The $1/Hour Robot Is Coming: Four Industry Leaders Explain What’s Next
In a Nutshell
Four robotics leaders agree that 2025-2028 will mark the shift from research prototypes to commercially deployed robots in industrial inspection, warehouses, and homes. The key technical bet is that human-form robots (1X Neo, Agility Digit) will unlock massive internet-video training data and teleoperation scaling, while Boston Dynamics Spot and Anyotics ENM prove today’s quadrupeds already deliver ROI via 8-20× daily automated inspections that replace dangerous, dull work. Pricing models converge on labor-cost benchmarks ($1-3/hour target) with robot-as-a-service subscriptions, and all four companies explicitly reject military weaponization while warning that U.S. and allied nations must secure domestic manufacturing to prevent Chinese dominance.
These notes were generated by AI and may contain inaccuracies.
The video is recorded at the Machina conference in Paris, France, focused on AI in the real world.
AppLoven started with an $8 domain and no VC funding, becoming one of the largest ad platforms globally. The same engine now powers AppL ads for e-commerce. Ads run inside mobile games reaching over a billion people with full-screen, distraction-free attention. The platform finds buyers and optimizes for profit. One cookware brand scaled from $4 million to $16 million, turned profitable, and is on pace for $80 million this year. Visit apploven.com/allin to launch campaigns.
Dr. Peter Funkhouser is co-founder and CEO of Anyotics, which produces the ENM robot. He has worked in the robotics space for nearly 20 years. The company spent its first five years as a research lab.
The robot is described as a "dog-based robot" or "dogobot." The four-legged form factor became standard because four legs provide superior mobility, balance, stability, and dexterity. Four legs enable climbing stairs and navigating environments where humans can go. The wide footprint and multiple footholds provide stability on slippery floors, rain, snow, and grass.
Humanoid robots (Optimus, Neo, Atlas) use two legs for working at eye level and in narrower spaces like coffee shops. Quadruped robots are preferred for facilities with sufficient space where stability is paramount. The speaker jokes that cafes should use "centaurs" as the new standard.
Anyotics has deployed hundreds of robots over five years for critical infrastructure inspection. The robots cost low hundreds of thousands of dollars to purchase, with tens of thousands annually in service contracts. They are not for home use but for industrial applications.
The robots carry extensive sensors including thermal cameras, acoustic microphones, and gas concentration detectors. They detect micro gas leakages, temperature anomalies, and equipment overheating beyond human perception. The monetary benefit comes from avoiding downtime, as critical assets can lose hundreds of thousands of dollars per hour when stopped.
The robots operate for one to two hours per mission before returning to docking stations for charging. Some customers run 40 missions per day. They do not hot-swap batteries. The focus is on hands-free autonomy where customers want data and insights rather than managing robots.
Real-time processing for obstacle avoidance and data quality must occur on the robot due to connectivity concerns. Cloud computing handles contextual analysis and historic downtime analysis. Maximum operational time is in the eight-hour range with charging included. Customers have shifted from manual inspections once or twice daily to eight, fifteen, or twenty times daily.
Offshore deployments on oil rigs and wind energy facilities are particularly valuable because helicopter flights cost tens of thousands of dollars. The robots operate on transformer stations where windmills connect. They function in extreme temperatures from -20°C in Norway to +40-60°C in deserts, and in explosive atmospheres where sparks are prohibited.
The company receives requests for military applications but is not pursuing them. Military requirements differ significantly: millisecond response times, remote control with human-in-the-loop, different communications, and different autonomy needs. The company maintains a non-military mission focus.
Zero percent of the robot is sourced from China. Components come from the US, Europe, and local sourcing. The company builds significant portions in-house. China produces robots that walk beautifully and perform backflips but lack autonomy, inspection intelligence, and workflow integration. Data security concerns include ISO certification for cybersecurity and avoiding control of critical infrastructure by external parties.
NATO members are increasing defense spending. The speaker personally believes Europe has a responsibility to build its own military technologies. However, Anyotics is not entering this space due to different product requirements and team specialization needs.
Bert Borick is founder and CEO of 1X, makers of the Neo household robot. The company has sold pre-orders with a guaranteed shipping date of 2026.
The first 10,000 pre-orders sold out in the first few days. The payment model includes a $500 monthly subscription. The product is positioned for early adopters willing to accept initial limitations.
Neo will be launched as a platform allowing third-party developers to build skills, similar to an app store. Developers can create specialized capabilities (example: salad-making skills) that users can subscribe to. The platform extends beyond consumer applications to create labor abundance across society.
The company expects to ship robots with near full autonomy in 2026, though some teleoperation or guidance may be needed initially. The robots may fall during early deployments. The goal is to ship a fully autonomous experience that remains useful.
Teleoperation allows remote operators (potentially in the Philippines at $3/hour) to control robots for specific tasks. This mirrors existing models like remote cashiers in retail. The platform supports multiple approaches: full automation, occasional teleoperation intervention for data improvement, or complete teleoperation depending on use case complexity.
