Construction / Robotics · Deep dive

Gravis Robotics

Four-year-old Zurich autonomy company that spun out of Marco Hutter's ETH Robotic Systems Lab (the HEAP walking excavator, the In-Situ Fabricator) and, on August 17, 2026, announced a $200M all-SoftBank Series A at a reported $1B post-money — the largest Series A in construction-robotics history — to sell a machine-agnostic retrofit kit (Gravis Rack) and operator co-pilot (Gravis Copilot) that turn Caterpillar, Volvo, Develon, Hitachi, JCB and John Deere excavators into autonomous or semi-autonomous machines.

emerging

The question that decides it: Does a machine-agnostic retrofit kit distributed through rental-fleet channels (Flannery in the UK, presumably United Rentals or Sunbelt in the US) win the autonomous-earthmoving category before Caterpillar extends its 550-truck Cat Command mining autonomy franchise into excavators, Komatsu's July 31, 2026 partnership with AIM Intelligent Machines produces a Smart-Construction-native retrofit, and Bedrock Robotics' $270M Series B (August 2026, CapitalG/Valor Atreides) locks up the largest US infrastructure GCs first — and if Gravis wins the pilot phase, does SoftBank's $200M cheque fund a genuine standalone software business, or a rich-enough asset for Volvo, HD Hyundai or Cat to acquire once the technology is proven?

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

Snapshot

Gravis Robotics is a four-year-old ETH Zurich spinout that on August 17, 2026 announced a $200M all-SoftBank Series A at a reported $1B post-money — the largest Series A in construction-robotics history per the company, and Europe’s newest robotics unicorn per Dealroom. It sells a rooftop retrofit kit called Gravis Rack (LiDAR, cameras, RTK/GNSS, edge compute) and an in-cab operator-assist product called Gravis Copilot, both built on a decade of autonomous-excavator research inside Marco Hutter’s Robotic Systems Lab — the same lab whose HEAP walking excavator built a 65-metre dry-stone wall autonomously at Oberglatt in 2022. Named deployments span Holcim quarries at Lee Moor, Taylor Woodrow’s Manchester Airport infrastructure job, HD Hyundai’s Develon and Hitachi as OEM partners, and Flannery as the UK rental-channel launch partner. The bet is machine-agnostic autonomy across a mixed fleet before Bedrock Robotics ($270M Series B in the same month), Komatsu’s newly announced AIM partnership, and Caterpillar’s mining-autonomy franchise close the window.

Founding story

The founder set is unusual. Ryan Luke Johns did undergrad at Princeton and an M.Arch at Columbia — he is an architect first — then went to ETH Zurich for a doctorate inside Hutter’s lab, where he taught a walking excavator to grab arbitrary stones and build structures from them, holding a Guinness World Record for the largest dry-stone wall built by a robot (the 65.5-metre, 938-block retaining wall at Circularity Park in Oberglatt). Dominic Jud is the hydraulics half: his PhD was the closed-loop control of HEAP itself — a 23-axis modified Menzi Muck M545 with force-controllable hydraulic cylinders in the chassis, plus the full LiDAR / IMU / GNSS / joint-sensor stack. Marco Hutter, their PhD adviser, has been building this factory for a decade: the ANYmal quadruped (ANYbotics), the In-Situ Fabricator with Gramazio Kohler, HEAP itself.

The company incorporated in 2022, sat quietly on ETH grant capital, and did not do a public priced round until November 28, 2025 — a $23M raise co-led by IQ Capital (UK) and Zacua Ventures with Pear VC, Holcim MAQER Ventures, Nesma & Partners’ Imad, Sunna Ventures and Armada Investment Group. Nine months later SoftBank wrote a $200M cheque as sole investor at a reported $1B post-money. SoftBank managing director Dai Sakata framed it publicly as central to the fund’s Physical AI thesis. That framing matters — it signals Gravis is being valued as an AI company not a construction supplier, and it is priced accordingly.

How it works

Gravis Rack is a flat rectangular appliance that bolts onto the excavator’s roof. Inside sits the edge-compute board that runs the perception and control stack; on top and on a pair of rear masts sit LiDAR sensors, cameras, RTK/GNSS antennas and Wi-Fi. The kit can run offline when site connectivity is thin. It is machine-agnostic by design: reference deployments span Caterpillar, Volvo, Develon, Hitachi, John Deere, JCB, Case, Sumitomo and Yanmar excavators — the “mixed fleet” claim is the core commercial pitch and the anti-OEM wedge.

