Back to Knowledge
Insights 9 min read

Rebuilding a $10 Trillion Industry

Ortal Sasson Hendin
Ortal Sasson HendinPrincipal
Raz Mangel
Raz MangelGeneral Partner
Josh Trup
Josh TrupPrincipal

Published on November 22, 2026

Why AEC Is One of the Most Exciting Opportunities in the AI Era

Earlier this year we published the ten themes we believe will define early-growth investing in 2026. In a series of follow-on posts we are going deeper on the verticals where those themes converge most powerfully. This is the first, and one we are particularly excited about.

AI's impact varies enormously across industries. In some it compresses value that took years to build; in others it cracks problems that resisted every prior wave of technology and changes how an industry actually operates. Architecture, engineering and construction (AEC) sits firmly in the second group. The way we build a house, a hospital or a highway today is essentially the way we built one twenty years ago. We believe it will look completely different a decade from now, and the companies driving that change are being built today.

It's Time to Change the Status Quo

Construction is one of the largest industries on earth: roughly $10 trillion of work done a year, heading toward $14 trillion by 2037, and, on a broader measure of construction-related spending, about 13 percent of global GDP. It has also long been among the least digitized; in McKinsey's ranking of US sectors only agriculture came lower. Large projects tend to run about 20 percent behind schedule and as much as 80 percent over budget, and rework caused by poor data and miscommunication alone costs the US industry an estimated $31 billion a year. While productivity across the US economy has risen almost fourfold since 1950, construction productivity has fallen below where it started. Everyone on a project knows before the first pour that it will cost more, take longer and get partly rebuilt. No technology wave has changed that.

Not for lack of trying. CAD replaced the drafting table, BIM added a 3D database, cloud project management moved the paper onto a phone. Each wave digitized the documents around the work and wrapped itself around the existing process instead of replacing it; the expensive part, the engineering judgment that takes twenty years on jobsites to acquire, stayed in people's heads. Even the first LLM startups after ChatGPT hit the same wall: a model that can summarize a spec cannot design a system that fits the space, carries the load and passes code. AEC is uniquely hard: every project is a one-off prototype, built outdoors, by dozens of firms with misaligned incentives, under thousands of simultaneous spatial, regulatory and cost constraints. That complexity is a large part of why the industry lagged, and why the payoff is so large now that the technology is finally good enough.

What is different this time is where the new generation is aiming: not at the paperwork, but at the expertise-heavy work that always required senior talent. Judgment that took two decades to acquire is becoming trainable; catching a clash before it becomes rework is becoming automatable. In some disciplines AI is beginning to resolve entire design problems in a single pass, a complete electrical system for a data center generated overnight, a structural scheme that recalculates as the geometry moves. Expertise is migrating from people into systems, with licensed engineers still signing off on the result, and that changes where value sits in this industry.

Multiple Forces Converging

New AI modalities that can handle the complexity. Vision models read drawings. Emerging 3D foundation models are learning spatial relationships and structural geometry. Physics-aware systems simulate loads, generate designs built to meet code and reason about constructability. Agents connect these capabilities across a workflow, so the output is an engineered result rather than a summary.

Physical AI reaching the jobsite. Sensors, compute and actuators have fallen in cost, and the autonomy stack matured in vehicles is beginning to handle the unstructured, changing terrain of a construction site. Physical work is becoming automatable at a price and reliability that begin to make sense for a contractor.

Structural macro tailwinds. The US industry needs roughly 349,000 additional workers in 2026, most of them to replace retirees, and 456,000 in 2027; about a fifth of US electricians are over 55. Across Europe, electricians are in shortage in 21 countries and plumbers in 20. In the skilled trades this is demographic, not cyclical. Meanwhile the long-term pipeline of work is enormous: the $1.2 trillion Infrastructure Investment and Jobs Act, an estimated $106 trillion of global infrastructure investment needed through 2040, and a data-center boom that has made data centers the largest category of US private office construction. Demand for buildings is rising while the supply of people who know how to build them is not keeping up. Software and machines are among the few things that can close that gap at scale.

Across the Entire Building Lifecycle

Innovation is emerging in every phase, from feasibility through construction and into operations: AI-native software displacing legacy platforms, AI companies going after budgets that used to go to engineering-services firms, and robotics-as-a-service changing the jobsite from the ground up.

Where We See the Most Interesting Opportunities

Several areas stand out to us, each for a different reason.