The platform includes data collection equipment such as gloves with tactile sensors matching the robot's capabilities. Developers can gather data in their environments, fine-tune models with dense captioning provided by the system, and deploy customized solutions. The platform supports running third-party models including those from OpenAI, Claude, or independent world model labs.
The company believes its own model will be the best but will allow other models to run on Neo. This openness is necessary because no single general model currently solves all robotics challenges. The focus remains on solving embodied AGI to create labor abundance while enabling ecosystem development.
The speaker describes a personal use case: using Neo for remote presence while traveling, wearing a hat on the robot to maintain company operations, inspect parts, and interact with team members.
Teleoperation remains essential regardless of autonomy advances. Applications include factory environments where operators can remotely navigate, pick up parts, participate in meetings, and perform tasks. Remote power stations located over an hour from human presence benefit from having robots on standby for occasional maintenance tasks like flipping switches. These one-off scenarios occurring every few months don't justify full automation.
The "expert in place" model enables teleporting the world's best experts to solve situations anywhere. This applies to surgeons and other specialists who can provide expertise remotely without physical travel.
Neo has become so capable with new hands that teleoperation cannot fully utilize the hardware capabilities. The fidelity of the hand exceeds what teleoperators can leverage. Full haptic systems would slow operations and become clunky. The approach now focuses on gathering data by having humans wear robot sensors transparently without disturbing natural movement.
The decade-long bet involves training robots similar enough to humans to utilize all available human video data. The WX World Model Lab was established to leverage scaling laws on this approach. Building the world model at scale involves multiple data sources beyond factory sensor data.
The data pyramid has four tiers:
- Teleoperation data: highest quality, small fine-tuned datasets where operators attempt tasks repeatedly, selecting successful samples
- Sensor data: humans wearing robot sensors during tasks, closer to robot embodiment but not identical
- Egocentric video: human point-of-view footage, valuable because robot hands match human hands
- General video data: YouTube and internet video of the world, vastly larger than other sources
True intelligence requires multiple orders of magnitude more data than egocentric or sensor data collection can achieve. AI breakthroughs occur when new massive data sources become usable. The approach depends on utilizing general internet video data, requiring robots to match human form and mechanics as closely as possible, including nonlinear properties like tissue deformation, friction, and impact energy.
The differentiation from other robotics companies involves pre-training models on internet video data with humans as the cross-embodiment reference rather than other robots. All data will eventually become robotics data with actions, tactile feedback, and forces, but requires first creating base models good enough for deployment and data collection.
Current development trajectory suggests hard takeoff—robots building robots, data centers, chip fabs, mining, and refining—within under 10 years. The current bet is 3 years, though 5-7 years remains acceptable. This enables true abundance of labor and self-sufficient systems.
Physical AI will progress science by enabling models to build and conduct experiments and observe results. The timeline is years rather than decades. Neo can be ordered online with transparent pricing similar to Tesla ordering.
Boston Dynamics has transitioned from research and development to real-world deployment. The company has changed ownership from independent venture-backed to Google, then to Masayoshi, currently owned by Hyundai. Spot has over 500 customers across 46 countries and is the most deployed mobile autonomous robot globally.
Industrial inspection represents the primary use case, including acoustic monitoring, gauge reading, and vibration detection. The robot can perform inspections during the day and security perimeter work at night. Customer ROI must be achieved within two years. The robot records multiple sensors including video, vibrations, radar, and acoustics.
Spot uses a capex model priced between $100,000 for base configuration to $300,000 fully loaded with services and integration. The business model for Atlas will likely shift to robot-as-a-service. Mean time between intervention exceeds 3,000 hours. Battery life is approximately 90 minutes with automatic charging station swaps.
The focus is on augmenting human labor rather than replacement, enabling humans to perform knowledge work instead of inspections. Value creation includes identifying issues like air leaks that would cost $3 million daily. The approach emphasizes removing dangerous, dull, and dirty jobs that harm human bodies.
Boston Dynamics employs two brain systems: the physicality control brain managing dynamic movement, reliability, and manipulation lives on the robot, while the reasoning layer for semantic environment understanding can operate in the cloud through partnerships with Google DeepMind and other AI partners. Customer-specific workflow information lives between robot and cloud.
100% of Boston Dynamics robots are built outside China and Taiwan. The CEO states Chinese humanoid robotics should not be allowed in the United States due to security concerns and IP protection, citing semiconductor industry lessons. A national robotics strategy is needed to bring manufacturing to the US and allied countries.
Boston Dynamics maintains an anti-weaponization stance for industrial focus, though explosive ordnance disposal (EOD) represents an accepted use case. The company currently performs EOD work. The CEO acknowledges that if China deploys weaponized robots, the company would need to respond, though the decision timing remains uncertain.
