Energy / Construction robotics · Deep dive

Gritt

San Francisco physical-AI startup that bolts Kawasaki arms onto rented skid steers to place utility-scale solar panels — an eight-person crew goes from 800 to 3,000-4,000 panels a day — out of stealth July 2026 with a $26M Series A and 2.8 GW under contract.

emerging

The question that decides it: Gritt's wedge is placement only — its AI drives rented skid steers and off-the-shelf Kawasaki arms to set panels that human crews still fasten, on racks humans still built, over posts humans still drilled. Does retrofitted generalizable AI let Gritt absorb fastening, racking and piling fast enough to own the install stack, or does the placement-only scope cap its take per watt while vertically integrated rivals (AES's in-house Maximo, Terabase's Terafab line) and post-2025 US tax-credit phase-outs shrink the market it's racing to automate?

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

Snapshot

Gritt is a San Francisco robotics startup, founded in 2022 by two Carnegie Mellon-trained roboticists, that came out of stealth on July 21, 2026 with a $26 million Series A led by Obvious Ventures. Rather than building robots, it retrofits rented skid steers and off-the-shelf Kawasaki industrial arms with its own AI models, turning commodity construction equipment into autonomous solar-panel placement systems. The company claims an eight-person crew that manually installs about 800 utility-scale panels a day can place 3,000 to 4,000 alongside its machines — a 4-5x multiplier — and says it is contracted to support 2.8 gigawatts of installation over 18 months for customers including three of the top ten US power construction companies (all company-reported, July 2026). It matters now because the US solar build-out is labor-constrained — a projected 53,000-worker gap against 2026 targets (pv magazine, April 2026) — and because a crowded field from AES’s in-house Maximo to SoftBank-backed Terabase is racing at the same work with opposite hardware philosophies.

Founding story

Puneet Puri (CEO) and Vishal Dugar (CTO) came up through Carnegie Mellon’s Robotics Institute. Per company and press accounts (July 2026), Puri built early inspection drones for industrial systems and added a Stanford GSB stint; Dugar worked on autonomous trucking fleets and self-piloting aircraft — outdoor autonomy problems that cannot assume a structured environment. That shared scar tissue is the founding logic: industrial robots have historically failed on construction sites precisely because sites are chaotic, and the founders’ 2022 bet was that modern AI models had finally changed that equation. Puri’s stated thesis to TechCrunch (July 2026): construction-speed automation requires intelligence generalizable across varied outdoor environments, not a bespoke machine per task. Andrew Beebe of Obvious Ventures, who led the Series A, drew the cultural contrast explicitly — not rocket-budget perfectionists, but people who scale dirty, dull, dangerous jobs. The company spent roughly four years in stealth, shipping systems onto live sites before announcing anything.

How it works

The mechanics are the differentiator, so be precise about them. Gritt does not manufacture a robot. It takes equipment already on every utility-scale solar site — rented skid steers and telehandler-class machines — and mounts off-the-shelf industrial robotic arms (Kawasaki is the named supplier, July 2026) on them. Gritt’s proprietary layer is the AI: perception and control models that let the assembly autonomously pick large glass modules off delivery stacks, carry them across rough, open terrain, and position them on the mounting racks with what the company calls sub-millimeter accuracy. Human workers then fasten the panels. That last clause is load-bearing: as of July 2026 Gritt’s robots place; people still bolt, and people still built the racks and drilled the posts beforehand.

The claimed advantage of retrofit over purpose-built is threefold. Capital: no factory, no fleet on the balance sheet — base vehicles rented, arms catalog items. Deployment speed: going from 2 systems to 48 within six months (target stated July 2026) is only plausible because the hardware is procurable, not manufactured. Generalization: the founders say the training pipeline that took weeks to learn cinder-block stacking later learned a rebar-tying demo in a day (TechCrunch, July 2026) — each new task gets cheaper, so the platform can crawl from placement to fastening, post-drilling, rack assembly and eventually non-solar work. The two deployed systems collect field data continuously. Company-reported record: tens of thousands of panels placed autonomously, zero breakages (July 2026) — meaningful, since a cracked module is pure loss, but unaudited.

Product and business overview

What Gritt sells is three layers. First, the placement system: retrofitted vehicle-plus-arm units doing the pick-carry-place cycle alongside a reduced human crew. Second, the AI platform — the actual company — a task-training pipeline the founders claim generalizes across manipulation jobs; solar placement is the wedge, with fastening, post drilling, rack building and rebar tying named as the roadmap (July 2026). Third, an aspirational site-intelligence layer: Puri describes the sensor-laden systems flagging open trenches ahead of storms or missing inventory — physical AI as jobsite management, not just labor (TechCrunch, July 2026). Only the first layer generates revenue today. Customers at stealth exit: three unnamed top-ten US power construction companies, plus the 2.8 GW of contracted support — roughly 4-6% of a year of US utility-scale additions at the EIA’s 44-GW 2026 forecast. Real, not rounding error, but far from ubiquity.

Business model and pricing

Here the public record thins out, and that itself is a finding. Gritt has published no pricing: no per-panel rate, no per-watt fee, no day-rate, no revenue-share structure. The shape of the disclosed contracts — “supporting installation of 2.8 GW” — plus the rented-equipment model strongly implies service-style revenue (robots-as-a-service or per-project contracting) rather than hardware sales, and coverage (TheNextWeb, briefs.co, July 2026) frames the model as customers not needing to buy equipment. But no outlet has printed the number that matters: what Gritt charges per installed watt versus the labor it displaces. Until it discloses take-per-watt, margin, or even a contract value, its unit economics are unverifiable — noted because the asset-light story (rented skidders, catalog arms) is only cheaper if the AI’s reliability premium prices above the rental and integration costs it stacks on top.

Traction over time

Two honest caveats. Every number in the table is company-reported as of the July 21, 2026 announcement; stealth means no independent time series exists. And the implied leap — 2 systems to 48 in six months, a 24x scale-up — is the steepest ramp anyone in this sector has attempted; Terabase and AES took years to reach fleet operation. The 2.8 GW is contracted support, not completed work, with terms (exclusivity, minimums, cancellation) undisclosed.

Market analysis

The structural story is strong. The US solar industry employed over 280,000 workers but needs roughly 355,000 by late 2026 to hit 60-70 GW annual deployment — a 53,000-worker gap — and 27% of utility-scale firms call installation hiring “very difficult” (pv magazine / IREC-derived data, April 2026). The EIA forecasts 44+ GW of new utility-scale solar in 2026. Panel installation is also an injury-and-turnover machine: repeated overhead lifting of ~100-pound modules in desert heat, at sites deliberately located where nobody lives. That is the labor pool Gritt monetizes.

The countervailing force is policy. The July 2025 US budget law phased out federal solar tax credits for projects placed in service after 2027 unless construction begins by mid-2026 — pulling a surge of starts forward (part of why 2026 targets are so labor-strained) while clouding demand beyond 2028. Gritt’s near-term market is inflated by exactly the deadline rush that may deflate its medium-term one — presumably why the roadmap leans so hard on “then, everything else.” The global build-out (Australia’s A$100M Solar Scaleup Challenge, 2025; China’s module-maker robots) keeps the placement market international even if US policy wobbles.

Competitive intel

The frontmatter carries the full set; the analytical shape is this. Gritt is the only notable player betting on retrofit of rented equipment — everyone else builds hardware. AES’s Maximo is furthest along in deployment (100 MW installed at Bellefield by March 2026, four-robot fleet, over a module a minute) and represents the worst structural outcome for vendors: developers automating in-house. Terabase ($200M raised through January 2025, SoftBank-led) sells a whole field-factory workflow, out-funding Gritt six-to-one. Charge Robotics ($39M, SOLV Energy-validated) automates tracker-section assembly upstream of placement. Cosmic Robotics ($4M seed, April 2025) is the direct placement rival with purpose-built hardware. Luminous (ARENA-backed, on Neoen and Engie farms in Australia, 2025-26) and Trina’s Trinabot show the model working abroad. Built Robotics ($112M) proved the retrofit thesis in pile driving. Gritt’s edge: the fastest capital-light scaling path and a generalization story none of the single-task machines have. Its exposure: placement is the most contested step in the stack, its 4-5x claim must beat AES’s and Cosmic’s in the bake-offs EPCs will inevitably run, and it holds no hardware moat by design — Kawasaki arms are available to everyone.