MEP engineering. Mechanical, electrical and plumbing is one of the areas we find most compelling. It is where a building's tightest spaces meet its densest rules: thousands of components routed through congested ceilings and risers, each governed by code, each colliding with structure and with every other trade. It is a major source of clashes and rework, and it resisted automation because the constraints are physical and simultaneous. That is exactly the problem the new modalities are built to solve. Endra and Augmenta generate building systems, starting with electrical, directly from a building model, designed to meet code and avoid clashes from the start; Spacial turns an architect's drawings into permit-ready structural and MEP plans for homes. The timing matters: in a data center the electrical and cooling systems are the building, and we believe MEP capacity, in design and in the field, is becoming a constraint on how fast AI infrastructure itself gets built.

Delivering the engineering. The same logic applies wherever engineering expertise is sold as a service. Exodigo, a Greenfield portfolio company, fuses multi-sensor data with AI to map what lies underground before anyone digs, replacing exploratory excavation and a big portion of the preconstruction civil engineering work and design. Structural Intelligence applies AI agents to structural production: analysis, drafting and documentation. The most ambitious of these companies do not just sell tools to engineers. They sell the engineering.

Over the last century, complex civil infrastructure projects have been plagued by rising costs, schedule overruns, and persistent uncertainty.

Exodigo is changing that trajectory by fundamentally rethinking project development through AI-Assisted Design (AIAD). Our ability to model subsurface conditions enables us to advance multiple design alternatives simultaneously, evaluate trade-offs in near real time, and move from concept to detailed design faster than traditional methods allow.

The result is a construction-ready package, compatible with any software platform, that reduces uncertainty, mitigates risk, and enables safer, smarter, and more predictable infrastructure delivery.

Jeremy SuardJeremy SuardCEO & Co-Founder

A new generation of design platforms. Revit was designed more than two decades ago around a desktop application and a file. In the AI era that architecture becomes a liability. A cohort of startups is rebuilding BIM as a cloud-native platform where the model is one continuously updated database instead of a document. One example is Motif, founded by former Autodesk executives, which is building a modern building design platform for the AEC industry; others include Arcol, Snaptrude and Qonic. Until recently this was not practical: a BIM platform is an enormously complex piece of software, and rebuilding one for the cloud was beyond the reach of a startup. That has changed, and the timing is now right to challenge one of the most entrenched products in the industry.

Automating the work on site. Buildots and OpenSpace turn helmet and 360-degree cameras into a continuous comparison of what was built against what was planned; Trunk Tools gives the field an agent that has read every document on the project. Robots arrive in two forms: retrofit autonomy for existing machines, as Bedrock does for excavators, and purpose-built robots for repetitive trades, Monumental laying brick, Dusty printing layouts on the slab, Built driving solar piles. Reframe Systems and ICON go further and move the work itself into robotic factories and 3D printers.

Unique and Experienced Talent on a Mission to Disrupt AEC

The most telling signal is who is founding these companies. Motif's founders led Autodesk's product and platform organizations. Augmenta's CEO helped build Autodesk's generative-design tools. Hypar's co-founder created Dynamo, the visual programming tool Autodesk built into Revit. Two of Higharc's co-founders came from Autodesk research and engineering, Skema's CTO from Revit, and Swapp's founders sold their previous company to Autodesk. The people who understand the incumbent's architecture best are betting that the next platform will be built outside it.

Incumbents Are Not Standing Still

Autodesk, a $7.2 billion-revenue company, previewed an agentic Autodesk Assistant across its clouds in September. Procore unveiled its Helix intelligence layer and agreed to acquire DroneDeploy. Trimble announced the acquisition of Document Crunch; Caterpillar bought Skycatch. Consolidation will continue.

Incumbents typically hold the strongest control points in this industry: the design file, the project record, decades of workflow lock-in, and distribution into every large contractor and design firm. That is real, and it is why earlier waves produced so few new platforms. This time the ground under those control points is shifting. The new entrants are built around models rather than documents and have no license base to protect, so automation is their product rather than a threat to it. Some of them sell into engineering and construction budgets, a far larger pool than software budgets. And the data they generate, coordinated designs, as-built reality, decisions and outcomes, compounds into a different kind of moat than a file format. We believe that leaves real room to build large companies with durable moats alongside the incumbents.

Looking Ahead

AEC is entering its defining technology cycle: the problems are old, the market is enormous, the labor gap is structural, and for the first time the technology is good enough. But technology alone will not decide the winners. This is a fragmented, project-based industry with cautious buyers, and the companies that break out will be the ones that crack go-to-market. That is exactly where we focus our work with founders at Greenfield.

Our landscape maps more than 70 innovators today; a decade from now, several of them will be the way the built world gets built. If you are building in this space, we would love to connect with you.