Cupix is repositioning CupixWorks from a construction-focused 3D digital-twin platform into what AEC Magazine reports the company calls Enterprise Spatial Intelligence, extending the platform past practical completion and into operations and maintenance. AEC Magazine reported the move in a piece posted 5 October. “The upcoming version 5.0 is intended to extend the platform across the asset lifecycle, from construction into operations and maintenance,” it reports.
A note on sourcing before the substance, because it bears on how firmly any of this should be read. AEC Magazine attributes both the version number and the phrase to Cupix itself. On the company’s own homepage, neither is something a visitor would see: the page’s only “5.0” strings are coordinates inside SVG path data, and “ENterprise Spatial Intelligence” — with that capitalisation — sits in the page source inside an element the site’s own stylesheet hides. Cupix’s press-release and news pages both return 404. The page a visitor actually reads calls the product a “Construction & Facility Management Spatial Intelligence Platform” in its title and a 4D digital twin in its body. So the version number and the ESI label in this piece are AEC Magazine’s account of Cupix’s own terms, and are carried here as that.
One further caution about that homepage, because this piece nearly drew the wrong conclusion from it. Its markup contains a product-navigation entry reading “Compass and Cupix MCP” — two of the features at the centre of the story — but the link points at /platform/compass-mcp-prototype, which returns a 404, as does its sibling /platform/siteview-prototype. That is unbuilt navigation in a page’s source, not evidence that anything has shipped.
What Gets Joined to What
The claim in the reporting is not that Cupix captures more. It is that the capture stops being a standalone record.
AEC Magazine lists the inputs as 360-degree, drone and laser scan data; BIM and other project data from Autodesk Forma, Procore and Revizto; schedules from Excel, Microsoft Project, Primavera P6 and Oracle Primavera Cloud; and enterprise asset-management data from platforms including SAP, IFS and IBM Maximo.
That last category is the one that marks the shift. Forma, Procore, Revizto and P6 are all tools of the delivery phase — design, coordination, programme. SAP, IFS and IBM Maximo are where a building’s owner tracks equipment, work orders and maintenance over a service life measured in decades. A reality-capture platform integrating with the first group is documenting construction. One integrating with the second is proposing to be part of the operating record.
Cupix’s homepage already advertises integrations with BIM360, Autodesk Forma, Procore, Revizto, Oracle and ArcGIS, so the delivery-side connections are established. The enterprise asset-management side is what the 5.0 reporting adds.
The Handover Problem, Stated Plainly
The reason this is an interesting direction rather than a feature list — and what follows here is this publication’s reading, not the company’s claim — has to do with what happens to construction information at handover, which is mostly that it stops being used.
A project generates an enormous and expensively assembled picture of itself: models, scan data, progress captures, submittals, the record of what was actually built where and when. At completion, the party that paid for all of it receives a set of documents and, typically, a facilities team that works from a different system entirely. The spatial record goes into an archive. Five years later, when someone needs to know what is behind a wall before cutting into it, the answer is usually a site visit and an educated guess.
AEC Magazine’s account of the pitch addresses exactly this: by linking spatial records with asset data, construction history can be carried into operations to support maintenance and future renovation. It also reports Cupix’s claim that the combined dataset can help teams align design, schedule and site progress, and automate comparison of actual conditions against project intent so that deviations surface rather than waiting on manual review.
That is a familiar promise. What makes it more plausible than the generic version is the integration target: the value of knowing what is behind the wall only materialises if the answer appears inside the system the maintenance technician already opens, which is Maximo or SAP, rather than a digital twin the facilities team may not hold a licence for.
Compass, and the MCP Connector
Two of the named features are worth separating from the rest.
Compass, per AEC Magazine, lets users pose questions in natural language and receive answers linked to the relevant captures, dates and locations, searching across captures, notes, model data and assets. That the answers are tied back to a specific capture, date and location is the part that matters — a spatial query layer whose answers carry their provenance is a different product from one that returns a confident paragraph.
