The Seam — Monthly Deep Dive — Who Owns Your Project Data When AI Touches It? — August 2026
the seam
The Seam · Deep Dive
Monthly Deep Dive · August 2026 · Where AI meets the built environment
This month's deep dive
Who Owns Your Project Data When AI Touches It?
For forty years your drawings, models, and field records sat in files you controlled — mostly because no one else could read them. AI made them readable. In three weeks this summer, three of the industry's largest platforms spent to own that data, not the models that read it.
Why this topic
I picked this from the running agenda because the dominant AEC-AI story of the summer wasn't a model release — it was a wave of consolidation around data. Prefer the group chew on something else next month — permitting and code compliance, the blurring of design and construction, AI and infrastructure? Just reply and say so.
Part 1 · Context
Why this lands now
Between July 14 and July 29, three of the built environment's biggest platforms made acquisitions, and all three were justified the same way: they were buying data, not intelligence.
On July 14, RealPage acquired Cherre (4B+ entities, $4T in tracked assets). CEO Dirk Wakeham was blunt about why: "It's not the models that are the problem… it's mostly that they don't have accurate, reliable data." RealPage was sitting on 25 petabytes of its own data that was "largely unorganized and somewhat unusable." On July 23, Carrier closed its full acquisition of 75F, the AI building-automation company (1,800+ installs) whose software turns sensor streams into operating decisions. And on July 29, Procore agreed to buy DroneDeploy for ~$845M — bringing ~20 trillion sq ft of visual site data and 100,000+ labeled safety issues to a corpus that already holds ~400M photos, 126M drawings, and 10M RFIs. Just before, Trimble moved to buy Document Crunch — the same logic, applied to contracts. Procore/DroneDeploy →
Read together, these aren't four deals. They're one bet placed four times: that in AEC the durable advantage is no longer the algorithm — rented, commoditizing, improving weekly for everyone at once — but the proprietary corpus the algorithm is pointed at.
Practitioners agree, and they're anxious. Revizto's 2026 Bridging the Gap survey (2,006 professionals, 600 CIOs) found 96% of technology leaders concerned about who owns and controls their data; only 10% see AI value with no barriers left. Unanet's 2026 report tells the other half: 75% of firms now use AI, but only 29% trust the data feeding it. Adoption has outrun data-readiness two-to-one. Revizto survey →
And the clock has a regulatory hand now. The morning this publishes — August 2, 2026 — is the day the EU AI Act's general obligations take effect and enforcement begins, including transparency duties. The rules that bear most directly on AEC design (high-risk systems) were pushed to December 2027 — so this isn't the day the hammer falls on a spec tool. But it is the day data provenance stops being a purely internal question. Set against the Anthropic Economic Index — architecture & engineering at ~5% observed AI exposure against ~85% theoretical, the widest gap of any major industry — the stakes get sharp. That gap is either the biggest greenfield in professional services or the largest unpriced exposure on your balance sheet — and which one depends on who controls the data as it closes. Anthropic Index →
Part 2 · What's New
Two roads out of the same problem
Road one — rent the intelligence, feed the platform
The acquisition wave is the visible edge of a consolidation strategy: put your projects on one platform and its agents read your drawings, triage your RFIs, flag your risks — no infrastructure required. The convenience is real. So is the gravity. Revizto caught the tension: 39% of CIOs plan to consolidate onto fewer platforms even as 96% worry about control — firms walking toward the lock-in they say they fear, because the alternative is more work. Autodesk's new Neural CAD (a foundation model reasoning over CAD geometry, aimed at Fusion and Forma) is a genuine step change, but AEC Magazine's Martyn Day named the catch: it "depends on customers entrusting even more of their engineering knowledge, workflows and intellectual property to a single platform." AEC Mag →
Road two — bring the AI to your data
The counter-move matured from slideware into shipping product, and its slogan is nearly identical everywhere. The Open Design Alliance — a 1,200-member non-profit that, by charter, cannot be acquired — is releasing self-hosted MCP servers for DWG, IFC, and STEP, letting firms run AI agents against their own CAD/BIM data entirely inside their walls; Day called it "about as close to a firm-level data sovereignty play as this industry has produced." Days later Revizto shipped an Object Properties API + MCP server: "Your AI comes to your data. Your data never leaves your environment." Research is catching up too: ARCHER showed a self-hosted open-weights model can hit 97.8% of frontier accuracy at ~1/4 the cost on code compliance — and, crucially, that its verdicts are auditable, because the checking code is released rather than sealed. ODA → · ARCHER →
The model is a commodity. The causal substrate is the moat.
— Antony Slumbers, "The Office That Earns Its Rent," June 2026
Neither road is obviously right yet, and most evidence is vendor-shaped. Self-hosting is real but partial (ODA's Revit write-back is a long way off; "run your own model" assumes people to run it). The platform road delivers value today at the cost of a dependency that compounds. What changed this summer isn't that the question got answered — it's that it got expensive enough that everyone has to have an answer.
