
If you cannot answer those from memory, that is itself the finding. It is also the cheapest moment this will ever be to start.