Construction · Deep dive

Bedrock Robotics

A San Francisco startup founded by ex-Waymo autonomy leaders that bolts a brand-agnostic sensor-and-compute kit onto existing 20-to-80-ton excavators, turning contractors' own machines into supervised-autonomous, 24/7 earthmovers rather than selling them new hardware.

emerging

The question that decides it: Bedrock's whole thesis is that a brand-agnostic retrofit — an $80M-a-year engineering team's sensor kit bolted onto a contractor's existing Caterpillar, Deere or Komatsu excavator in a single shift — delivers a lower fully-loaded cost per operating hour than the factory-integrated autonomy Caterpillar (11B+ tonnes already moved by its autonomous mining fleet) and John Deere are building natively, while earthmoving sites stay chaotic enough — buried utilities, shifting terrain, weather, mixed human-and-machine workflows — that removing the cab operator is genuinely hard. **What has to be true either way:** the retrofit wins only if one remote supervisor can safely oversee many operator-less machines at a cost per cubic yard below both a human operator and an OEM's own kit, AND if dirt work turns out to be a structured-enough domain that supervised autonomy reaches true operator-less scale before an OEM bundles equivalent autonomy into the machine at the point of sale and prices the retrofit out.

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Bedrock Robotics is a two-year-old San Francisco company trying to automate earthmoving without selling anyone a new machine. Its product, the Bedrock Operator, is a sensor-and-compute retrofit kit that clamps onto a contractor’s existing 20-to-80-ton excavator — a Caterpillar, John Deere or Komatsu — in a single work shift, turning it into a supervised-autonomous digger that can run around the clock while a remote human oversees several machines at once. Founded in 2024 by four alumni of Waymo’s self-driving program (led by ex-Waymo trucking and Anki chief Boris Sofman), Bedrock emerged from stealth in July 2025 with $80M and, just seven months later in February 2026, raised a $270M Series B co-led by CapitalG and Valor’s Atreides AI Fund at a roughly $1.75B valuation — putting total funding above $350M. As of April 2026 it had “dozens” of machines on live US sites and was targeting its first fully operator-less excavator deployments during 2026. What it has not disclosed is revenue.

Founding story

The premise is a contrarian read of a decade of autonomous-vehicle disappointment: robotaxis took twenty-plus years and billions because the open road is unforgiving — pedestrians, unprotected left turns, an infinite tail of edge cases and a regulator watching every mile. A private, fenced construction site has none of that. It is chaotic in a more bounded way — varied terrain, buried utilities, dust and weather — but there is no public to endanger, no city to negotiate with, and an owner who will pay for a machine that works nights and weekends. Boris Sofman spent nearly five years at Waymo leading its autonomous-trucking program and core perception before concluding the machine-learning-first stack Waymo pioneered was, if anything, better suited to dirt than to downtown.

Sofman is not a first-time founder. He has a Carnegie Mellon robotics Ph.D. and previously co-founded and ran Anki, the consumer-robotics company behind the Cozmo and Vector toy robots, which shut down in 2019. He started Bedrock in May 2024 with three co-founders: fellow Waymo veterans Kevin Peterson (CTO) and Ajay Gummalla (VP of engineering), plus Tom Eliaz (VP of engineering), who had led teams at Segment and Twilio and brought the cloud and data-infrastructure chops for the fleet software. The team moved fast on purpose — the company has said it had something operating autonomously at its test site within roughly six months of founding — betting that the labor math would sell itself. The industry backdrop it cites: a shortage of around 500,000 construction workers, roughly 40% of the workforce expected to retire within a decade, and about 1.25 billion hours a year spent operating heavy equipment.

How it works

The Bedrock Operator is a bolt-on, not a rebuild. A sensor bar mounts on the excavator’s cab roof carrying 360-degree LiDAR, dual RTK-GPS antennas, an inertial measurement unit and eight high-definition cameras — enough for centimeter-level situational awareness of the machine and the pit around it. Inside the cab sits an industrial computer (built on NVIDIA Jetson compute) running the perception, planning and control stack, which actuates the machine through a drive-by-wire harness that moves the hydraulics — boom, stick, bucket, swing and tracks. Installation takes a few hours to a shift and requires no permanent modification, so a contractor can pull the kit and return the machine to a human operator, or resell it, without penalty. That reversibility and brand-agnosticism is the commercial wedge: a contractor upgrades the fleet it already owns rather than buying a new autonomous machine.

