SoftBank Group spent eight years building and then dismantling its stake in the robots most people picture when they hear the company's name. Twelve days after selling its final slice of Boston Dynamics to Hyundai, the Japanese investor is reportedly circling a very different kind of machine: an excavator that can dig a trench with no one in the cab.
According to people familiar with the discussions cited by Bloomberg, SoftBank is in talks to acquire Gravis Robotics AG, a Zurich-based startup building autonomous hardware and software for excavators and other heavy construction equipment. A deal could value the four-year-old company at more than $500 million — roughly ¥82 billion — though people close to the matter cautioned that no final decision has been made on size or structure, and the transaction may be completed in stages through secondary share purchases and fresh capital. SoftBank and Gravis both declined to comment publicly on the reports, which surfaced July 28.
From dry-stone walls to a $500M price tag
Gravis was spun out of ETH Zurich in 2022 by Ryan Luke Johns, Dominic Jud, Marco Tranzatto and Burak Çizmeci, working alongside their lab mentor Marco Hutter, the ETH professor who directs the university's Robotic Systems Lab. The company's pitch is deceptively simple: rather than build a robot from scratch, it turns machines contractors already own into autonomous ones.
Its technology takes the form of modular retrofit kits that bolt onto standard earthmoving equipment, fusing LiDAR, cameras, GNSS positioning and hydraulic sensors so an ordinary excavator can perceive its surroundings in 3D and directly control its own hydraulics — trenching, grading and managing stockpiles with minimal human input. Operators supervise the work through a tablet interface that switches between fully autonomous and augmented-manual modes. The founding team's academic pedigree is not just marketing: an ETH group tied to the same lab holds a Guinness World Record for the largest dry-stone wall built by a robot, a six-metre-high, 65-metre-long structure assembled by an autonomous excavator.
The startup raised $23 million in November 2025, in a round led by Zacua Ventures and IQ Capital, with backing from Armada Investment Group and Swiss building-materials giant Holcim. Gravis says it is now operating across seven countries spanning the UK, EU, US, Latin America and Asia — among the widest footprints in autonomous excavation.
"The fastest path to autonomy is delivering productivity today," chief executive and co-founder Ryan Luke Johns said in announcing that round. "By giving operators real-time 3D intelligence and the ability to shift seamlessly between autonomy and augmented control, we cover more of the work, accelerate adoption, and create the data pipeline needed to learn new capabilities from the industry's hardest jobs."
Why this matters: Son's pivot to unglamorous robots
The Gravis talks are the clearest signal yet of how Masayoshi Son is reshaping SoftBank's robotics bet. On July 16, the group sold its remaining 9.65% stake in Boston Dynamics to Hyundai Motor Group for about $325 million, exercising a put option baked into the 2021 sale agreement. That closed the book on SoftBank's investment in the humanoid and quadruped robots — Atlas, Spot, and the earlier consumer flop Pepper — that were built for spectacle as much as utility.
What replaces them is markedly less photogenic. Since 2025, SoftBank has folded roughly 20 robotics holdings, including Agility Robotics, Berkshire Grey, Skild AI and AutoStore, into a holding company called Robo HD, and agreed to buy ABB's industrial robotics division for $5.4 billion. Gravis would reportedly sit not in Robo HD but inside Roze, a new US-based entity that Bloomberg reports SoftBank eventually intends to list. The through-line is a wager on "physical AI" — the idea that the next frontier for artificial intelligence is not chatbots but machines that move dirt, stack pallets and weld chassis. Autonomous earthmoving targets a real and worsening pain point: chronic labor shortages on construction, mining and agriculture sites that struggle to fill seats in the cab.
Not everyone is convinced the math works. World Construction Today reported that market observers are questioning whether $500 million is justified for a company this young with a single disclosed venture round — though no analyst has attached a name to that skepticism on the record. Integration risk looms too, given how differently ABB's structured factory-floor robotics behave from Gravis's unstructured outdoor autonomy. Construction and mining have been among the slowest industries to adopt robotics precisely because muddy, ever-changing job sites are far harder to automate than an assembly line.
What to watch next
Watch for confirmation of deal terms and whether the transaction is staged or clean, since a phased structure would let SoftBank hedge on the contested valuation. Watch, too, for how Gravis is positioned alongside the ABB robotics business under Roze ahead of any public listing. The deeper test is the one Son cannot buy his way past: whether physical AI performs as reliably on a rain-soaked job site as it does on a temperature-controlled data-center floor.
"The fastest path to autonomy is delivering productivity today. By giving operators real-time 3D intelligence and the ability to shift seamlessly between autonomy and augmented control, we cover more of the work and accelerate adoption."- Ryan Luke Johns, CEO and co-founder, Gravis Robotics