Part 3 · AEC Implications
What it means from each seat
For architects
Your drawings and models are the corpus everyone is buying. Read the data clause in the tools you already use — KP Reddy calls it the industry's "Facebook moment: you already agreed to it." The exposure is subtle: not that AI produces a bad building, but that the judgment that produced good ones migrates out of your practice and into a platform you rent.
For engineers
ARCHER reframes the licensure question: can you accept responsible charge for a compliance result from a model you cannot inspect? If not, auditability isn't a nice-to-have — it's a precondition for using AI in the part of the work your license covers. Self-hosted, inspectable checkers make "responsible charge over an AI output" thinkable; a sealed black box does not.
For contractors
You generate the richest data — photos, logs, RFIs, schedules — and Procore's DroneDeploy buy means your field record is the asset being consolidated. Two questions bite: who owns the derived data (labeled safety issues, productivity signals), and when you switch platforms, does your history come with you — or stay behind as someone else's training set?
For owners & developers
RealPage/Cherre and Carrier/75F are about your data — feasibility, leasing, operations. The developer who owns a clean substrate walks into the design meeting knowing more than the design team. Slumbers' warning lands here: think hard before renting your operational intelligence from a consortium whose anchor is one of the largest landlords on the planet.
Two things cut across every seat. The contracts haven't caught up — AIA and ConsensusDocs say almost nothing about who owns AI-derived data, so the defaults are set silently by EULAs no one on the team has read. And consolidation and portability pull in opposite directions — you can't fully own your substrate and also enjoy one vendor doing everything. Where you stand on that line is now a strategic decision, not an IT one.
A regional note, since this feeds the Aug 12 breakout at Innovation Depot: Russ Gibbs at Brasfield & Gorrie is Birmingham's most public AEC-AI voice (and on Auburn CARV's advisory board); Sandeep Ahuja's cove in Atlanta is the region's most visible AI-native practice; Auburn CARV is the nearest serious research concentration. A 20-person firm and a national contractor answer this question very differently — which is exactly what makes it worth arguing about in a room that holds both.
Part 4 · Discussion Prompts
Five questions for the room
01
Rent or own — and can a small firm even choose?
Is owning a "data substrate" realistic for a 20-person practice, or a strategy only Procore-scale players can afford?
02
Have you actually read the clause?
Do you know whether the AI tools you use today train on what you feed them — and can anyone really "own" design intent?
03
Can you consolidate and stay portable at once?
39% of CIOs want fewer platforms; 96% fear losing control. They're the same people. Convenience, or trap?
04
Can you stamp what you can't inspect?
Can a licensed engineer accept responsible charge for a code result from a model they can't see inside?
05
Is your data really the moat?
If a competitor had all your files tomorrow, how much of your advantage lives in judgment and relationships instead?
Part 5 · Pre-Read Packet
Read before August 12
Two to ground the question, three on what happened this month, one to argue with. About an hour total.
Foundational · 8 min
The Great Brain Robbery — Richard Harpham, AEC Magazine. The existential version: when AI trains on your drawings and designs autonomously, the risk is that professionals surrender the reasoning that made their work valuable. Read →
Foundational · 25 min
Your Data Is the Product — And You Already Agreed to It — KP Reddy, Insights by KP. The plainest construction-industry read on what "we train on your data" means — reframed through the subcontractor-vetting instinct the built world already trusts. Read →
This month · 5 min
Procore to Acquire DroneDeploy for $845M — Matthew Thibault, Construction Dive. The deal that makes the thesis concrete: your project files are not just deliverables, they are the corpus. Read →
This month · 6 min
RealPage Acquires Cherre — Philip Russo, Commercial Observer. Home of the summer's defining quote: "it's not the models that are the problem… they don't have accurate, reliable data." Read →
This month · 10 min
ODA Opens Its CAD and BIM Tools to AI — Martyn Day, AEC Magazine. The clearest look at road two: self-hosted MCP for DWG, IFC, and Revit from a non-profit that can't be bought. Read →
Read to argue with · 12 min
The Office That Earns Its Rent — Antony Slumbers. The contrarian case: the fight over data architecture is "an argument about plumbing." The moat isn't who holds the data — it's what the data can prove. "Rent the model. Own the substrate." Read →
Share & print
The full deep dive with all sources, as a web page: Read the deep dive →
A one-page handout to print for the table on Aug 12 (stats + questions + QR): Open the handout →
The Seam · AI + AEC · A personal project by Bruce Lanier
Monthly Deep Dive · August 2, 2026 · Where AI meets the built environment
Not affiliated with Birmingham AI or the AEC breakout group. Reply to request next month's topic.
Don't miss what's next. Subscribe to Bruce Lanier: