Supply chain / Warehouse robotics · Deep dive

Gather AI

A CMU-spinout physical-AI platform that flies commodity drones through warehouses, reads pallet imagery with computer vision, and reconciles the picture against the WMS — pitched as a hardware-agnostic 'curious' AI for the 90% of DCs that never automated.

emerging

The question that decides it: Does drone-plus-computer-vision inventory reconciliation deliver enough persistent labor arbitrage to justify the hardware and services CapEx, once 3PL customers demand shared savings and the incumbent WMS suites (Manhattan, Blue Yonder, SAP EWM) bundle equivalent computer-vision reconciliation into software their operators already own?

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

Snapshot

Gather AI is a Pittsburgh physical-AI company that sells warehouse-inventory reconciliation: commodity drones and forklift-mounted cameras fly and roll through DCs, capture pallet imagery, and pass it to a computer-vision stack that reads LPNs, UPCs, lot codes and text and matches every location against the warehouse management system. Founded in 2019 as a CMU Robotics Institute spin-out by three PhDs who had built one of the first safe autonomous helicopters, it raised a $40M Series B in February 2026 led by Keith Block’s Smith Point Capital Management, bringing total funding to $74M. The bet: computer-vision inventory intelligence — not humanoids, not new ASRS — is the near-term physical-AI wedge in the 90% of warehouses that never automated.

Founding story

Sankalp Arora, Daniel Maturana and Geetesh Dubey met at the CMU Robotics Institute on a DARPA program to build a full-scale autonomous helicopter — a project that flew on FBI training grounds at Quantico, won the Howard Hughes and AUVSI Xcellence Awards, and was nominated for the Collier Trophy. Arora built the safety and sensor-planning stack; Maturana, author of VoxNet, owned perception. Teaching an aircraft to survey unfamiliar terrain safely is structurally the same problem a warehouse drone faces at smaller scale.

They spun out in 2019 with a deliberately narrow first act: use off-the-shelf commercial drones instead of building aircraft, and solve the boring, high-frequency problem of pallet-location accuracy inside 3PL and manufacturing DCs. That framing still shapes the company — hardware treated as replaceable, perception + orchestration software as the durable asset. Arora repeats the “curious drones” line in every interview: the robot doesn’t just execute a script, it queries the environment for what the operator asked for and adapts.

Two details matter. The 2019 start meant three years of angel and CMU-adjacent seed capital before institutional venture arrived in the October 2022 Series A. And in June 2023 Gather AI acquired the assets of Ware Robotics, its closest US software-focused drone rival, taking customer count to a stated 25 and roughly doubling headcount — a consolidation that has since defined its US posture.

How it works

The physical loop is three steps. A commercial off-the-shelf drone — Gather AI supports multiple platforms and released a “Starling 2 Logis” reference drone in 2025 — takes off from a dock, follows a predetermined aisle route, and localizes itself against ArUco fiducial markers (printed barcode-like patterns) affixed to racking or floors. As it flies, it photographs every pallet location, streaming imagery back. A single drone scans up to 900 pallets per hour, roughly 15x faster than a manual cycle-count operator. Aisles do not shut down; staff stay 15 feet from the drone but keep working the rest of the aisle.

The intelligence loop is where the actual product lives. The computer-vision stack reads multiple barcodes, LPNs, UPCs, lot codes and free-form text out of each image, infers case counts for full and partial pallets, detects empty slots, and computes location-occupancy percentage. Results are time- and location-stamped in a dashboard that surfaces three buckets: matches with the WMS, empty locations, and exceptions requiring human review. Operators triage exceptions there — the human never opens the drone footage until the software flags it — and push corrections back into the WMS, MES or ERP.

The 2026 evolution — Gather AI 2.0, funded by the Series B — decouples the stack from drones entirely. The same pipeline runs on forklift-mounted cameras, fixed cameras and worker wearables, feeding a continuously updated “digital twin” rather than a periodic scan. A natural-language Copilot layer in enterprise rollout for 2026 lets an operator ask, “how many pallets of SKU X are in row 34?” The drone was the on-ramp; the durable product is any-camera-anywhere perception plumbed into the WMS.

Product and business overview

Gather AI sells three interlocking things. First, the drone-based inventory scan — the original wedge, still the most-cited use case, with the clearest ROI story (company-cited 99.9% inventory accuracy, 80% reduction in manual cycle-count labor, ROI in under six months). Second, the Physical Intelligence Platform — the hardware-agnostic perception + reconciliation software, sold as a platform layer beneath any camera source. Third, adjacent SKUs: inferred case counts, location-occupancy analytics for slotting optimization, and the 2026 Copilot NL interface. Named to Fast Company’s Most Innovative Companies list for 2026, Gather AI has expanded beyond 3PLs into manufacturing, retail, aerospace and automotive. The story it wants investors to internalize is a category migration — from a drone-hardware vendor competing with Verity and Dexory, to a warehouse-intelligence software company competing with the perception features WMS incumbents will inevitably bundle.

Business model and pricing

Gather AI does not publish pricing. Third-party directories (SoftwareAdvice, Gartner Peer Insights) describe an initial setup fee plus a recurring subscription with optional premium modules; customer references frame the math as sub-six-month payback against fully loaded cycle-count labor cost. That is consistent with a per-facility subscription tiered by drones or SKU volume plus a services line for onboarding (ArUco marker installation, WMS integration, flight-path mapping). Decoupling from proprietary hardware is a margin story: if the company isn’t building drones, it isn’t carrying the industrial-manufacturing gross-margin drag Verity and Dexory do. Latka reports ~$15M ARR at ~$270M valuation as of 2026 (unverified; directional). Bookings +250% YoY implies momentum; the missing disclosures — NRR, ACV, services-vs.-software mix, gross margin — are the ones any software-vs.-hardware framing will eventually have to defend.

Traction over time

Named 2026 anchor customers include GEODIS, NFI Industries, Kwik Trip, Barrett Distribution Centers, Axon, Langham Logistics and dnata; earlier references include DHL, Metro Supply Chain and (via a 2025 partnership) Burwell. All metrics are company-supplied and unaudited; no revenue, gross margin, live-facility count or ACV has been disclosed.

Market analysis

The pain is real. Inventory inaccuracy costs the global economy an estimated $1.1 trillion annually, with $400B of that in retail lost revenue and roughly 1-3% of sales at the individual retailer (NRF and industry-analyst estimates, 2023-2025). Gather AI’s own materials frame a $150B annual loss for misplaced-pallet errors specifically. The vendor market is smaller and more contested: the global WMS market is around $4.8B in 2026, forecast to $10.9B by 2031 at ~18% CAGR (MarketsandMarkets, 2025), increasingly consolidated around Oracle, Blue Yonder, SAP, Manhattan and Tecsys (~33% share together). The drone-scanning niche where Gather AI, Verity, Dexory and Corvus play is a small fraction — likely low hundreds of millions in software + services spend today — with the addressable base expanding as the 80% of industrial facilities with no automation start to buy something. The tailwind is chronic labor shortage, projected to reach ~2 million unfilled US industrial roles by 2030 (NAM).

Competitive intel

Three tiers. First, direct drone rivals. Verity AG is the most credible global competitor: Kiva co-founder Raffaello D’Andrea, $76M raised, vertically integrated drones that fly in pitch-dark warehouses without human supervision, and a European heavy-industrial customer roster (DSV, Maersk, Ingka, Samsung SDS). Attacks on lights-out autonomy depth. Dexory is the best-funded player in the segment ($205M raised through Oct 2025) but is a 12m tall autonomous ground robot, not a drone; its US expansion is a direct threat on any RFP that prefers ground robots and the deepest capital pool. Corvus Robotics (~$28M raised) matches the drone footprint at smaller scale and pitches infrastructure-free autonomy — sharp counter-positioning against Gather AI’s fiducial-based localization.

Second, adjacent physical-automation vendors. Symbotic ($1.8B revenue 2025, >$5B backlog) sells the whole automated warehouse; where it wins, a case-handling ASRS already knows what is in every slot, shrinking greenfield for a visibility vendor.

