Robotics / physical AI
The credible next big thing, if embodied data, hardware cost and reliability improve together.
Robotics is the credible next big thing because it is where the AI stack meets physical scarcity: labor, factories, warehouses, energy networks, farms, roads, skies and infrastructure. But the useful version is not a sci-fi humanoid in every home. The practical version starts in manufacturing and drones, where tasks are bounded, ROI is visible and fleets can learn from repeated work.
Manufacturing
Factory cells, mobile robots, inspection, palletizing, welding support, line tending.
Drones
Inspection, agriculture, mapping, public safety, security, delivery on constrained routes.
Humanoids
Useful where facilities are built for people, but gated by uptime, safety and dexterity.
The robotics wave is not one thing
Physical AI is a bundle of adoption paths. Industrial robot arms, autonomous mobile robots, cobots, warehouse systems, inspection drones, agricultural drones and humanoids all sit under the robotics story, but they do not mature at the same rate. The near-term winners are the systems with constrained environments, clear ROI and limited degrees of freedom. The long-term story is more general machines that can cross task boundaries.
Manufacturing is the first serious beachhead
Factories already buy robots. That matters. Robotics does not need to persuade manufacturing that automation exists; it needs to make automation more flexible, cheaper to deploy and easier to reprogram. IFR's 2025 data showed 542,000 industrial robots installed in 2024, more than double the number from ten years earlier. That is the opposite of a pure hype market: there is already budget, procurement muscle, integrator knowledge and a measured installed base.
The useful path is cells, fleets and inspection before humanoid everything
The cleanest manufacturing path is not a general humanoid replacing every worker. It is AI improving specific production cells: visual inspection, bin picking, palletizing, welding support, line tending, mobile material movement, predictive maintenance and exception handling. These jobs are repetitive enough to measure, valuable enough to budget for and bounded enough to make safety cases. Humanoids may become useful where the environment is built for people, but the first money is still in narrow, boring automation.
Drones are robots with an easier body
Drones are the underrated physical-AI branch because they avoid many humanoid constraints. They do not need legs, hands, doors, stairs or human-scale manipulation. They need perception, navigation, fleet management, battery logistics and regulatory permission. That makes the adoption path clearer for infrastructure inspection, agriculture, public safety, mapping, security and selected delivery routes.
BVLOS is the drone equivalent of cloud getting APIs
Most valuable drone work needs beyond-visual-line-of-sight operation. Without BVLOS, many commercial drones remain expensive remote cameras. With routine BVLOS, drones become scheduled infrastructure: inspect power lines, survey construction, monitor crops, respond to incidents and move small payloads. The FAA still treats package delivery through Part 135 certification plus BVLOS waivers or exemptions, and broader BVLOS rulemaking is the key regulatory hinge.
Why manufacturing and drones belong in the same lineage
Both have the same Wardley pattern: start with constrained tasks, build fleet telemetry, standardize deployment templates, then compound learning across many sites. A robot arm in a factory and a drone inspecting utility lines look different, but both become products when the vendor can sell a repeatable operating pattern instead of a bespoke research project.
What will fizzle inside robotics
The home general-purpose robot will probably disappoint in the near term. Full humanoid autonomy in messy human spaces is much harder than demo videos imply: safety, liability, manipulation, uptime, maintenance and unit economics all stack against fast consumer adoption. The robotics wave can be real even if the domestic servant robot narrative fizzles.
The actual next-big-thing test
Robotics becomes the next big thing only if it moves from custom deployments to productized operating templates. Manufacturing and drones are the best tests because they can show measurable throughput, fewer dangerous human hours, lower inspection cost, better asset uptime and repeatable fleet learning. If those curves bend, physical AI becomes a platform wave. If they do not, it remains a collection of impressive demos and expensive pilots.
Manufacturing automation incumbents
These companies matter because factories already trust them. If physical AI crosses into manufacturing, much of the revenue may pass through incumbents, integrators and plant-standard vendors before it reaches humanoid startups.
