On a 130-acre patch of desert outside Phoenix, an excavator is moving dirt without a human in the cab. The site, a future manufacturing facility being built by Sundt Construction, is the proving ground for what Bedrock Robotics calls the industry's largest known supervised autonomy deployment for mass excavation. The company says the machines have already moved more than 65,000 cubic yards there [Equipment World, 2026].
That is the bet, in physical form. Bedrock, founded in 2024 and headquartered in San Francisco, is not building a new excavator. It is building a kit of sensors, compute, and software that turns existing heavy equipment into autonomous machines [Bedrock Robotics, 2026]. The wedge is mass earthmoving, the repetitive, skill-intensive work of pushing dirt around a job site, which Bedrock argues should free skilled operators for more complex tasks [The Robot Report, 2026]. The customer is the general contractor. The pitch is throughput on a labor-constrained site.
Investors have written large checks against that pitch. In February 2026, Bedrock disclosed a $270 million round led by CapitalG and the Valor Atreides AI Fund [The New York Times, Feb 2026]. That followed an $80 million seed and Series A in July 2025 led by Eclipse and 8VC [RoboticsTomorrow, Jul 2025]. The cap table also includes Emergence Capital, Two Sigma Ventures, Xora, C4 Ventures, and Valor Equity Partners. For a two-year-old hardware-software company, the pace is aggressive.
Seed / Series A (Jul 2025) | 80 | $M
Series B (Feb 2026) | 270 | $M
Why the bet could be big
Construction is one of the largest sectors of the global economy and one of the least automated. Skilled operator shortages are a recurring complaint from contractors, and earthmoving is the kind of bounded, outdoor, GPS-friendly task that maps reasonably well to autonomy stacks built for highways. CapitalG, Alphabet's growth fund, and Valor have backed the thesis that the cost curve on lidar, compute, and perception software has finally bent far enough to make a retrofit kit economical on a piece of equipment that already costs hundreds of thousands of dollars [The New York Times, Feb 2026]. The retrofit angle matters: Bedrock does not have to convince a contractor to replace a fleet, only to upgrade it.
The Sundt deployment is the early evidence. Equipment World reported the Phoenix project as a supervised autonomy run, meaning humans are still on site overseeing the machines, but the work itself is being executed by the software [Equipment World, 2026]. Engineering News-Record covered the same milestone, framing it as one of the more ambitious live-site autonomy programs in the industry [Engineering News-Record, 2026]. If Bedrock can move from supervised autonomy to lights-out earthmoving on commercial schedules, the addressable market expands quickly from pilot sites to multi-billion-dollar infrastructure programs.
The team
The founding team is the reason the rounds came together so fast. CEO and co-founder Boris Sofman previously ran Waymo's trucking division and earlier founded the consumer robotics company Anki, with a robotics PhD from Carnegie Mellon [Forbes, Jul 2025] [Joel on the Beach Podcast, 2026]. CTO and co-founder Kevin Peterson and VP of Engineering Ajay Gummalla are also Waymo veterans; Gummalla spent seven years as a director of systems and programs at Waymo and 13 years before that as a hardware tech lead at Google [Equipment World, 2025]. Co-founder Tom Eliaz, also a VP of Engineering, helped scale Segment's engineering through its $3.2 billion acquisition by Twilio [C4 Ventures, 2026]. The operational bench has filled out around them: Laurent Hautefeuille as COO, Bryan Hammes as VP of Finance, Steve Lin as Head of Operations, and Matthieu Guilbert as Head of Robotics [Bedrock Robotics, 2026].
That is a lineup built for the part of the problem that breaks most autonomy startups, which is not the demo but the deployment: parts, service, uptime, and customer success on machines that work in the dirt for ten hours a day.
The honest counterfactual
Bears will point at Built Robotics and SafeAI, both of which have been chasing autonomous heavy equipment for years and have not yet produced a category-defining commercial winner. Construction sites are unstructured, weather-exposed environments where edge cases are constant, and the path from supervised autonomy to unsupervised commercial work is exactly where prior efforts have stalled. The bull answer, as Bedrock's investors have framed it, is timing and team: a perception and planning stack built by people who shipped Waymo's trucking program, applied to a narrower task (mass excavation) on a customer site (Sundt's Phoenix project) that already has measurable throughput [Equipment World, 2026] [The New York Times, Feb 2026]. The bet is that the autonomy stack is finally good enough, and that retrofit economics beat purpose-built robots.
What to watch
The next 12 months will turn on three things. First, whether the Phoenix deployment with Sundt converts from supervised autonomy to a commercial contract with measurable cost-per-cubic-yard numbers. Second, whether Bedrock signs a second and third named general contractor, which would suggest the kit generalizes beyond a friendly pilot. Third, whether the company starts building a service and parts footprint in the regions where its customers actually dig, because autonomy uptime in construction is a logistics problem as much as a software one. The $270 million gives Bedrock several years of runway to answer those questions [Construction Dive, Feb 2026].
The investors have priced in a winner. The dirt in Phoenix will say whether they are right.