Third, and ultimately more important: the WMS incumbents. Manhattan, Blue Yonder and SAP EWM already own the WMS. All three are racing to bundle visual cycle counting, computer-vision pick verification and camera-fed reconciliation natively. When the reconciliation target ships the reconciliation tool, the surface area for a standalone vendor tightens. Gather AI’s drone-decoupling and Copilot roadmap are, read charitably, an attempt to be indispensable at the perception layer before that window closes; read uncharitably, a bet on being acquired. The real baseline in every deal is not a rival vendor — it is Zebra handheld scanners and scheduled cycle counts: free at the margin and embedded in every WMS SOP.

History and evolution

• 2019 — Arora, Maturana and Dubey spin Gather AI out of CMU Robotics Institute; angel-funded seed period follows.
• Oct 2022 — $10M Series A led by Tribeca; first institutional round.
• Jun 2023 — Acquires assets of Ware Robotics; customer count ~25; headcount roughly doubles.
• Mar 2024 — $17M Series A-1 extension led by Bain Capital Ventures; total $34M.
• Apr 2024 — Launches inferred case count and location-occupancy analytics.
• 2025 — Launches Starling 2 Logis drone; expands verticals; deepens 3PL set (GEODIS, NFI, Barrett, Kwik Trip).
• Feb 2026 — $40M Series B led by Smith Point Capital Management with Bain, Tribeca, Bling, Dundee, XRC and new investor The Hillman Company; total $74M; positioning pivots from “drone company” to “Physical Intelligence Platform”; Fast Company Most Innovative 2026.

What people say

The case for. Customer references are consistent and specific — GEODIS, NFI, Kwik Trip and Axon cited as reference deployments hitting 99.9% inventory accuracy, 80% reduction in manual counting, payback in under six months (company / customer statements, 2025-2026). Modern Materials Handling, DC Velocity, The Robot Report and Supply Chain Management Review have covered the Ware acquisition and the hardware-agnostic pivot favorably, treating Gather AI as the consolidated US winner in software-first drone inventory. Bain Capital Ventures published a thesis backing the “commodity drone plus proprietary AI” architecture as the reason it led the 2024 extension. Glassdoor sentiment skews positive on mission, colleagues and remote-first flexibility (4.1 average, 67% recommend, small sample, 2026).

The complaints. Two threads. First, worker-facing conditions in customer warehouses: Glassdoor reviews describe being pushed to work in sub-50-degree warehouses without heating and facilities with poor air quality, together with below-market compensation, no promotion structure and limited diversity. That is not a review of the software; it is what happens when a small SaaS company sells into 3PL environments where the customer sets the physical conditions and the implementation team lives on-site. Second, and more strategic: the missing disclosures. Gather AI publishes no revenue, no live-facility count, no ACV, no NRR, no gross margin, and — despite the hardware-agnostic pivot — very little detail on services-vs.-software split. In a category where Verity is bigger in Europe and Dexory is better capitalized globally, the absence of unit-economics disclosure is a legitimate diligence gap. The ROI numbers are all company-supplied; there is no third-party audit of the 99.9% accuracy figure.

Outlook: the open question

The last five years vindicated the opening bet. Off-the-shelf drones plus a competent perception stack do reconcile inventory faster than manual cycle counts, customers pay, and the drone-vs.-drone field narrowed rather than exploded — Ware got acquired, Corvus is still small, Verity is European.

But the question in 2026 is not whether drone inventory works. It is whether customer willingness to pay for it survives two squeezes. The bull case requires three things within 24 months: (a) the 3PL customer base — GEODIS, NFI, Barrett, Kwik Trip, dnata — renews and expands rather than commoditizing the vendor as scanning becomes a cost-of-doing-business line item pushed back into shared-savings pricing; (b) the Physical Intelligence Platform genuinely lands as forklift-camera and wearable revenue at margins meaningfully above the drone-services line, giving Gather AI a moat that is not a specific piece of aerospace; and (c) Manhattan, Blue Yonder and SAP EWM remain slow enough at bundling native computer-vision reconciliation that Gather AI becomes either an entrenched platform or an attractive acquisition before their roadmaps close the gap. The bear case is the mirror: 3PL renewals hit price concessions, the digital-twin pivot lives mostly in the deck, and the WMS suites bundle visual cycle counting into software the DC already owns.

