SoftBank bets $200 million on robots that drive diggers

Key points

• SoftBank invested $200 million in Gravis Robotics' Series A round in 2025, which the company claims is the largest in construction robotics history.
• The Gravis Rack, a bolt-on hardware and software kit, has been installed on excavators from ten manufacturers including Caterpillar, John Deere, Volvo and Hitachi.
• Gravis says its system improves productivity by up to 30% compared with a skilled human operator at peak performance.
• The company, founded in 2022 as a spin-out of ETH Zurich university in Switzerland, has already deployed its technology across four continents.
• A separate $8 million contract with the UK government will put autonomous excavators to work on Britain's $716 billion infrastructure pipeline.

A digger does not steer around the world. It tears it apart. That basic fact is why construction sites have resisted the wave of automation that swept through warehouses, car factories and delivery fleets over the last decade.

Gravis Robotics, a Zurich-based startup that teaches excavators to operate themselves, thinks it has cracked the problem. And SoftBank, the Japanese investment giant behind some of the biggest technology bets of the past twenty years, just backed that belief with $200 million.

What does Gravis actually make?

The core product is a retrofit kit called the Gravis Rack, a combination of sensors, computers and control hardware that bolts onto excavators and other heavy earthmoving machines that a contractor already owns. Once fitted, the machine can work under three modes: a human operator stays in the cab and receives real-time 3D guidance, the operator hands control to the AI entirely, or anything in between.

The kit plugs into machines from at least ten different brands. Caterpillar, Case, John Deere, JCB, Hitachi, Volvo and others are all on the list. That matters because most construction companies own a mixed fleet, and nobody wants to scrap working equipment just to switch software providers.

When full autonomy is switched on, one person can supervise an entire fleet of machines at once instead of sitting inside a single cab.

How does it learn to handle unpredictable ground?

Teaching a robot to drive a car is hard enough. Teaching one to dig is harder, because the ground fights back differently every few centimetres, shifting from soft clay to rock-packed soil without warning.

Gravis trained its software in simulation first, moving what it describes as billions of cubic yards of virtual material before the system ever touched real earth. The idea is to close the "sim-to-real gap", the difference between a tidy digital practice world and the chaotic reality of a live jobsite. Skilled human operators read the machine through vibration and engine noise; the Gravis AI reads the same signals through the machine's own sensors at speeds no human can match.

The company reports productivity gains of up to 30% compared with a peak-performing human operator, though independent verification of that figure has not been published.

What does the SoftBank money pay for?

Gravis plans to hire engineers, expand to more jobsites worldwide and accelerate the rollout of its kit. The company also recently won an $8 million UK government contract to supply autonomous excavators to Flannery Plant Hire, Britain's largest equipment rental provider, as part of a national infrastructure push.

For construction workers, the near-term picture is probably less about replacement and more about supervision. Someone still needs to watch the machines, read the site and make judgment calls. For everyone else, faster, cheaper infrastructure projects are the potential upside, whether that means new homes, upgraded power grids or the data centres that AI companies are racing to build.