Construction eats startups for breakfast.
No sector has absorbed more venture capital with less to show for it per dollar spent. Since 2017, billions have flowed into contech. Most of it is gone. And yet right now, in early 2026, a new wave of AI-native construction companies is raising at valuations that would have been unthinkable five years ago. FieldAI just closed $405M at a $2B valuation, backed by Bezos, Nvidia, and Temasek. OpenSpace has deployed $199M and is acquiring competitors. Buildots is buying workforce platforms.
So which wave gets it right? To answer that, you have to understand what the last one got wrong.
Era One: The Ignored Industry (Pre-2017)
Before 2017, construction was a venture dead zone. The industry was fragmented, slow to adopt software, and deeply relationship-driven. GCs ran projects on spreadsheets and gut feel. Subcontractors communicated by phone. Drawings lived in truck cabs.
This wasn't ignorance. It was risk-adjusted skepticism. Construction moves in cycles tied to macro conditions, not product release schedules. Sales cycles are long. Procurement is Byzantine. The buyer (the GC) is rarely the user (the foreman). Software companies that tried to crack the industry in the 2000s largely gave up.
The two companies that didn't give up — and survived into the next era — were Procore and Autodesk. Both understood something the later wave forgot: workflow software that fits into how construction actually works beats revolutionary software that requires construction to change.
Era Two: The SoftBank Era (2017–2021)
Then Katerra happened, and everything broke loose.
Katerra was founded in 2015 with a genuinely audacious thesis: vertically integrate the entire construction value chain. Design, manufacture, assemble — all in-house, all optimized, all software-driven. SoftBank wrote an $865M check in 2018. Total raise: $1.6 billion.
The thesis wasn't stupid. Construction is extraordinarily inefficient. Labor productivity has barely moved in fifty years while manufacturing productivity has tripled. The opportunity was real.
The execution was catastrophic.
Katerra tried to reinvent everything simultaneously — design software, prefab manufacturing, site assembly, IoT, project management. They built factories before proving unit economics. They took on diverse project types (offices, hotels, apartments, single-family homes) that shared almost no prefabricated components. A wall panel optimized for a three-story walkup doesn't work in a ten-story tower. They burned through four CEOs in six years. They filed Chapter 11 in June 2021.
The lesson everyone drew was wrong. The takeaway wasn't "don't bet on construction tech." It was "don't bet on companies that try to be the entire stack before proving any single layer of it."
Meanwhile, the era produced some genuine exits. Autodesk acquired PlanGrid for $875M in late 2018 — PlanGrid had done the unglamorous work of digitizing construction drawings and making them usable on-site. Aconex, an Australian document management platform, gave the industry its first true billion-dollar exit when Oracle acquired it for $1.2B. Trimble bought e-Builder for $500M and Viewpoint for $1.2B. These weren't moonshots. They were workflow tools that construction workers actually used every day.
Procore went public in 2021 at a $7B+ valuation. It had spent fifteen years doing one thing: making it easier for GCs to manage projects, documentation, and financials in one place. It never tried to rebuild the construction industry. It just made the existing one run better.
The Pattern
Look at the winners from Era Two and a clear pattern emerges.
Every company that won was a wedge player. They found one painful, specific workflow — drawing management, document control, project financials, RFI tracking — and owned it completely before expanding. Procore started with project management. PlanGrid started with blueprints. BuildingConnected (acquired by Autodesk for $275M) started with bid management.
Every company that lost was a stack player. They tried to own the entire value chain from day one. Katerra is the canonical example, but it wasn't alone. Dozens of "construction OS" and "full-stack builder" companies raised tens of millions between 2018 and 2022 and are now either dead or limping.
The pattern holds because construction is not a software problem. It's a coordination problem at physical scale, in dynamic environments, with fragmented ownership structures, union rules, local codes, and weather. You can't software-first your way to a solution. You have to earn trust from the boots on the ground, one workflow at a time.
Era Three: AI Finds the Wedge (2022–Now)
Funding peaked in 2021–2022, then corrected hard. But the correction didn't kill the sector — it culled the bad ideas. Seed rounds in contech are now at all-time highs. Q1 2025 saw $3.55B invested in construction technology, with 55% going toward AI and robotics.
The new wave looks different from Era Two. The companies raising big are not trying to replace construction. They're trying to automate specific, painful tasks that currently eat labor and time.
FieldAI is building AI for physical robots navigating unstructured jobsite environments — no GPS, no predefined paths. That's a hard technical problem with a specific wedge: the parts of a site where you currently need a human to just walk around and observe. OpenSpace converts site video into navigable 360 maps and integrates with BIM and scheduling tools. Buildots attaches cameras to workers' hard hats and matches footage against BIM models to automate progress tracking. GreenLite applies AI to permitting review — one of the most paper-intensive, delay-prone parts of any project.
None of these companies are trying to be Katerra. Each has a specific, painful, measurable problem. Each can show ROI within weeks, not years. Each layers into existing workflows rather than replacing them.
This is the right playbook.
What This Means for Founders
If you're building in contech, the question is not "how do I digitize construction?" It's "what specific thing happens on a jobsite that takes too long, costs too much, or fails too often — and can I prove I've fixed it in 90 days or less?"
The companies that win in Era Three will be the ones that pick the right wedge:
- High frequency — something that happens on every project, not once per build
- Measurable — the ROI is provable in weeks, not a case study in two years
- Workflow-native — the tool fits into how crews already work, not how software engineers think they work
- Data-generating — every use creates proprietary data that makes the product smarter
The companies that lose will be the ones that confuse a compelling thesis with a business. Construction will absorb your thesis and hand it back to you in pieces.
Katerra had $1.6 billion and a board full of smart people. It still couldn't make a wall panel work across two building types.
Respect the industry. Find the wedge. Prove the unit before you scale the stack.
The winners always did.
References
- FieldAI raises $405M to build universal robot brains — TechCrunch
- Autodesk to acquire PlanGrid for $875M — Construction Dive
- What Happened to Katerra: How It Raised $1.6B and Still Failed — Failory
- Procore IPO S-1 Breakdown — Meritech Capital
- OpenSpace acquires Disperse amid M&A uptick — Construction Dive
- Six AI Construction Tech Startups Secure Major Funding in Early 2026 — IndexBox
- ConTech Industry Landscape and Projections 2025–2030 — PropTech Jobs
- M&A Activity Spells a New Future for the Built World — BuiltWorlds