A fourth watch item is harder to price. Dexory has more than twice Gather AI’s capital and is actively selling in the US; Verity has the D’Andrea pedigree and European anchors. Gather AI is not the best-funded player in its own category, and $40M is not enough to outspend either rival on international expansion. That is why the platform pivot and Copilot roadmap are load-bearing — the story that says Gather AI is playing a different game than the two competitors that would out-capitalize it in a straight hardware race. Whether it is a category or a positioning is what the next 18 months answer.

How a challenger would attack it

Kill the fiducials, then kill the services line. Gather AI’s drones localize against ArUco markers glued to racking — an installation-and-maintenance burden Corvus already counter-positions against with infrastructure-free flight, and one that inflates the setup fee and onboarding services (marker installation, flight-path mapping, WMS integration) buried in Gather AI’s undisclosed pricing. A challenger ships marker-free SLAM autonomy with genuinely self-serve deployment: a dock, a subscription, live in a week — Verity’s one-week deployment claim, delivered at US 3PL prices. The second vector is the forklift, not the drone: Gather AI 2.0’s own pivot concedes the durable product is any-camera perception, but forklift-mounted cameras piggyback on hardware the DC already owns and scan continuously during normal work, no flight windows, no 15-foot standoff rules, no rotor downwash near pickers. A camera-only entrant with no aerospace heritage undercuts on price and skips the drone CapEx conversation entirely. Third, exploit the disclosure gap in competitive deals: no audited accuracy figure, no NRR, no live-facility count — against Dexory’s ~$205M war chest and Verity’s DSV/Maersk/IKEA roster, a challenger that publishes verified accuracy benchmarks and reference-checkable renewal data wins RFPs on trust. Gather AI’s $74M cannot fund a two-front fight on both hardware autonomy and platform software.

Same playbook, new buyer

Drone-plus-CV reconciliation has been sold almost entirely to 3PLs and manufacturing DCs — pallets in racking, matched to a WMS. The same perception loop transfers to asset-intensive environments with worse visibility and richer budgets. The sharpest shift is outdoor and yard inventory: container yards, lumber and steel stockyards, auto-auction lots, and equipment-rental fleets have no WMS-grade location data at all, and the buyer is measured on asset utilization rather than cycle-count labor — a bigger, less contested ROI. Second: cold storage, where sub-zero aisles make manual cycle counting genuinely miserable and expensive; Gather AI’s Glassdoor complaints about sub-50-degree warehouses hint at the labor economics without productizing them. Third: retail backrooms and store-level inventory, where the $400B retail-accuracy problem lives but the drone form factor and 3PL sales motion don’t fit — fixed cameras plus the same reconciliation stack, sold to retail ops. Gather AI won’t chase these soon: its Series B story, reference customers, and Copilot roadmap are all warehouse-WMS-shaped, its integration IP is racking-and-pallet specific, and with two better-funded rivals in its core segment it cannot afford to open a second front.

Sources and further reading

• TechCrunch — Gather AI, maker of ‘curious’ warehouse drones, lands $40M led by Keith Block’s firm (9 Feb 2026)
• Business Wire — Gather AI Raises $40M Led by Smith Point Capital Management (9 Feb 2026)
• Supply Chain Management Review — Gather AI’s next chapter: from drone vision to warehouse intelligence (2026)
• Automate.org — Industry Insights: Gather AI’s CEO on what it takes to make robots curious (2026)
• TechCrunch — Gather AI secures new cash to scan inventory in warehouses using drones (6 Oct 2022)
• The Robot Report — Gather AI acquires assets of Ware Robotics (Jun 2023)
• Axios Pro — Gather AI raises $17M Series A extension led by Bain Capital (27 Mar 2024)
• Bain Capital Ventures — Drone vision software for warehouse inventory (thesis post, 2024)
• Tech.eu — Swiss autonomous inventory drone maker Verity extends Series B to $43M (11 Jul 2023)
• SiliconANGLE — Autonomous warehouse robotics platform Dexory raises $80M (2 Oct 2024)
• UKTN — Warehouse intelligence group Dexory scores £123m investment (15 Oct 2025)
• The Robot Report — Corvus Robotics soars to new heights with Series A round for drone inventory (2024)
• Glassdoor — Gather AI employee reviews (accessed Aug 2026)
• MarketsandMarkets — Warehouse Management System market size, share, latest trends 2025-2030