The autonomy stack is closer to a fleet-of-narrow-behaviours model than a general jobsite AGI. Gravis publicly covers trenching, bulk excavation, truck loading, grading, stockpile management and quarry hopper feeding — repetitive, geometrically bounded tasks where the site geometry can be surveyed once and treated as a slow-updating digital twin. The company says its models are trained heavily in simulation to close the sim-to-real gap; every deployed machine also runs continuously as a site sensor, feeding back survey and hazard data. Positioning is RTK/GNSS-anchored; hydraulic joint-space control is the piece Jud brought straight from HEAP.

There are two operating modes. In Copilot mode the human operator stays in the cab with 3D guidance overlays, hazard detection and machine assistance — a Trimble-Earthworks-plus-obstacle-avoidance step-change. In full-autonomy mode the operator is out of the cab supervising multiple machines from a control desk. Reported productivity uplift versus manual operation is 30%.

Product and business overview

Two named products. Gravis Rack is the hardware-plus-autonomy retrofit — sold with installation, calibration and ongoing software updates, priced for the fully autonomous use case. Gravis Copilot, launched at CONEXPO in Las Vegas in March 2026, is the in-cab assist version — same hardware, less ambitious behaviours, priced for the much larger installed base of contractors who will not yet remove operators from cabs but will pay to keep them.

Distribution is emerging in three shapes. First, direct sales to large industrial buyers — Holcim in cement/aggregates (also an investor via MAQER Ventures), Taylor Woodrow in UK civils, Nesma in Saudi contracting (also an investor via Imad). Second, OEM co-marketing — the CONEXPO demos with Develon (HD Hyundai) and Hitachi position Gravis as a preferred autonomy partner without an exclusive lock-in on either side. Third, the rental channel — the November 2025 Flannery partnership packages a Gravis-equipped machine as a turnkey rental in the UK, which is the single most-scalable go-to-market for a retrofit kit and the piece SoftBank’s $200M will fund at scale.

Business model and pricing

Undisclosed publicly. Sacra’s August 2026 note describes a mixed gross-margin picture: hardware and installation are lower-margin, the recurring software / subscription layer is higher, and early deployments carry field-engineering cost that will amortize as channel partners take more of the burden. Built Robotics historically charged contractors an hourly fee to license Exosystem’s autonomy software — the closest US analog — and Bedrock’s operator-less-hour framing suggests the category is coalescing around per-machine-hour or per-tonne pricing. Gravis has not confirmed the primary meter. Expect a hybrid: hardware sale or lease of the Rack plus per-hour or per-machine-per-month software fee, discounted through Flannery-style rental bundles.

The unstated economics: the Copilot product is the wedge, the fully autonomous Rack is the compounding revenue, and the rental-channel bundle is what turns a $200M raise into a real installed base.

Traction over time

Two data points to flag. First, the gap between incorporation (2022) and the first priced round (Nov 2025) is unusually long — the company built product on grant capital and quiet angel money before taking a lead investor, which is why the November 2025 round was already accompanied by seven-country deployment. Second, the nine-month gap between the $23M seed and the $200M SoftBank round is very short: the round was clearly negotiated on the back of the CONEXPO-March-2026 US-expansion narrative and the Bedrock $80M-Series-A / $270M-Series-B pressure on the same category.

Market analysis

The addressable market is real and estimation is noisy. Fortune Business Insights pegs autonomous construction equipment at ~$17.7B in 2026 growing to $35.2B by 2034 (~9% CAGR); Grand View at ~$16.4B growing to $32.5B by 2033 (~10.2% CAGR); GM Insights near ~$18.2B in 2026 to $25.9B by 2030 (~9.2% CAGR). Asia-Pacific dominated 2025 share at ~42% per Grand View; North America contributed ~$3.33B (~20% share). The wider global construction equipment market is roughly $220-250B per multiple 2025 sources.