Operationally, a contractor uploads a grading or excavation plan to Bedrock’s Fleet Portal and geofences the work area. The autonomous excavator then digs, swings and loads human-driven dump trucks on its own, while the software classifies terrain, detects obstacles, plans the dig sequence against a digital site model, and logs productivity metrics — cycle times, cubic yards moved, fuel burned. The machine is designed to run 24/7, pausing only for fuel and maintenance. Crucially, this is supervised autonomy: a 5G or Starlink link streams telemetry and video to Bedrock’s cloud, where a remote human supervisor can monitor and take over during edge cases. The economic bet is the same one Waymo made — one supervisor amortized across many machines, no operator in any cab, eventually undercutting one operator per machine.

Product and business overview

Today Bedrock sells one thing well — autonomous retrofits for large excavators — and has telegraphed a roadmap toward a broader “site OS.” The near-term product is the Operator kit plus the cloud fleet portal that plans work, supervises machines and reports productivity. The declared expansion is from single-machine control to fleet orchestration: dozers, wheel loaders and articulated haul trucks working in a coordinated, connected fleet, which turns Bedrock from a per-machine autonomy vendor into a site-level automation layer with adjacent revenue in grading, compaction and material handling. On top of that sits a potential software/analytics tier — selling the telemetry the machines already generate back as productivity dashboards and AI scheduling, and a data flywheel in which every deployed machine’s work cycles improve the models for all of them. Management has also floated pairing autonomy with electric-drivetrain retrofits, though that is aspiration, not product.

Business model and pricing

Bedrock sells directly to construction contractors and equipment-rental companies, and combines hardware with software subscription. The public structure is two-part: a one-time hardware/installation fee for the physical kit, then an ongoing monthly subscription covering the software platform and the remote-supervision service. Hardware provides upfront cash; the subscription and supervision layer is the recurring, higher-margin annuity. Bedrock has not published per-machine prices. The nearest public benchmark comes from rival Built Robotics, which charges roughly $3,000 per machine per month for its Exosystem plus per-hour usage fees — a useful anchor for what contractors currently pay for third-party autonomy, though Bedrock’s own numbers are undisclosed. The go-to-market motion is land-and-expand: contractors typically start with one or two machines, judge the results on cost per cubic yard and uptime, then scale. The cost side that has to be managed down is real — sensor and compute hardware, cloud infrastructure, and the salaried human supervisors who are the whole point until the machines are reliable enough to thin them out.

Traction over time

Named early partners span four contractors across four states: Sundt Construction, Zachry Construction Corporation, Champion Site Prep and Capitol Aggregates, in California, Arizona, Texas and Arkansas. The most concrete proof point to date is the November 2025 job in which a Bedrock-equipped machine moved more than 65,000 cubic yards of material on a 130-acre manufacturing site under supervised autonomy. Every one of these figures should be read for what it is: pilot-scale, human-supervised, and pre the milestone that actually matters — a truly operator-less excavator running a commercial job, which the company has promised for 2026 but had not confirmed at scale as of mid-2026. No revenue has been disclosed.

Market analysis

Estimates for the autonomous construction equipment market vary widely by methodology. MarketsandMarkets, in a February 2026 update, sized it at ~$4.40B in 2024 growing to ~$9.77B by 2030 (a ~14.2% CAGR). Grand View Research is far more bullish, projecting ~$23.89B by 2030 at ~9.6% CAGR from a larger base, while Mordor Intelligence pegs ~$5.31B in 2025 rising to ~$9.49B by 2030 (~12.3% CAGR). The spread — roughly $9.5B to $24B by 2030 — reflects disagreement about scope more than about direction. The structural forces are unusually clean for a robotics category: an aggravating labor shortage (a cited ~500,000-worker gap and ~40% of the workforce nearing retirement), the ~1.25 billion hours a year spent in heavy-equipment cabs, a hyperscale construction boom (data centers, battery plants) that demands round-the-clock excavation, and financing pressure — developers paying high single-digit interest want jobs finished faster and more predictably. The counterforce is that autonomy in the physical world has repeatedly cost more and taken longer than forecast, and construction’s edge cases (buried utilities, weather, shifting soil, mixed human-machine sites) are real even if they are more bounded than city driving.

