The short answer
You can buy AI-powered building automation through six channels: your incumbent BMS manufacturer and its authorised dealers, an independent system integrator, a software vendor selling directly, a cloud marketplace such as Azure or AWS, an ESCO under a performance contract, or a public procurement framework. Which one you use depends less on the vendor list and more on which layer of the stack you are actually missing. Most portfolios already own the controls. What they lack is the data and action layer above them.
Why “where to buy” is harder than it looks in this category
Search for building automation vendors and you get three unrelated kinds of business in the same results: equipment manufacturers who design controllers, local firms who install and program them, and software companies that sit on top of whatever is already there. They are not substitutes for each other, and they do not sell through the same channel.
The market is also more fragmented than its brand recognition suggests. The top ten building automation and control vendors accounted for roughly 21% of total market revenue in 2024, which means the majority of spend goes to regional integrators, specialist software firms, and mid-tier manufacturers who never appear in a category-leader list.
The second complication is that “AI-powered” now covers a very wide range of capability. Memoori’s 2026 landscape research on AI in commercial buildings found that most AI deployed in buildings watches rather than acts: in indoor environment applications, the share of deployed AI that actually closes a control loop sits well below half, and in predictive maintenance, diagnostic alerting dominates over automated workflow integration. Two products can both be marketed as AI-powered building automation while one raises a ticket and the other changes a setpoint.
So before comparing suppliers, work out which layer you are buying.
The three layers you are buying across
Field and control layer. Controllers, sensors, actuators, panels, BACnet/Modbus/KNX networks. This is capital equipment with a 15 to 20 year life. You buy it from a manufacturer through an authorised dealer or an integrator.
Data layer. The part that normalises every system in the portfolio into one consistent model so that analytics and agents can run across buildings rather than per building. This is software, bought as SaaS, and it is where open standards matter most.
Agent and action layer. Optimisation, fault detection, autonomous supervisory control, agentic workflows. This depends entirely on the quality of the layer beneath it. Buying it without a data layer is why so many pilots stall at one building.
Most procurement failures in this category come from buying layer three from a supplier who only really sells layer one, or buying layer one again when the actual gap was layer two. This is the layer people mean when they say agentic building operations, and it depends entirely on the quality of the layer beneath it.
The six places you can actually buy
1. Your incumbent BMS manufacturer
Siemens, Johnson Controls, Honeywell, Schneider Electric, ABB, Trane and Bosch all now ship AI-branded cloud platforms alongside their controls. Siemens has been pushing generative AI query features into Building X, and Johnson Controls continues to expand OpenBlue at portfolio scale, including a commitment announced in March 2026 to scale OpenBlue AI across 5,000 buildings in the UAE with automated peak-load curtailment.
In most markets you do not buy from the manufacturer directly. Installation and ongoing service run through an authorised dealer or branch office, and the technician who arrives typically works for a firm covering one metro area rather than a national account team.
Buy here when: you are replacing controls anyway, you are single-vendor across the portfolio, and lifecycle service matters more than portability. Watch for: the AI layer is usually scoped to that manufacturer’s own equipment. If 40% of your portfolio runs someone else’s controls, you are buying a partial view.
2. An independent system integrator or master systems integrator
Integrators program the controllers, handle protocol translation on legacy equipment, and keep the system running after commissioning. Independents generally cost less per hour than OEM dealers and give more room to negotiate scope, though support quality varies more between firms.
The MSI or SSI model puts one party in charge of coordinating hardware and software across the whole project. It works well on new build with a large integration scope. Its known weakness is that the MSI does not control what it integrates. It stays dependent on third-party vendors, which shows up as version drift, module incompatibility, and a stack of layers that each add margin without adding capability.
Buy here when: you have a mixed estate, legacy protocols, or a new build where someone has to own coordination between subcontractors. Watch for: ask who owns the integrations at the end of the contract. If the answer is the integrator, you have replaced vendor lock-in with integrator lock-in.
3. An independent software vendor, direct
Building operating systems, analytics and FDD platforms, and autonomous optimisation vendors sell SaaS directly to owners and operators. They are hardware-agnostic by design and are bought on subscription rather than capex, which changes both the approval path and the speed of deployment.
This is the channel to use when the controls are fine and the problem is that nothing above them talks to anything else.
Buy here when: you have more than one BMS brand in the portfolio, or you want to run the same optimisation logic across every building regardless of what is installed. Watch for: ask whether the data model is open and documented. A platform built on an open ontology such as RealEstateCore means your normalised data stays usable if you change application vendors later. A proprietary internal model means it does not.
4. Cloud marketplaces
Azure Marketplace, AWS Marketplace and Google Cloud Marketplace list an increasing number of smart building and energy platforms. Buying through a marketplace lets you transact under an existing cloud agreement, draw down committed cloud spend, and skip a separate vendor onboarding and security review in many organisations.
Worth knowing: nearly every major commercial buildings automation vendor depends on Microsoft Azure OpenAI services for its underlying AI infrastructure. The incumbents mostly do not own their foundation models. That is not a problem in itself, but it does undercut the idea that any single controls vendor has a proprietary AI advantage.
Buy here when: your organisation has committed cloud spend to burn, or procurement moves faster through an existing cloud agreement than through a new supplier. Watch for: marketplace listings vary in what they include. Confirm whether implementation and integration are in scope or quoted separately.
