{"slug": "startup-spotlight-verne-robotics-rents-robot-labor-by-the-hour", "title": "Startup Spotlight: Verne Robotics rents robot labor by the hour", "summary": "Verne Robotics, founded by Neil Nie and Aditya Jha, is building an hourly robotics-as-a-service model around its dual-arm Nemo robot, which learns packaging and fulfillment workflows from demonstrations instead of requiring months of custom integration. Y Combinator lists the San Francisco company as an active Summer 2025 batch member with seven people, and its profile says Verne Robotics has worked with a biotech unicorn, a biotech nonprofit and a direct-to-consumer apparel brand. The founders interviewed with Y Combinator for the Summer 2023 batch, were rejected, launched three additional failed products, applied four times and interviewed three times before joining the Summer 2025 batch.", "body_md": "# Startup Spotlight: Verne Robotics rents robot labor by the hour\n\n**Verne Robotics, founded by Neil Nie and Aditya Jha, pairs dual-arm hardware, diffusion policies and field data to automate packaging without a large upfront robot purchase.**\n\n        By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)\n        · Published \n\nPrimary source: [Y Combinator](https://www.ycombinator.com/companies/verne-robotics)\n\n## Why it matters\n\nVerne Robotics is testing whether rapid robot training can support an hourly service model instead of a large automation purchase. That shifts deployment risk toward the robotics vendor and makes uptime, intervention rates and utilization central to the business.\n\n[Neil Nie (@neil_nie_)](https://x.com/neil_nie_?ref=runtimewire) and [Aditya Jha (@nolitadrip)](https://x.com/nolitadrip?ref=runtimewire) are building [Verne Robotics](https://vernerobotics.com/?ref=runtimewire) around a practical test for the latest wave of robot learning: whether a small machine can be taught useful work quickly enough that a customer can rent its labor by the hour.\n\nThe San Francisco founders are starting with packaging, fulfillment and other back-of-house jobs that demand repetitive handling without the uniformity of a traditional assembly line. Verne Robotics' dual-arm Nemo robot is designed to move into an existing facility, learn the customer's workflow from demonstrations and begin operating without months of custom integration.\n\nThat deployment model is the core of the pitch. Robot foundation models draw attention for generality, yet buyers still have to make the system work inside a particular warehouse, beside a particular table, handling a shifting mix of products. Verne Robotics is betting that owning the hardware, data collection, model training and field deployment can compress that last mile enough to make robotics-as-a-service economically viable.\n\n[Y Combinator currently lists](https://www.ycombinator.com/companies/verne-robotics?ref=runtimewire) Verne Robotics as an active Summer 2025 company with seven people. The accelerator profile says Verne Robotics has worked with a biotech unicorn, a biotech nonprofit and a direct-to-consumer apparel brand. Most performance and customer figures remain claims from Verne Robotics, with one named deployment offering the clearest view of what the system does in practice.\n\n### Four applications before Verne\n\nNie and Jha met in high school while tinkering with robots, according to their [original Y Combinator launch](https://www.ycombinator.com/launches/O3T-verne-robotics-robot-arms-that-learn-new-skills-in-hours?ref=runtimewire). Their path to Verne Robotics included several attempts at building something else.\n\nIn a [founder account published by Verne Robotics](https://www.linkedin.com/posts/vernerobotics_story-time-may-10th-2023-it-was-our-first-activity-7353085800160350211-U-7i?ref=runtimewire), Nie wrote that the pair first interviewed with Y Combinator for its Summer 2023 batch while working on a consumer app. Y Combinator passed. Nie and Jha subsequently launched three additional products that failed, applied four times and interviewed three times.\n\nThe repeated rejections forced a narrower answer to a basic founder question: what problem would they keep working on after the current product failed? For Nie, the answer was robot learning. He returned to robotics and AI research at Stanford before the pair applied again with Verne Robotics and joined the Summer 2025 batch.\n\nNie brought a research-heavy route into the business. Verne Robotics' [founder page](https://www.vernerobotics.com/company?ref=runtimewire) says he studied at Columbia and Stanford, working with Shuran Song and Jiajun Wu and participating in Stanford's Vision and Learning Lab, led by Fei-Fei Li. His Columbia work included [research on robots learning the articulated structure of objects through interaction](https://sfa.cs.columbia.edu/?ref=runtimewire), such as discovering how drawers, doors and scissors move.\n\nNie also worked on perception and interaction at Apple and helped develop Apple Vision Pro, according to Verne Robotics and Y Combinator. He left a computer science PhD program at UC Berkeley to build Verne Robotics.