The $100B Drafting Job That Autodesk Has No Incentive to Automate

01 /The Concept

A LiDAR scanner captures a mile of road in a matter of hours. Converting that scan into CAD deliverables takes a drafter about a week. The data is already digital. The features are already there, at centimeter resolution, in the point cloud. Getting them onto a drawing still requires a human to trace each one by hand.

Civil infrastructure CAD has looked roughly the same since the 1990s. A survey crew captures a LiDAR point cloud of a road corridor – dense, centimeter-accurate data describing every physical feature of the site. A drafter then manually traces each feature – curbs, utility fixtures, painted lane markings, drainage inlets – into the deliverables that engineers, permitting agencies, and construction teams use downstream. Autodesk and Bentley Systems charge by the seat. Their business model runs on drafters spending time inside the software. A tool that eliminates that time doesn't improve their product – it cannibalizes the behavior their revenue depends on.

Mach9, founded in 2021 by Alexander Baikovitz, Haowen Shi, Michael Mong, and Zachary Sussman – roboticists who met at Carnegie Mellon and spent years building hardware surveying systems before pivoting to software – automates the extraction step. Digital Surveyor, the company's flagship, ingests a LiDAR scan and generates a first-draft extraction using AI models calibrated to surveying-grade centimeter accuracy. The engineer reviews, edits, and refines. The output flows into whatever CAD environment the team already uses.

That framing matters. Asking a civil engineering firm to change its CAD environment means accepting liability risk, retraining cost, and workflow disruption across every active project. Mach9 doesn't ask for that. The AI step is invisible to the downstream project.

Usage-based pricing – per mile of road or acre surveyed rather than per seat – removes the ceiling on productivity gain. Nearly 100 customers are in private beta across more than 2,500 projects. The company claims 10x faster workflows at one-fifth the cost of conventional extraction.

Mach9 raised a $34 million Series A in October 2026, co-led by First Round Capital and Bessemer Venture Partners, bringing total funding to $49 million.

02 /Why It Matters

Survey deliverables are the measurement layer that every downstream phase of an infrastructure project depends on. Errors at extraction don't stay local – they cascade. A mislocated curb affects drainage design. An incorrectly placed utility coordinate delays permits. In federally funded projects, measurement errors trigger compliance reviews that stall work for months.

This is why the accuracy threshold in civil engineering is different from most software categories. Centimeter precision isn't a feature request; it's a qualification floor. Mach9's founders spent years in 3D sensor data before they understood the failure modes well enough to train models that clear the bar professional surveyors require.

The competitive dynamics of the incumbent market run deep. AutoCAD is specified in project contracts. It's taught in civil engineering programs. Bentley MicroStation is embedded in government transportation procurement at levels a startup can't easily circumvent. But neither product takes a LiDAR scan and produces a draft extraction – that step has never been part of what they sell, because their business models don't benefit from building it.

The less obvious argument for Mach9's durability is data. Every project that runs through the platform produces a paired dataset: raw scan alongside survey-grade CAD deliverables, validated by a licensed engineer. Building that dataset requires capture hardware and professional validation. At scale – thousands of projects – it becomes a training corpus that a new entrant would need years to reproduce. Each additional project doesn't just add a customer; it widens the gap between Mach9's models and any future competitor starting from scratch. Mach9 acknowledges this directly: surveys stay in the platform as a growing system of record. That's not just a product feature; it's a compounding moat.

Adjacent bets in the same infrastructure digitization space include CivilGrid (subsurface asset mapping from scan data) and Blueprints AI (permit-ready architectural drawings). Both suggest that infrastructure design is splintering into AI-native point solutions across the project lifecycle before any platform consolidates them.

03 /Opportunities

The drafting bottleneck is the clearest current instance of a pattern worth recognizing: incumbent software vendors in precision-dependent professional workflows have structural reasons not to automate the labor their pricing models depend on. Where that constraint exists, the barrier to a new entrant isn't technical – it's the switching cost that keeps buyers from testing alternatives.

Mach9's approach to the switching cost problem is worth studying. It doesn't compete with AutoCAD; it feeds AutoCAD. The AI extraction layer produces a draft that flows into the tool clients already pay for, and the value is captured before the incumbent software is even opened. This design – AI that feeds into rather than replaces an entrenched professional tool – is applicable anywhere the workflow input is a complex sensor dataset and the output is a structured professional deliverable.

Structural engineering, geotechnical reporting, and environmental site assessment all share this shape. The accuracy requirements are similar; the incumbent tooling is similarly unchanged; AI-native competition is similarly thin. Each market is smaller than civil CAD but also less defended.

The data moat argument applies across all of them. In any domain where AI-generated outputs must be validated by a licensed professional before use, the accumulation of raw input plus verified professional output becomes a training asset that compounds. The first mover that reaches meaningful project volume builds a dataset that later entrants can't buy.

The actionable direction: find a precision professional workflow where incumbents have priced themselves out of the automation opportunity, the output requires professional validation, and each project produces a validated training pair. That paired dataset – not the software – is what any durable business in this pattern is actually built on.