FieldAI's map-free robot software draws a $10 billion term sheet

Engineers often have to hand a robot a map of the site it will work on, and a construction site keeps changing after that map is drawn ^[5]. Drawing the map is slow work, and slower the larger the area ^[5]. On a project plan the mapping pass looks like a setup cost paid once. On a site that changes regularly, the static map drifts into what SiliconANGLE calls potentially significant accuracy gaps ^[5].

FieldAI says its models skip that step. They also work without GPS, an internet connection or travel paths a user has set ^[6]. The company says the models adjust to risk as well, for example slowing a robot in a factory section where the lights are failing ^[8]. SiliconANGLE's account says this significantly cuts the work of setting up a robot's navigation stack and lowers the cost of automation ^[16]. Neither report includes a setup time or a dollar saving.

The thing being pitched is a "universal general-purpose brain" for humanoids, robot dogs, drones and industrial rovers, from a company that builds none of them ^[3]. The thing being done is narrower. In March FieldAI added support for Boston Dynamics' four-legged Spot so industrial customers can automate equipment inspection ^[9]. It also turns robots' camera, lidar and radar data into continuously updated digital twins, and customers use those twins to test site changes before making them ^[10].

The valuation is harder to tie to the mapless feature. FieldAI's $10 billion is about 74 times its $135 million figure ^[13]. Skild AI, valued at $14 billion ^[4], reported $100 million in run-rate revenue last month, about 140 times ^[14], on what The Next Web calls a stricter measure ^[11]. FieldAI's figure counts "revenue and in customer contracts" together, according to SiliconANGLE ^[2], so the two multiples do not compare cleanly. Physical Intelligence, valued near $11 billion, is the third of the US robot-software companies that The Next Web puts at roughly $35 billion combined ^[4].

For the operator deciding whether to pay for this, two questions sort the sites: how often the ground changes, and how large it is ^[5]. A small site that rarely changes can live with a mapped robot, and the mapless claim buys little there. A large site that changes often needs the most re-mapping, so it gains the most if FieldAI's claim holds. The other two quadrants come down to the vendor's price. The forcing number is the hours a crew spent on its last re-map multiplied by the re-maps the project schedule implies, and it caps what mapless navigation can save on that site.

I think the case is strongest on construction and energy sites, the large and shifting ground where FieldAI already has customers ^[2]. The tradeoff is that a buyer there takes the setup savings on FieldAI's word until its own crew logs the hours.