5. An ESCO or energy-as-a-service provider
Energy service companies bundle the technology into a performance contract, where some or all of the cost is funded from guaranteed savings. The AI layer comes as part of a package rather than as a line item you select.
Buy here when: there is no capex available and the business case is purely energy. Watch for: measurement and verification terms, contract length, and what happens to the software licence when the contract ends. Savings-share models can make the vendor’s incentives diverge from yours on comfort and IAQ.
6. Public procurement frameworks
For public sector, university, healthcare and municipally owned portfolios, most of this is bought through framework agreements, dynamic purchasing systems, or open tender. In the EU, above-threshold contracts go through TED. Several national frameworks already include smart building software categories, which is faster than running a standalone tender.
Buy here when: you are a public body or a publicly owned property company. Watch for: write the interoperability requirement into the specification rather than the vendor name. Requiring open standards and documented APIs is legal and enforceable. Requiring a brand is neither.
Choosing a channel: a short decision path
| Your situation | Channel to start with |
| Controls are end of life across the portfolio | OEM via authorised dealer, plus an independent data layer |
| Mixed BMS brands, controls are fine | Independent software vendor, direct |
| New build, multiple subcontractors | MSI or SSI, with an owner-side data layer specified |
| Legacy protocols, phased retrofit | Independent integrator specialising in modernisation |
| No capex, energy-only business case | ESCO or EaaS |
| Committed cloud spend, fast approval needed | Cloud marketplace |
| Public sector | Framework agreement or open tender |
Before you shortlist anyone: audit what you already own
The single most common procurement mistake is evaluating solutions without a current inventory of the systems already deployed across the portfolio and who is using them. If a BAS is scheduled for replacement in two years, a product that integrates tightly with that BAS is a weak long-term choice, and you cannot know that without the inventory.
A usable pre-procurement audit covers: BMS brand and version per building, protocol coverage, point counts and naming conventions, existing service contracts and their end dates, who holds the credentials, and what data you can already export today without paying anyone.
That audit is also what makes vendor demos comparable. Without it, every vendor will demo against their own reference building.
Questions to ask every supplier
- Does your AI recommend, or does it write back to the BMS? Ask for a named site where it closes the control loop.
- Which of my BMS brands do you support natively, and which need a gateway?
- What data model do you normalise into, and is it open or proprietary?
- If I replace your application in three years, what happens to my structured data?
- Who owns the integrations at contract end?
- How do you prove savings? Which M&V protocol, measured against which baseline?
- What is the cybersecurity posture on write access, and does it support BACnet/SC?
- What is the per-building onboarding time once the first building is live?
- What is the total five-year cost including integration, licences and service, not just year one?
- Can I speak to a customer with a portfolio of my size and mix?
Question one separates the market faster than any other. A supplier who only talks about dashboards and alerts is selling a system of record. A supplier who changes building state is selling a system of action. Both are legitimate purchases. They are not the same purchase, and they should not be compared on the same scorecard.
Red flags
- The AI capability cannot be demonstrated on a building that is not a reference site.
- Integration is quoted as “included” with no point count or scope.
- The data model is described as proprietary and treated as a feature.
- Savings claims are modelled only, with no measured customer result.
- The write-back architecture cannot be explained by the person selling it.
- Pricing per building falls sharply at volume but onboarding time does not.
What it costs and how long it takes
Expect three separate cost lines: software subscription, one-time integration per building, and ongoing service. The integration line is the one that varies most and the one most often underestimated. Ask for it per building, per protocol.
On timelines, a realistic pattern for a mid-size portfolio is one building live in four to eight weeks, a validated use case by the end of the first quarter, and portfolio rollout starting in quarter two once the per-building onboarding cost is known rather than estimated. Treat any proposal that promises portfolio-wide autonomous control in the first quarter as a scoping problem rather than a capability claim.
Frequently asked questions
Can I buy AI building automation without replacing my BMS? Yes. Independent software platforms connect to existing BMS and IoT systems through open APIs and connectors, which is the normal approach for mixed or legacy estates. Replacing controls is a separate decision driven by equipment age, not by AI capability.
Should I buy from my existing BMS vendor or an independent platform? Buy from the incumbent if your portfolio is genuinely single-vendor and you value one lifecycle service relationship. Buy independent if you have more than one controls brand, since a manufacturer’s AI layer is generally strongest on its own equipment.
Is it cheaper to buy through a cloud marketplace? List pricing is usually the same, but marketplace purchases can draw down committed cloud spend and often move faster through internal procurement. Confirm whether implementation is included.
What is the difference between building automation and agentic building operations? Building automation executes rules and schedules that someone configured. Agentic building operations means software that observes portfolio state, decides on an action against a goal such as energy cost or IAQ, executes it through the control system, and verifies the result. The difference is whether the system acts.
How do I avoid vendor lock-in? Specify an open data model, contract for data export in a documented format, confirm integration ownership at contract end, and keep the data layer procurement separate from the application layer procurement.
Where ProptechOS fits
ProptechOS is bought as the data and action layer, directly, on subscription. It connects existing BMS, EMS and IoT systems through open APIs and RealEstateCore-certified connectors, so it runs across a mixed estate without rip-and-replace. Because the normalised model is the open RealEstateCore standard rather than a proprietary schema, the structured data stays yours and stays usable if the applications above it change.
