The Excavator With a Second Job

Gravis Robotics, a Zurich-based construction robotics company, has raised $200 million from SoftBank in a Series A round the company describes as the largest ever for a construction robotics startup. The money funds a retrofit system that turns existing excavators and other heavy machinery into supervised, semi-autonomous equipment, letting operators step out of the cab and manage fleets remotely.

That's the headline. The quieter detail is what the machines do while they're working. Every Gravis-equipped machine can simultaneously act as a site sensor, collecting data for automated surveying and hazard mapping while construction work is being performed. Call it moonlighting machinery — equipment that clocks in for one job and quietly does a second, uncontracted one on the side.

For anyone who runs field programs — utility mapping crews, telecom infrastructure audits, asset inventory teams — this isn't a construction story. It's a preview. The tools doing the physical work are starting to double as the tools that document it, and that changes how you budget a field program before it changes anything else.

A Machine That Digs and Documents

Gravis was founded in 2022 as a spin-out of the Swiss Federal Institute of Technology Zurich, building software-defined systems for heavy excavators. Its core product, the Gravis Rack, is a bolt-on control system rather than a new machine. It's a hardware-agnostic control system that can retrofit machinery from manufacturers including Caterpillar, John Deere, JCB, Hitachi and Volvo for operator-assisted or autonomous operation. Contractors keep their existing fleets; they just add the autonomy layer.

The scale of the new funding is what got attention. SoftBank is investing $200 million in Gravis Robotics — as the largest Series A in construction robotics history, accelerating the company's mission in physical AI. The company says it has already deployed its systems across four continents and proven them across diverse real-world machinery and sites. Part of that deployment includes a government-backed CAM Pathfinder project with Flannery Plant Hire, the country's largest equipment rental provider in the U.K. — the kind of infrastructure asset inventory work that field teams recognize instantly, just automated at the machine level instead of run as a separate survey pass.

The Second Job Nobody Negotiated

Moonlighting machinery refers to equipment whose primary contracted task is physical work — digging, grading, lifting — but which produces a secondary stream of usable field data without a separate crew, sensor deployment, or site visit. That secondary output was never in the original spec. It shows up anyway, because the machine is already moving through the site with sensors on board.

Consider a highway-widening project where an excavator, retrofitted with autonomy software, logs its position, bucket load, and terrain profile every few seconds while moving earth. A survey crew would ordinarily walk that same ground separately, on its own schedule, with its own line item. Here, the positional and condition data arrives as a side effect of the actual construction, not as an added task competing for budget and daylight hours.

Field Data Collection Has Always Had a Staffing Problem

Anyone who has run a utility mapping program, a telecom infrastructure audit, or an asset inventory sweep knows the real cost isn't the analysis. It's getting a qualified person to the right location with the right equipment, repeatedly, across hundreds or thousands of sites. That's the bottleneck moonlighting machinery is quietly attacking from a different angle: fold the sensing into equipment that's on-site for another reason anyway.

It's the same logic behind Agtonomy's passive-collection updates to its off-road platform — machines doing agricultural work while logging field conditions nobody had to dispatch a separate team to capture. Gravis applies the same principle to earthmoving. The industries look different. The staffing math is identical.

Except the Data Still Needs a Home

Except moonlighting has a catch.

Raw sensor streams from a fleet of excavators aren't automatically insight. They're volume. Someone still has to structure the position logs, tag the hazard flags, reconcile the terrain scans against a known baseline, and decide what counts as signal versus noise from a machine that was never designed as a survey instrument in the first place. That translation step — from incidental capture to usable field record — is the part that the translation tax exists to describe: the hidden cost of turning raw operational exhaust into something a client or engineer can actually use.

Collection was never the hard part. It rarely is.

Somebody Still Verifies the Ground Truth

The productivity numbers attached to this round are worth a second look before anyone treats moonlighting data as automatically trustworthy. Gravis says its systems can improve jobsite productivity by up to 30% compared with peak manual operation in some applications. That's a real claim, and it might hold up. But those figures are based on Gravis's own deployments and have not been independently verified.

That's not a knock on Gravis specifically. It's the standing caution for any field program that starts generating data as a side effect rather than a designed measurement. Volume isn't verification. A machine that logs its own position a thousand times a day still needs someone checking that the position, the timestamp, and the hazard flag actually mean what the dashboard says they mean — the same discipline behind closing the desk distance between assumption and ground truth in any research program, physical or otherwise.

The Bigger Bet Behind the Round

Global infrastructure is in the middle of a massive expansion, and across every sector, construction has become the primary bottleneck. That shortfall stems from a severe labor crisis, and heavy construction remains one of the least automated major industries in the world, still largely run on machines that operate the way they did fifty years ago. SoftBank isn't just betting on faster digging. It's betting that a machine already doing physical labor can absorb a second job — documentation — without anyone having to staff for it separately.

That's a bet worth watching outside construction, too. If an excavator can survey the ground it's disturbing, why should a retail audit only capture one data type per store visit? Why should a telecom infrastructure check require a separate mapping pass when the primary inspection could log location and condition automatically? Why treat incidental data collection as a bonus instead of designing for it from the start?

None of those questions have tidy answers yet. But the direction is clear enough: the equipment doing the work and the equipment documenting the work are converging, one retrofit and one funding round at a time.

Start a Project

TeamPL runs field data collection programs — egocentric video, in-store audits, mystery visits, GIS surveys — designed around a specific research question, not a generic panel. If this topic touches a program you're planning, tell us what you're trying to learn.

Every machine has one job on paper. The interesting ones are already working a second.

Frequently Asked Questions