Two structural forces expand it. The US construction industry needs roughly 349,000 net new workers in 2026 per ABC’s 2026 outlook, with 40% of the current construction workforce approaching retirement; average equipment-operator age is 47, with an estimated 120,000 experienced operators exiting by 2028 per K2 Staffing. Skilled excavator operators are the scarcest of the group. On the supply side, physical-AI investment scale has stepped up — SoftBank Group’s Physical AI thesis, NVIDIA’s NVentures backing Bedrock, and the Komatsu-AIM strategic partnership together move the reference valuations by an order of magnitude.

Competitive intel

Full list in frontmatter. The competitive geometry has four faces.

The most immediate is Bedrock Robotics — same wedge, better US distribution, four ex-Waymo founders, $270M Series B in the same month as Gravis’s Series A at $1.75B (higher than Gravis’s reported $1B), with the first operator-less US infrastructure deployments announced August 17, 2026. Bedrock is not the follower; in the US it is arguably the leader. Gravis’s edge is Europe / Middle East, mixed-fleet breadth and hydraulic-autonomy heritage; Bedrock’s edge is the US GC relationships and Waymo pedigree.

The second is OEM-native. Cat MineStar Command for hauling has >550 autonomous mining trucks and >2B tonnes hauled — an incumbency proof that once the OEM ships autonomy, the retrofit market compresses. Komatsu on July 31, 2026 announced a strategic partnership with AIM Intelligent Machines for physical-AI autonomous bulldozers and excavators, already in US commercial deployment. Volvo Autonomous Solutions ships autonomous quarry haulers today. Each can bundle native autonomy into a new-machine sale.

The third is Built Robotics, whose apparent pivot toward autonomous solar-farm construction (with a reported ~114-employee WARN realignment) leaves the retrofit-excavator arena that it pioneered largely to Gravis and Bedrock — bullish for both but a warning about the pilot-to-fleet gap that killed Built’s momentum in the general market.

The fourth is the status quo: Trimble Earthworks, Leica MC1 and Topcon MC-Max machine-guidance kits sit on tens of thousands of existing excavators today at a fraction of a Gravis-Rack price. Copilot has to beat a Trimble-equipped operator by a wide enough margin to justify swap-in, or it fights Trimble for the same slot.

History and evolution

• 2015-2022 — Marco Hutter builds out ETH’s Robotic Systems Lab; ANYbotics spins out in 2016; HEAP walking-excavator programme runs 2016-2025.
• ~2018-2022 — Ryan Luke Johns’ ETH PhD on autonomous excavation with non-standard stones; Dominic Jud’s PhD on 23-axis hydraulic autonomy of HEAP.
• Nov 2022 — HEAP autonomously builds the 65m, 938-block dry-stone retaining wall at Oberglatt Circularity Park; Gravis Robotics incorporated in Zurich the same year.
• 2023-2024 — Early Holcim pilot at Lee Moor quarry, UK, under the Holcim MAQER Ventures programme.
• Nov 28, 2025 — $23M seed round co-led by IQ Capital and Zacua Ventures announced; already deployed across seven countries.
• Mar 3-8, 2026 — US commercial expansion at CONEXPO-CON/AGG Las Vegas; Gravis Copilot debut with Develon (HD Hyundai) and Hitachi partner demos.
• Jul 31, 2026 — Komatsu / AIM Intelligent Machines strategic autonomy partnership announced (competitive pressure).
• Aug 17, 2026 — SoftBank $200M Series A at reported $1B post-money; on the same day Bedrock Robotics announces its $270M Series B and first operator-less US excavator deployments.

What people say

The case for. Coverage is broadly positive. Construction Dive frames the round as validation that “physical AI is on investors’ radars” and notes BuiltWorlds’ 2026 Equipment & Robotics benchmarking data showing robotics adoption up 45% year-over-year and the pilot-only group nearly tripling. The Robot Report, Engineering News-Record and Silicon Republic all called the round the largest Series A in construction robotics history. Sifted and Dealroom framed it as Europe’s newest robotics unicorn. Sacra’s August 2026 note credits Gravis for the machine-agnostic breadth — same software stack across many machine brands and sizes without custom reprogramming — and for the tri-track expansion logic (more task types, more verticals, more geographies). Named references are unusually strong for a company this young: Holcim (also an investor), Taylor Woodrow at Manchester Airport, HD Hyundai’s Develon and Hitachi as demo partners, Flannery as the UK rental channel. Ryan Luke Johns himself is a rare founder profile — Guinness World Record for the largest robot-built dry-stone wall, architect-turned-roboticist, ETH-Hutter pedigree.

