McCarthy and Palantir just showed what enterprise construction AI actually looks like
McCarthy Building Companies signed a multi-year deal with Palantir to build Pulse, an AI operations system connecting estimating, field ops, contracts, and logistics. Here's what that architecture signals for the rest of the industry.
Most AI in construction is still a point tool: a faster takeoff here, an AI submittal reviewer there. McCarthy Building Companies just went a different direction.
On June 4, McCarthy and Palantir Technologies announced a multi-year, multi-million-dollar strategic partnership. The center of it is a system called Pulse — McCarthy's AI-native operations platform, built on Palantir's enterprise data infrastructure. Palantir's prior customer list runs through intelligence agencies, hedge funds, and the Department of Defense. That it's now inside a commercial GC's operations is the relevant signal.
What Pulse is and what it does
Pulse is built on Palantir's Artificial Intelligence Platform (AIP), with McCarthy's construction workflows modeled in Palantir's Ontology — the layer that represents your business as a connected data graph, not a set of siloed databases.
For a superintendent or PM on an active job, Pulse is designed to deliver:
- Real-time insight across the project's current state, pulling from multiple data sources simultaneously
- Scenario planning — running "what happens if steel is three weeks late" against the actual schedule, crew commitments, and downstream trade sequence
- Risk analysis that surfaces constraints before they cascade across the critical path
- Decision orchestration connecting data from estimating, contracts, buyout, QA/QC, logistics, and equipment into a single decision context
The scope here is what sets this apart from most AI deployments in the industry. Procore's AI agents work within Procore (see our coverage of the contech AI acquisition wave last month). Pulse is built to connect data across McCarthy's entire enterprise — not just one platform.
Why the connected-data model is different
The core idea in Palantir's Ontology is that your business data — projects, people, materials, milestones, contracts — is mapped as a connected graph rather than isolated tables. When AI reasons over that graph, a procurement delay on one job can automatically update the resource forecast on another. A QA flag in one division can trigger a protocol review across similar scope elsewhere. A change order hits the buyout model without anyone manually bridging systems.
McCarthy's 160 years of construction expertise are being encoded directly into that Ontology — meaning the system isn't generic. It's trained on the specific workflows, constraints, and decision patterns that McCarthy's teams actually use.
Palantir describes this as a "tool factory": an environment where teams can build and configure AI applications on top of the shared data model, rather than swapping out one-off tools every budget cycle.
What Pulse won't fix
Enterprise AI platforms require real investment in data modeling before they deliver — you have to map your business in detail before any AI reasoning happens on top of it. McCarthy's build represents months of setup work that most GCs won't staff or fund in a single year.
The system doesn't replace field judgment. Scenario planning outputs are only as good as the inputs: if a superintendent's baseline schedule was built on optimistic assumptions, the AI model inherits those assumptions. Pulse helps a PM see options faster. It doesn't read the room on a pour going sideways.
And Pulse is McCarthy's. Other GCs won't get access to McCarthy's custom Ontology. This is a bespoke implementation, not a product on the market.
What this tells the rest of the industry
For large GCs in the ENR Top 100: the McCarthy/Palantir model is the emerging template for how enterprise AI actually gets deployed in construction — implementation partnerships, not software subscriptions. ROI timelines for this kind of build are long; expect 12–18 months before any published outcomes.
For mid-size GCs running 10–50 active projects: the lesson is architectural, not product-specific. The contractors getting the most from AI are the ones where data from estimating informs field decisions and vice versa. If your project data is split across three platforms that don't communicate, layering AI on top doesn't solve that — it exposes it.
The Procore AI product launched in May attempts a similar connected-data approach at a price point accessible to smaller shops. If Palantir's model works for McCarthy, Procore's version becomes easier to justify for the tier below.
Forward this to the person on your team who's still arguing AI is overhyped.
Construction AI Brief covers what actually matters in construction AI, three issues a week. Subscribe at constructionaibrief.com.