Competitive intel

The field splits into OEM-native and retrofit camps, and Bedrock fights on both fronts. Its own camp — retrofit — is led by Built Robotics, the pioneer and closest analog (~$112M raised, published ~$3,000/machine/month pricing, deep in utility-scale solar), alongside Teleo and the cautionary SafeAI, absorbed by Pronto in July 2025. The graver threat is the OEM camp: Caterpillar (11B-plus tonnes moved autonomously, Cat Command, autonomous excavators/trucks/dozers at CES 2026), John Deere (native perception kits, an autonomous dump truck) and Komatsu (IMC 3.0, a Pronto tie-up) all own the machine, the dealer network and the customer, and can offer autonomy as a factory option — the exact dynamic that could price a third-party kit out of an OEM’s installed base. Bedrock’s differentiation is a fresher ML-first stack from people who shipped Waymo, brand-agnosticism across every OEM’s fleet, and reversibility. Whether that is a durable moat or a temporary head start before the OEMs integrate equivalent autonomy is the whole question.

History and evolution

• May 2024 — Boris Sofman, Kevin Peterson, Ajay Gummalla and Tom Eliaz found Bedrock Robotics in San Francisco; the team reports autonomous operation at its test site within roughly six months.
• Jul 2025 — Emerges from stealth with $80M in combined seed (led by Eclipse) and Series A (led by 8VC), with Two Sigma Ventures, Valor Equity Partners and NVIDIA’s NVentures participating; already piloting on live sites in Arizona, Texas and Arkansas.
• Nov 2025 — Completes a large supervised-autonomy deployment on a 130-acre manufacturing site, moving 65,000-plus cubic yards.
• Feb 2026 — Raises a $270M Series B co-led by CapitalG and the Valor Atreides AI Fund at ~$1.75B post-money; total funding passes $350M.
• Apr 2026 — Reports “dozens” of machines on active US sites and sets a 2026 goal of first fully operator-less excavator deployments.
• Jul 2026 — Roughly 56 employees with ~32 open roles across San Francisco and New York.

What people say

The case for. The bull case is pedigree plus timing. Investors — CapitalG (Alphabet’s growth fund), Valor, NVIDIA’s NVentures, 8VC and Eclipse — bet a $1.75B valuation on a two-year-old company largely because the team actually shipped autonomy at Waymo and Anki, and because the problem is well-posed: a genuine labor shortage, a fenced private domain without public-road liability, and customers who value uptime above all. CapitalG frames construction as a better-structured autonomy problem than city streets, and several investors describe being converted skeptics. The retrofit angle is the crowd-pleaser — upgrade the fleet you already own, in a shift, reversibly — and Sofman argues the economics are compelling because faster, more predictable earthmoving ripples through a project’s financing costs.

The complaints and the skeptic case. The valuation is the loudest objection: ~$1.75B for a company with dozens of supervised machines, no operator-less commercial deployment yet confirmed at scale, and no disclosed revenue is priced for a future that autonomous-vehicle history says arrives late and over budget. The technical skeptics point at exactly the edge cases Bedrock waves away — buried utilities, changing weather, unstable soil and mixed human-machine sites where one safety incident during an early deployment could freeze the whole category’s adoption. Sacra’s own risk section flags regulatory and union barriers, DOT and procurement processes that favor established OEMs, and the OEM-competition risk directly: if Caterpillar or Komatsu accelerate or acquire a retrofit specialist, they can undercut Bedrock with integrated hardware-software. The SafeAI-into-Pronto absorption is a live example that standalone retrofit economics are not guaranteed. And there is the structural question no press release answers: is a retrofit team, however good, going to out-engineer Caterpillar’s 30 years of autonomy data and dealer network on the very machines Caterpillar builds?

Outlook: the open question

Bedrock is the best-funded, best-pedigreed entrant in a category that is suddenly crowded and suddenly credible — and it is being asked to prove two hard things at once. What would have to be true for it to work: first, the retrofit has to win on unit economics — one remote supervisor overseeing many operator-less machines has to drive cost per cubic yard below both a human operator in the cab and an OEM’s factory-integrated kit, and stay there as the OEMs cut prices to defend their installed base. Second, earthmoving has to turn out to be a structured-enough domain that supervised autonomy actually crosses into reliable operator-less operation — not just impressive supervised demos like the 65,000-cubic-yard job, but machines running commercial dirt work with the human moved fully out of the loop and thinned across a fleet. If both hold, the reversible upgrade-your-own-fleet pitch is a genuine wedge into a multibillion-dollar market with a real labor crisis behind it, and the team compounds a data flywheel the OEMs will struggle to match on software. What would sink it: driverless dirt slips the way driverless roads did — the edge cases prove less bounded than the pitch, a safety incident chills adoption, or Caterpillar and John Deere simply integrate equivalent autonomy at the point of sale and price the retrofit out of their own machines, relegating Bedrock (and Built, and Teleo) to the aftermarket long tail while SafeAI-style consolidation thins the field. The evidence to watch is narrow and concrete: whether the first truly operator-less excavator deployment lands in 2026 as promised, the machine count and repeat-expansion within the named contractors, any disclosed cost-per-hour or cost-per-cubic-yard figures, and the first move by an OEM to bundle or acquire. Supervised demos are cheap. Operator-less economics, at scale, against Caterpillar, are the whole game.