The United States must protect intellectual property and bring robotics manufacturing into the United States or allied countries, as China cannot be allowed to win in robotics the way it has in semiconductors.
Boston Dynamics is based in Waltham, Massachusetts, outside of Boston, and is considering opening a West Coast facility. The company offers some remote work options and is actively hiring.
Professor Jonathan Hurst is co-founder and Chief Robot Officer at Agility Robotics. He holds a PhD in robotics from 2008 and has over 20 years of experience in the field.
Twenty years ago, robotics was primarily focused on automation systems in industry, while research communities worked on humanoid robots and autonomous mobile robots. The focus was on building intelligence and hardware capable of real-world impact. Recent breakthroughs have moved robotics from research topics into direct real-world applications. Universities have seen exponential growth in robotics programs due to student demand.
Previous humanoid robot efforts represented false starts, similar to AI winters. The key difference now is that robots can perform useful tasks in human spaces, not just appear humanoid. Large language models have solved perception challenges - robots can now identify objects on a table and understand context without narrow programming. This represents a major inflection point enabling broader context awareness and generality.
Unlike language models trained on internet data, robot control requires torque commands and sensor input data that doesn't exist in training sets. Multiple approaches are being used including learning from demonstration, teleoperation, animation input, and motion capture. However, human control of robots has limitations as it doesn't optimize for robot capabilities.
World models represent the next frontier, with companies using motion capture and human operators wearing gloves to control robots remotely for tasks like making salads or pouring water. World models allow simulation of warehouses and physics for millions of iterations computationally, though significant sim-to-real gaps remain due to unmodeled factors like condensation and wave dynamics.
There is no singularity or single crossing-over moment expected. Progress resembles a snowball gaining momentum rather than a discrete breakthrough. The combination of all tools - world models, real-world practice, and iterative improvement - will drive advancement. Robots require significantly more data and examples than humans to learn, though once one robot learns a skill like playing violin, all robots of that type can share that knowledge via Wi-Fi.
Digit robots currently perform multi-purpose workflows including picking bins and totes, palletizing, and depalletizing. The humanoid form factor is justified by the need for two arms to handle large objects, whole-body control for dexterity, and balancing capabilities. Digit V5, releasing later this year, will be the first balancing humanoid robot that can operate without physical barriers between robot and humans in warehouse settings.
Industrial facilities like Tesla and Toyota factories have safety lines that shut down operations if crossed. Agility Robotics spent 2-3 years working with Amazon to meet safety requirements, requiring holistic redesign of every robot system. Digit V5 represents a key scaling moment by meeting these safety standards without physical barriers.
Robot bill of materials currently costs tens of thousands of dollars. Target pricing aims for car-comparable costs. Digit V5 robots can operate 20 hours per day due to fast charging, providing 78,000 hours annually or 40,000 hours over a 5-year lifespan. The value proposition centers on replacing $20-40/hour human labor with robots costing significantly less per hour once scaled.
Agility offers both capital expenditure purchases and robot-as-a-service models, with the latter providing lower barriers to entry and reduced risk. Pricing follows human labor costs as the benchmark, creating an inelastic price structure. The company maintains substantial room for profit margins while delivering customer savings.
Major facilities like Amazon already have majority robotic workforces through AMRs, conveyor belts, and industrial arms. The transition to 51% robotic workforce in package sorting has essentially occurred. Future developments include autonomous trucks with lights-out autonomous package sortation facilities. Humanoids are most valuable for human environments rather than highly optimized specific tasks.
Agility explored package delivery with Ford seven years ago, demonstrating Digit getting out of vehicles and delivering packages to front porches including stairs. While this remains on the roadmap, it's not considered the best initial market entry point compared to warehouse and industrial applications.
The 3Ds of robotics - dull, dirty, and dangerous jobs - represent the primary use case. The goal is for future generations to view current manual labor jobs the way we view 1900s coal mining, as work that shouldn't be done by humans. Quality of life improvements will mirror the transformation seen since 1900.
Current students have advantages during this period of technological change, as established professionals must adapt to new methods. Core engineering skill sets will remain applicable across evolving career paths. Beyond PhD and master's programs, blue-collar equivalents include robot operators, assemblers, builders, and maintenance technicians.
Two arms provide sufficient capability for handling large objects while maintaining simplicity. Adding third, fourth, or fifth arms increases coordination complexity without clear utility justification. Design decisions prioritize the simplest possible robot capable of required tasks.
Favorite robots include Wall-E and Eve for their dedication to their designed purpose, and Baymax from Big Hero 6 for demonstrating clear helpful programming. The Baymax example illustrates the importance of software safeguards and emergency stops to prevent robots from causing harm.
Agility Robotics operates three locations: Salem, Oregon (original facility), Fremont, California (new facility focused on robot behavior development), and Pittsburgh (near Carnegie Mellon). The company is actively hiring across these locations.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