\n\nJha supplies the product counterpart. Verne Robotics says he graduated from Cornell and worked on Microsoft's product organization, where he led Azure Copilot from private preview through general availability. The pairing gives Verne Robotics a founder responsible for robot-learning research and another who has already managed an AI product through the less glamorous process of turning a preview into something customers can use.\n\nTheir earlier failures matter because Verne Robotics requires restraint. A small robotics developer can spend years chasing a general-purpose machine without finding a workflow that pays for it. Nie and Jha instead started with contained, repetitive jobs where labor costs and throughput can be measured.\n\n### Teaching Nemo a job\n\nVerne Robotics introduced Nemo3 publicly in July 2025 as a bimanual robot for tasks such as vial packaging, kitting, parcel sorting and meal-kit assembly. In that launch, Verne Robotics said Nemo3 could handle objects within a 2.5-foot radius, carry payloads of up to four pounds and fit into different work environments.\n\nThe training process began with about 30 minutes of teleoperation data, according to Verne Robotics. Its software divided the demonstrated workflow into skills and used diffusion models to learn the associated actions.\n\nVerne Robotics' current materials refer to the product as [Nemo](https://www.vernerobotics.com/solutions?ref=runtimewire), a mobile, dual-arm system available with 700-millimeter or 1-meter arms. The shorter configuration is intended for compact spaces, while the longer version extends the robot's reach. Listed workflows now include high-mix bin picking, contract packaging, mixed-product kitting and apparel-return processing.\n\nThe physical configuration supports the commercial argument. A reconfigurable robot can move between workflows and facilities, spreading hardware costs across several jobs. Rapid training matters because every day spent collecting data and tuning a deployment adds engineering expense before the robot produces anything billable.\n\nVerne Robotics has yet to publish a controlled benchmark for its central claim that Nemo learns jobs in hours. The public materials do not provide task success rates, intervention frequency, pick speed or uptime. The phrase covers a wide range of possible outcomes, from learning a narrow sequence in a prepared workspace to handling a variable workflow across full shifts.\n\nThat gap does not erase the deployment evidence, but it sets the next standard Verne Robotics must meet. Industrial buyers will judge Nemo on completed units, error recovery, support costs and output per paid hour. A fast training session is valuable only when the resulting policy keeps working after the founders leave the building.\n\n### ABClonal is a production customer and case study\n\nVerne Robotics identifies [ABClonal](https://www.ycombinator.com/companies/verne-robotics?ref=runtimewire), a supplier of research reagents, as a production customer and case study. Nemo packages chemical reagents and ice packs for cold-chain order fulfillment, taking over work that would otherwise occupy laboratory personnel.\n\nVerne Robotics says the [ABClonal deployment](https://www.vernerobotics.com/results?ref=runtimewire) went from setup to operation in four days. Verne Robotics also reports that the robot gives ABClonal employees back five to 10 hours per week and saves $1,750 per month. Those figures come from Verne Robotics' case study rather than independent operational data, and the case study does not specify the number of robots, operating hours or period over which savings were measured.\n\nThe workflow is still a useful choice. Reagent packaging involves delicate containers, cold-chain materials and repeated manipulation, while remaining bounded enough for an early deployment. Employees can spend recovered time on laboratory work such as pipetting, aliquoting and experiments, according to the case study.\n\nVerne Robotics has continued testing outside biotech. In a [2026 field update](https://www.linkedin.com/posts/neilnie_one-of-the-hardest-problems-in-robotics-is-activity-7465382604406427648-gr2M?ref=runtimewire), Nie said Verne Robotics collected a few hours of data at Southern California customer sites before a robot began autonomously packing boxes of fabric bolts. The post did not identify the customer or publish production metrics, so the account is evidence of field activity rather than proof of scaled commercial operation.\n\n### Hourly pricing changes who carries the risk\n\nTraditional industrial automation often asks the buyer to fund hardware, integration and facility changes before savings arrive. Verne Robotics says its faster training process supports an hourly payment model, shifting part of the deployment risk back to Verne Robotics.\n\nThat alignment can be powerful. A customer purchasing robot hours can compare the expense directly with the cost and availability of manual labor. Verne Robotics has a reason to minimize setup time, interventions and downtime because an idle robot produces no useful work for the customer.