How a challenger would attack it

Out-focus it, out-price it, or become the OEM’s weapon against it. Bedrock’s $1.75B mark prices a general-purpose earthmoving platform, which forces it to chase breadth — excavators now, dozers, loaders and haul trucks on the roadmap — before it has confirmed a single operator-less commercial deployment. The Built Robotics playbook shows the counter: pick one repeatable, high-volume vertical (utility-scale solar trenching and piling, data-center pad prep, pipeline right-of-way) where the dirt work is genuinely structured, and grind to operator-less reliability there while Bedrock’s supervised fleet stays spread across four contractors in four states. Second, price transparency: Built publishes ~$3,000/machine/month; Bedrock publishes nothing. A challenger with open per-cubic-yard pricing and a performance guarantee turns every pilot bake-off into a procurement comparison Bedrock’s undisclosed economics lose by default. Third — the sharpest vector — partner with the incumbent rather than fight it: Komatsu already chose Pronto, and SafeAI’s absorption shows retrofit independents get bought or squeezed. A challenger that licenses its stack to Deere or Cat as the factory autonomy option inherits the dealer network, financing and installed base that Bedrock must sell around machine by machine. And Bedrock’s remote-supervision layer is a standing cost until machines are reliable; a rival that thins supervisors faster wins the only unit economic that matters.

Same playbook, new buyer

Bolt-on supervised autonomy fits any fenced site with repetitive cycles and a labor gap. The nearest transfers are aggregate quarries and sand-and-gravel pits — more repetitive than construction sites, already semi-automated in mining, but below the ticket size Cat’s mining division bothers with — plus ports and intermodal yards (Teleo’s terminal-tractor wedge), landfill compaction, and agricultural earthworks. Each is a bounded private domain with the same 24/7 uptime buyer and fewer buried-utility edge cases than civil construction. The buyer shift with the best economics is the equipment-rental channel: United Rentals or Sunbelt offering “autonomous excavator by the week” turns autonomy into a rental line item, reaches thousands of contractors who will never sign a Bedrock subscription, and matches the reversible-kit design perfectly — yet serving rental fleets at scale means field service, uptime SLAs and pricing discipline that a 56-person, R&D-weighted startup cannot staff while chasing its promised 2026 operator-less milestone. Geographically, Australia’s mining-adjacent civil sector and the Gulf’s mega-projects have acute operator shortages and money; Bedrock’s US-only deployment footprint and its need to prove the core thesis at home leave those markets to whoever moves first.

Sources and further reading

• Bedrock Robotics Raises $270 Million in Series B Funding (PR Newswire, Feb 2026)
• Bedrock Robotics’ $270M Series B paves the way for operator-less excavators (The Robot Report, Feb/Mar 2026)
• Ex-Waymo engineers launch Bedrock Robotics with $80M to automate construction (TechCrunch, Jul 2025)
• Bedrock Robotics funding, news & analysis (Sacra, 2026)
• Interview with Bedrock Robotics CEO Boris Sofman: ‘The economics are compelling’ (Robotics & Automation News, Oct 2025)
• Bedrock Robotics: Autonomy for the Built World (CapitalG, 2026)
• Breaking Ground with Bedrock (Eclipse Ventures, 2025)
• Built Robotics raises another $64M to make construction equipment autonomous (TechCrunch, Apr 2022)
• Caterpillar rolls out autonomous excavators, trucks, dozers (Interesting Engineering, 2026)
• Autonomous Construction Equipment Market worth $9.77 billion by 2030 (MarketsandMarkets, Feb 2026)
• Ex-Waymo engineers launch Bedrock Robotics (discussion) (Hacker News, Jul 2025)