The complaints. They are structural and they are pointed. BuiltWorlds’ own 2025 benchmarking notes that “the appetite and interest in robotics has grown larger than builders’ actual usage,” and contractors have repeatedly flagged a pilot-to-fleet gap on Built Robotics’ Exosystem — the closest historical analog — that Built has not fully bridged (the reported ~114-employee WARN realignment and pivot toward solar-farm autonomy is the tell). The Cat Command franchise (550+ autonomous mining trucks, 2B+ tonnes hauled) proves OEM incumbency compounds; a retrofit kit can win the pilot and lose the fleet as OEMs bundle equivalent capability into new-machine sales at zero customer capex. The Komatsu-AIM partnership announced July 31, 2026 forecloses a major OEM channel for Gravis. Bedrock’s $270M Series B in the same month as Gravis’s Series A, at a higher $1.75B valuation, with Waymo-alumni founders and first-ever operator-less US infrastructure deployments, means Gravis just paid a $1B price to be the number-two horse in its most important market. And on the numbers: PitchBook reports ~74 employees and RocketReach ~55 as of mid-2026 — a very small team to service the mixed-fleet promise, so a large fraction of the $200M is a hiring bet against Bedrock and AIM’s US teams. No published pricing, no disclosed ARR, no published autonomous-hours figure. Every headline number is a company or investor claim.

Outlook: the open question

Answered yes if: within 18-24 months Gravis has (a) more than a dozen named, contractually recurring rental-fleet customers via Flannery-equivalent partners in the US, UK and Germany; (b) a published autonomous-hours-per-machine-per-month figure competitive with Bedrock’s and Cat Command’s; (c) a safety-certification path (CE / OSHA / ANSI-B56-adjacent) that lets contractors legally remove operators from cabs at scale; and (d) enough distribution lock-in through OEM co-marketing (Develon, Hitachi and one of Volvo/John Deere/JCB) that the retrofit model survives Cat/Komatsu shipping native autonomy. Answered no if: Bedrock signs the top-10 US infrastructure GCs on operator-less excavation before Gravis’s US commercial team is stood up; Cat announces a Cat-Command-for-excavation SKU bundled into a new-machine sale in FY27; the Komatsu-AIM partnership matures into a Komatsu-native retrofit for the installed base at Cat-Command pricing; or the promised 30% productivity uplift on controlled sites does not survive contact with a rainy day and a mixed subcontractor crew — the pilot-to-fleet gap that has bitten Built Robotics. All four risks are live today.

How to attack it

The specific wedge: build a rental-channel-native, Copilot-tier retrofit priced below the Trimble-Earthworks upgrade, sold through United Rentals, Sunbelt and Ashtead as a machine-included subscription with a fixed-price safety and remote-supervision service on top. Gravis’s own beachhead moves (Flannery in the UK, presumably a US equivalent post-Series-A) prove that rental-fleet distribution is the real go-to-market for retrofit autonomy — the rental company owns the fleet, absorbs the capex, and bills the contractor per-day for a turnkey machine. A challenger who partners rental-native from day one (i.e. designs the Rack to be installed and swapped inside a rental yard by the yard techs, not by a company field engineer) can undercut Gravis’s field-engineering cost structure by design.

The specific weaknesses to exploit:

1. Retrofit vs OEM-native structural squeeze. Cat Command (>550 mining trucks, >2B tonnes) and the Komatsu-AIM partnership (Jul 31, 2026) point one way. A challenger that goes native with a smaller OEM (Volvo CE, JCB, Case, John Deere) on a preferred-integration basis can pre-empt Gravis’s mixed-fleet claim inside that OEM’s own dealer channel.
2. No published pricing, no published autonomous-hours. Contractors can’t underwrite on a company-claim 30% uplift. A rival that publishes hours, downtime, mean-time-between-safety-events and per-tonne pricing at pilot exit wins the procurement conversation.
3. US GTM standing start. Bedrock is US-native, Waymo-pedigreed, and shipping operator-less deployments today; Gravis announced US commercial only in March 2026. Every quarter Gravis lags in the US is a Bedrock lock-in on the top infrastructure GCs.
4. Small team, big promise. ~55-74 employees carrying a mixed-fleet, seven-country, three-product roadmap. A better-focused single-vertical (quarry autonomy, or trenching autonomy) attacker can out-execute inside its slice.
5. Safety certification is unfinished. Full operator-out-of-cab operation on active construction sites has no clean regulatory rail in most Western jurisdictions. A rival that goes safety-certification-first (ISO 17757-adjacent for earthmoving autonomy) sells to insurance underwriters before contractors and locks in a compliance moat.
6. Investor over-concentration. SoftBank as sole Series A investor is idiosyncratic risk — SoftBank has cooled on portfolios before. A challenger with a diversified strategic-and-financial cap table markets stability to industrial buyers who buy for a decade.