| Company | Lane | Why it matters | Traction proof |
|---|---|---|---|
| FANUC | Industrial robots / CNC / factory automation | One of the core incumbents in robot arms and factory automation; strong where reliability and integrator ecosystems matter more than demo novelty. | More AI-assisted setup, vision, maintenance and flexible cell programming without losing industrial uptime. |
| ABB Robotics | Industrial robots, cobots, mobile robotics, software | A broad automation supplier with credibility across automotive, electronics, logistics and general industry. | If ABB makes robot deployment feel more like a repeatable software product and less like bespoke systems integration. |
| Yaskawa Motoman | Motion control, welding, handling and industrial robots | Strong in servo/motion control and factory robot applications where precision and lifecycle support dominate buying decisions. | AI-enabled programming and easier changeovers in high-mix manufacturing. |
| KUKA | Automotive and general industrial automation | A major industrial robot brand with deep automotive/manufacturing deployment history. | Keeping relevance as factories demand more flexible automation and Chinese competition compresses hardware margins. |
| Universal Robots | Collaborative robots | The cobot reference point: useful when automation needs to fit smaller sites, lower-volume production and human-adjacent workflows. | Whether AI tooling lowers setup friction enough for more SMEs to buy robots without heavy integrator spend. |
| Omron / MiR | Autonomous mobile robots and factory logistics | Factory autonomy is not only arms. Moving material through plants and warehouses is one of the cleanest robotics ROI paths. | Fleet learning, traffic orchestration and integration into warehouse/factory software stacks. |
Physical-AI and humanoid entrants
These are the companies trying to turn foundation-model progress into mobile manipulation. The key question is not demo quality; it is whether they can produce safe, durable, paid work in repeatable industrial settings.
| Company | Lane | Why it matters | Traction proof |
|---|---|---|---|
| Figure AI | Humanoid robots for manufacturing and logistics | High-visibility physical-AI entrant; interesting because the target is paid industrial work, not consumer novelty. | Sustained factory/logistics deployments with uptime, safety and cost data that survive beyond pilots. |
| Agility Robotics | Digit humanoid for logistics workflows | Focused on tote movement and logistics, which is a narrower and more credible starting point than a general household robot. | Repeatable warehouse deployments where the robot handles dull movement work at acceptable operating cost. |
| Apptronik | Apollo humanoid; industrial and logistics pilots | Important because it sits near enterprise logistics, automotive partners and AI-model partnerships rather than only consumer theater. | Moving from announced pilots to measurable production workflows. |
| Boston Dynamics / Hyundai | Atlas, Spot and advanced mobile robotics | Technical credibility is very high; commercial question is whether Atlas becomes productized manufacturing capability, not just a robotics benchmark. | Factory deployments that trade spectacle for reliable repetitive work. |
| Tesla Optimus | Vertically integrated humanoid manufacturing bet | Tesla has factories, batteries, actuators, AI talent and a CEO narrative machine. That makes Optimus impossible to ignore and easy to overrate. | Internal Tesla factory usage with credible unit economics, not stage demos or production targets alone. |
| Unitree / Chinese humanoid ecosystem | Low-cost humanoid and quadruped hardware | China's supply chain can compress hardware cost quickly. That matters if robotics becomes a scale manufacturing contest. | Reliability, software autonomy and safety certification catching up with hardware price/performance. |
Drones, inspection and fleet autonomy
Drones are the practical robotics branch where body complexity is lower and the adoption question shifts toward regulation, fleet operations, software workflow and mission density.
| Company | Lane | Why it matters | Traction proof |
|---|---|---|---|
| DJI Enterprise | Drone hardware for inspection, public safety, mapping and agriculture | The hardware gravity well of drones. Any drone-market thesis has to account for DJI's cost, reliability and sensor ecosystem. | Whether non-Chinese alternatives can match capability/cost under procurement and security pressure. |
| Skydio | Autonomous drones for defense, public safety and infrastructure | The US autonomy-first counterweight to DJI, especially where security, domestic sourcing and obstacle avoidance matter. | Enterprise and government deployments that justify premium pricing versus DJI-class hardware. |
| Zipline | Autonomous drone delivery networks | The clearest example of delivery drones as logistics infrastructure rather than gadgets. | More high-frequency routes, regulatory expansion and economics that beat cars for small payloads. |
| Wing | Alphabet drone delivery | Strategic because it pairs autonomy, routing, consumer delivery and a large parent company that can fund long adoption cycles. | Routine city/suburb delivery density without noise, safety or unit-economic backlash. |
| DroneDeploy | Reality capture software for construction, energy and infrastructure | Shows the less glamorous drone thesis: the money is often in workflow data, inspection records and progress tracking, not aircraft sales. | More sites standardized on drone/ground reality capture as recurring operational software. |
| Anduril / defense autonomy stack | Autonomous defense systems and drones | Defense is a separate but powerful drone adoption lane: autonomy, sensors, fleet command and attritable hardware. | Procurement scale and operational feedback loops translating into reusable autonomy platforms. |
Manufacturing and drones are the right robotics subpages because they are already moving from custom systems toward repeatable products. Humanoids are the optional interface, not the whole thesis.