\n\nHourly pricing also places a demanding burden on Verne Robotics. Hardware depreciation, maintenance, onsite support, compute and human intervention all sit behind the hourly rate. Verne Robotics has not published that rate, minimum commitments or service guarantees. The viability of the model will depend on utilization: the same robot needs to spend enough time doing paid work to cover the engineering and hardware wrapped around each deployment.\n\nVerne Robotics appears to be building the data infrastructure for that operating model. A current [robotics data infrastructure role](https://www.ycombinator.com/companies/verne-robotics/jobs/9SnPqFO-robotics-data-infrastructure-engineer?ref=runtimewire) covers telemetry, images, video, force data, annotations and cloud-to-edge pipelines. A separate robot-operator position focuses on collecting demonstrations with bimanual systems. Together, the roles show how much human and software machinery still sits behind a robot that appears autonomous on a customer floor.\n\nNie has described production deployments as a learning loop. Mistakes, interventions and edge cases provide training data for task-specific policies and Verne Robotics' broader foundation models. In August 2026, [Together AI said](https://www.linkedin.com/posts/neilnie_we-train-our-own-foundation-models-at-verne-activity-7498787807906181120-NZjN?ref=runtimewire) Verne Robotics used a Y Combinator compute cluster to train a model three times larger than its previous model on almost 100 times as much data. Neither Together AI nor Verne Robotics published resulting accuracy or production-performance improvements.\n\n### A crowded race to close the deployment gap\n\nVerne Robotics is entering a market with well-funded research labs and established warehouse-automation vendors attacking adjacent problems.\n\n[Physical Intelligence](https://www.pi.website/?ref=runtimewire) develops generalist robot foundation models trained across different tasks and robot types. Its April 2026 pi0.7 release focused on following new instructions and combining learned skills in unfamiliar environments. [Dexterity](https://www.dexterity.ai/?ref=runtimewire) sells physical AI systems for logistics and says its robots have made more than 100 million autonomous production decisions. [Nomagic](https://nomagic.ai/news/zalando-to-install-up-to-50-ai-powered-nomagic-robots/?ref=runtimewire) is expanding a Zalando deployment to as many as 50 systems after a pilot that Nomagic says reached 100,000 picks per day.\n\nVerne Robotics' wedge is a smaller, vertically integrated system that Nie and Jha say can be adapted onsite and sold as an operating expense. The approach is aimed at jobs too variable for fixed automation and too narrow to justify a large custom robotics project.\n\nFunding records also describe a business at an early stage. [VCBacked estimates](https://www.vcbacked.co/company/verne-robotics?ref=runtimewire) that Verne Robotics raised a $500,000 pre-seed round in September 2025 and names Y Combinator, Pioneer Fund and Scale Asia Ventures as backers. [ZAKA VC](https://zaka.vc/?ref=runtimewire) and Crucible Capital have separately identified Verne Robotics as a portfolio company. Verne Robotics has not confirmed the database estimate, a valuation or complete round terms.\n\nThe capital question will become harder as deployments multiply. Robotics-as-a-service can lower the customer's upfront expense while requiring Verne Robotics to finance machines, replacement parts and support before recurring payments cover the investment. Better models help the economics only if they reduce setup labor and keep each robot productive across several workflows.\n\nNie and Jha have chosen a credible route through that problem: put robots into paying environments early, learn from the failures and price the result around work performed. Their first four attempts at starting a business did not survive. Verne Robotics emerged when they stopped treating the product as the objective and returned to the technical problem Nie was prepared to spend years solving.\n\nNemo now has to prove that the founders' speed carries from a four-day setup into months of reliable operation. If it does, Verne Robotics will have built something industrial robotics has repeatedly promised and rarely delivered to smaller operators: automation that arrives on the existing floor, learns the actual job and earns its keep one shift at a time.", "url": "https://wpnews.pro/news/startup-spotlight-verne-robotics-rents-robot-labor-by-the-hour", "canonical_source": "https://runtimewire.com/article/startup-spotlight-verne-robotics-rents-robot-labor-by-the-hour", "published_at": "2026-09-21 16:09:44+00:00", "updated_at": "2026-09-21 16:32:32.696488+00:00", "lang": "en", "topics": ["robotics", "ai-startups", "machine-learning", "ai-products"], "entities": ["Verne Robotics", "Neil Nie", "Aditya Jha", "Y Combinator", "Nemo", "Shuran Song", "Jiajun Wu", "Fei-Fei Li"], "alternates": {"html": "https://wpnews.pro/news/startup-spotlight-verne-robotics-rents-robot-labor-by-the-hour", "markdown": "https://wpnews.pro/news/startup-spotlight-verne-robotics-rents-robot-labor-by-the-hour.md", "text": "https://wpnews.pro/news/startup-spotlight-verne-robotics-rents-robot-labor-by-the-hour.txt", "jsonld": "https://wpnews.pro/news/startup-spotlight-verne-robotics-rents-robot-labor-by-the-hour.jsonld"}}