History and evolution

• 2022 — Founded in San Francisco by Puri and Dugar; enters stealth.
• ~2023-2024 — Seed round (undisclosed size and date) from First Round Capital, Climactic, Congruent Ventures, VSC Ventures.
• 2024-2026 (stealth) — Builds retrofit systems on rented skid steers with Kawasaki arms; deploys two systems to live utility-scale sites; places tens of thousands of panels, zero reported breakages; trains side demos (cinder-block stacking in weeks, rebar tying in a day) to prove generalization.
• Jul 21, 2026 — Exits stealth: $26M Series A led by Obvious Ventures with Union Square Ventures and Active Impact Investments; total funding reported as $32M (TheNextWeb), $32.4M (citybiz), and $34M (TechCrunch) — the outlets disagree, and the company has not published a reconciled figure. Discloses 2.8 GW contracted, three top-10 US power construction customers, and the 48-systems-in-six-months target.
• Jul 27, 2026 — Appears in Bricks & Bytes’ contech funding roundup as the week’s headline robotics raise.

No pivots, layoffs or public stumbles are on record — with the strong caveat that four of the company’s four years were spent in stealth, so the absence of visible history is structural, not evidential.

What people say

The case for. The one named external voice is investor Andrew Beebe (Obvious Ventures, July 2026), who frames the founders as rare operators combining AI and machine-vision depth with an appetite for dirty, dangerous, scalable work. TechCrunch’s Tim Fernholz spoke to one actual Gritt customer — anonymous, citing competitive reasons — who was enthusiastic on two specific grounds: remote sites where workers are nearly impossible to attract become buildable, and injuries fall because crews stop lifting 100-pound panels overhead all day (July 2026). Trade aggregators (Bricks & Bytes, July 27, 2026) flagged the 4x install-rate claim as the week’s standout. The customer roster itself — three top-10 US power construction companies signing before the company had a public name — is the strongest third-party endorsement available.

The complaints. There are no customer reviews, no G2 or Glassdoor presence, no substantive Reddit or forum threads — the company was in stealth until nine days before this page’s publish date; that absence is the documented state of the record, not an oversight. The criticism that exists is structural: every performance number (800-to-4,000 panels, zero breakages, 2.8 GW) is company-reported and unaudited; the total-funding figure itself differs by $2 million across TechCrunch, TheNextWeb and citybiz; coverage (briefs.co, July 2026) notes no outlet obtained installed cost per watt or bad-weather uptime; and the robots do not fasten panels, so the “installation” multiplier measures placement, one step of the job. The sector’s own reference point — AES’s “half the time, half the cost” Maximo claim — has never been independently verified either: self-graded homework is the industry pattern. Sector skeptics (Renewable Energy Magazine, May 2025) add the workforce question: solar installer is among the fastest-growing US jobs, and displacing its entry-level rung has labor-relations consequences no vendor has navigated publicly.

Outlook: the open question

Gritt works if two things prove true by early 2027: the 48-system fleet actually materializes with utilization on the contracted 2.8 GW, and the task stack expands beyond placement — fastening or post-drilling in production, not demo. If both land, the retrofit thesis is vindicated: capital-light scaling beats purpose-built fleets, the AI layer becomes the moat, and Gritt is positioned to follow Built Robotics’ path up the workflow with far better generalization economics — with the rebar and general-construction roadmap giving it an escape hatch from any post-2027 US solar demand cliff after the tax-credit phase-out (enacted July 2025).