The second is an AI-agent interface via what the reporting calls the Cupix Model Context Protocol connector. AEC Magazine’s word is connector rather than server. It is the second MCP connector to appear in this newsroom’s coverage inside a week, after Arcadis plugged its own into Autodesk Assistant. The pattern is becoming hard to miss: vendors holding a proprietary dataset are increasingly choosing to expose it to other companies’ agents through a standard connector rather than build the assistant themselves.
For a reality-capture company that logic is particularly strong, and it is the same argument OpenSpace made in September when it told the market that AI agents need eyes. If the useful thing you own is a verified record of physical conditions, then every agent that needs to check a physical fact is a potential consumer of your data, and the connector is cheaper to build than the chat interface nobody asked you for.
AEC Magazine also lists semantic classification of point clouds and a custom object locator for tracking selected objects in 3D.
Availability, Which Is the Part to Read Carefully
CupixWorks 5.0 is not generally available. AEC Magazine reports that prospective users can apply for access to an invitation-only beta.
This is worth stating as plainly as the feature list, because the feature list is what travels. Compass, the MCP connector, the Maximo and SAP connections and the semantic point-cloud work are, on the available reporting, capabilities of a version in restricted beta — not of software a contractor can buy this month. The unbuilt navigation entries noted above point, if anything, the other way. The reporting does not say which parts of 5.0 exist in what form, and this piece does not know.
The Market It Is Aimed At
AEC Magazine reports the initial targets as complex facilities: hyperscale data centres, pharmaceutical laboratories and semiconductor fabs, among others.
That is a coherent choice, and the common thread is not technical complexity so much as the cost of a wrong answer during operations. In a data centre or a fab, opening the wrong ceiling tile or shutting the wrong valve has consequences measured in revenue per hour, and the people who run them already buy expensive systems to avoid that. These are also the buildings most likely to be modified continuously after handover, which is precisely when a construction-era spatial record earns its keep.
It is the same customer gravity that has pulled the rest of this category toward the hyperscalers — the reason OpenSpace made a point of having documented 1,000 data centres, and part of what Procore was buying when it agreed to pay $845 million for DroneDeploy.
The Company’s Own Numbers
Cupix publishes a set of scale claims on its homepage, and they are reported here with attribution rather than audited: “Trusted by 70% of the ENR Top 30 General Contractors, Cupix has rapidly scaled to capture over 20,000 project sites globally since 2023, using AI to verify more than 1 million as-built BIM elements every single day.”
Cupix is not a single company but a group. Its terms of service name three contracting entities: Cupix America Inc., “a Texas corporation”, for customers in the United States, Canada, Europe and the Middle East; Cupix Asia Pacific Pty., Ltd., “an Australian corporation”, for Australia, New Zealand, Singapore and Hong Kong; and Cupix, Inc., “a South Korean corporation”, for Korea and Japan. The Korean business-registration block in the site footer gives an address in Seongnam-si, Gyeonggi-do. AEC Magazine’s piece names a list of large contractors among the platform’s users; this piece does not reproduce that list, because customer names are the claim class this newsroom holds to the strictest standard and a transcribed roster is the easiest place to introduce a misspelling or an entity error.
Analysis: The Category Is Running Out of Construction
What follows is this publication’s assessment, not reporting.
Reality capture for construction has been a solved commercial problem for a few years now, in the sense that the large contractors have bought it, the workflows exist and the differentiation between vendors has narrowed to integrations and analytics. That is a market with a ceiling, and the ceiling is the number of large projects under construction at any time.
Operations is a structurally larger market with a longer contract life, and it is where every player in this space has an obvious reason to go: the asset exists for decades after the project ends, and the owner who pays for maintenance is a more durable customer than the contractor who demobilises. Cupix moving toward Maximo and SAP is the same instinct that took OpenSpace toward agents and Procore toward owning the capture layer outright.
The hard part is not the integration. It is that the construction record degrades in usefulness the moment the building starts changing, and nobody is paid to keep it current after handover. A spatial twin that is accurate on the day of practical completion and never recaptured is a historical document, and a maintenance technician who finds it wrong once will not open it again. Whether Cupix has solved that, or any of the owner-side incentive problem underneath it, is not something the available reporting establishes — and with the product in invitation-only beta, it is not a question that can be answered from the outside yet.