A well-funded attacker with $80-120M, one OEM preferred-integration, and a United Rentals or Sunbelt exclusivity could catch Gravis in the US inside 24 months.

Adjacent-segment play

Same autonomy stack, different buyer. The most obvious adjacency is quarry / aggregates / cement autonomy end-to-end — Gravis already has Holcim as a customer and investor, and the quarry environment (fixed geometry, repetitive cycles, controlled access, no public safety exposure) is genuinely easier than an open construction site. A pure-play quarry-autonomy company sold to Holcim, Heidelberg, Cemex, Vulcan and Martin Marietta as a fleet-management-plus-autonomy platform is a cleaner sale than a mixed-fleet civils story and could be built on the same Rack. Holcim’s investment in Gravis is a hedge, not a preclusion — an attacker who takes Cemex or Heidelberg exclusive could carve the segment.

The second is mining haulage autonomy for the tier-2 mine operators Cat Command doesn’t serve. Cat Command’s economics work for majors (BHP, Rio Tinto, Fortescue). Mid-sized mines can’t underwrite the machine swap. A retrofit haulage stack — SafeAI is already here, thinly capitalised — priced for tier-2 gold, copper and iron-ore mines is a real segment.

The third is agriculture and forestry. Autonomous tractors (John Deere / Blue River) and forestry harvesters (Komatsu / Ponsse) are adjacent hydraulic-autonomy problems with a similar OEM landscape. The Gravis hydraulic-control stack (from HEAP) is genuinely transferable. Where the wedge does not generalize is on-highway trucking (already the crowded Aurora / Kodiak / Plus AI battleground) and mobile robotics indoors (no hydraulic advantage). And the deepest adjacency — safety certification and insurance underwriting for autonomous heavy equipment — is a wedge in its own right: the underwriter who prices operator-out-of-cab operation credibly gets to price every retrofit-autonomy provider’s insurance line, and no such player exists at scale today.

Sources and further reading

• Gravis Robotics gets $200M from SoftBank to retrofit excavators with self-driving AI systems — SiliconANGLE, Aug 17, 2026
• Autonomous equipment tech firm Gravis Robotics raises $200M — Construction Dive, Aug 17, 2026
• Gravis Robotics raises $200M for autonomous construction — The Robot Report, Aug 17, 2026
• $200M Funding Round Set for Automated Equipment Startup Gravis Robotics — Engineering News-Record, Aug 2026
• Gravis Robotics hits $1bn valuation with $200m investment from SoftBank — Sifted, Aug 2026
• Gravis Robotics analysis — Sacra, August 2026
• 200 million for autonomous excavators — ETH Zurich News, Aug 2026
• Foundation of the ETH Spin-Off Gravis Robotics — NCCR Digital Fabrication, 2022
• HEAP (2016-2025) — Robotic Systems Lab, ETH Zurich
• Autonomous excavator constructs a 6-meter-high dry-stone wall — TechXplore, Nov 2023
• Swiss construction startup Gravis Robotics raises €19 million — EU-Startups, Nov 2025
• Bedrock Robotics Raises $270 Million in Series B Funding — PRNewswire, Aug 17, 2026
• Komatsu and AIM Intelligent Machines enter Strategic Partnership — Komatsu Newsroom, Jul 31, 2026
• Caterpillar achieves 2 billion tonnes hauled with autonomous trucks system — Caterpillar, 2024
• Autonomous Construction Equipment Market Report — Fortune Business Insights, 2026