It stalls if placement-only scope meets fleet-scaling friction. The bear case is concrete: a 24x fleet ramp in six months requires field reliability, operator training and EPC integration that two deployed systems cannot yet evidence; placement alone may command too thin a slice of EPC budgets to support venture-scale revenue against AES automating in-house and Terabase selling the whole workflow; and with no proprietary hardware, a rival — or Kawasaki itself — can replicate the form factor, leaving defensibility resting entirely on model quality and field data from a fleet not yet built. Watch, over 12-18 months: the actual system count (48 by January 2027 is falsifiable), any named customer or audited throughput figure, a disclosed price per watt, the first production task beyond placement, and whether 2.8 GW contracted converts to megawatts completed.

How a challenger would attack it

Attack the seam Gritt left open: fastening. Gritt’s robots place panels; humans still bolt them, on racks humans built, over posts humans drilled. A challenger that ships place-plus-fasten as one motion — even at lower daily throughput — sells a bigger labor displacement per watt and makes Gritt’s 4-5x multiplier look like what it is: one step of the job. The second vector is Gritt’s deliberate absence of a hardware moat. The form factor is a rented skid steer plus a catalog Kawasaki arm; a fast follower — or Kawasaki itself, or an equipment-rental giant bundling autonomy into the fleet EPCs already rent — can replicate the physical stack in a quarter and compete purely on model quality against a company with just two systems’ worth of field data. Third, force the bake-off: every performance number Gritt has published is self-reported and unaudited, and no outlet has extracted a price per watt or bad-weather uptime figure. A challenger that publishes audited throughput, transparent per-watt pricing, and named customers turns Gritt’s stealth-era opacity into a liability exactly when EPCs run head-to-head trials. The window is now, while Gritt is consumed by a 24x fleet ramp no one in the sector has attempted.

Same playbook, new buyer

Take retrofitted placement autonomy to the buyers the solar scrum is ignoring. The whole competitive set — AES, Terabase, Charge, Cosmic, Luminous — is piled onto US utility-scale solar, a market whose post-2027 demand is clouded by the July 2025 tax-credit phase-out. The same rented-equipment-plus-catalog-arm model ports to adjacent repetitive-lift construction: precast and tilt-up concrete, warehouse racking, modular housing panelization — sites with the identical profile of heavy overhead lifts, remote locations and unhireable labor, and no robotics vendor in sight. Gritt itself names rebar tying and “everything else” as the roadmap, but its contracts, capital and credibility are all committed to hitting 48 systems on 2.8 GW of solar by January 2027; it cannot chase a second vertical mid-ramp. Geography is the other opening: Australia is literally paying for automation (the A$100M Solar Scaleup Challenge) and the Middle East and India build utility-scale solar in labor conditions worse than Nevada’s — markets where Luminous is the only mover and Gritt, with two US systems, won’t arrive for years.

Sources and further reading

• Gritt exits stealth with $34 million for robots to build solar plants — then, everything else (TechCrunch, July 21 2026)
• Gritt raises $32M for AI robots that bolt onto existing construction equipment to build solar farms faster (TheNextWeb, July 2026)
• Gritt Raises $32.4 Million to Scale AI-Powered Robotics for Infrastructure Construction (citybiz, July 2026)
• Latest Construction Technology Funding Rounds – 27th Jul 2026 (Bricks & Bytes, July 27 2026)
• Workforce gap of 53,000 threatens 2026 solar deployment targets (pv magazine USA, April 2 2026)
• Robots install 100 MW of solar panels on 1-GW AES project (Solar Power World, March 2026)
• Terabase raises US$130 million to scale up automated project construction work (PV Tech, January 2025)
• Cosmic Robotics gets funding to automate, accelerate solar installation (The Robot Report, April 2025)
• LUMI robots ScaleUp solar deployment with advanced automation (ARENA, July 2025)
• Charge Robotics 2026 Company Profile (PitchBook, accessed July 2026)