Robotics: what are startups building now?
In our robotics market deck, you will find everything you need to understand the market
SUMMARY
Robotics startups are building practical machines for narrow, valuable jobs today while spending heavily on humanoids and general robot intelligence that could make future machines much easier to retrain and redeploy.
The commercial proof is still strongest in specialized robotics. Warehouse systems, construction robots, agricultural machines and hospital robots already have the clearest records of repeated work, measurable savings and customer adoption.
Humanoids have moved beyond pure demos. Agility has contracted orders, Figure is producing hundreds of robots and working inside BMW plants, Apptronik is running training fleets, and AgiBot has pushed production into the tens of thousands.
The first useful humanoid jobs are surprisingly ordinary: moving totes, sequencing parts, pushing carts and handling repetitive material flows. That is probably a feature, not a weakness, because these jobs are easy for customers to value.
Home humanoids remain much earlier. 1X has reservations, pricing and a factory, but the category still lacks the kind of large, independently visible installed base that would prove ordinary households can rely on humanoids for months at a time.
Investors are also funding the software layer almost as aggressively as the robot makers themselves. Skild AI, Generalist, Physical Intelligence and FieldAI are betting that one intelligence stack can eventually control many robot bodies instead of every hardware company building its own brain from scratch.
That software race explains why data has become a strategic asset. Figure is building a large human-video data network, Apptronik is collecting demonstrations through Robot Parks, and model-first companies are trying to learn more from video because robot-generated training data is expensive.
Warehouses, farms, hospitals, construction sites and defense programs are emerging as unusually fertile markets because customers can attach a dollar value to a specific physical task. A robot that removes labor, chemicals, dangerous work or wasted staff trips has a much easier path to a budget.
The funding boom is therefore running ahead of the revenue base in some of the most ambitious categories. Billions are going into general-purpose robots and robot foundation models before those businesses have anything close to the operating history of specialized machines.
China currently has the clearest lead in humanoid manufacturing volume, while U.S. startups remain especially strong in model development, integrated humanoid systems and venture funding. The next meaningful comparison is less about who can build the most robots and more about which fleets get reordered because they actually earn their keep.
The robotics market now has two races happening at once: proving that a robot can do one job reliably enough to make money, and making robots learn new jobs cheaply enough that the same machine can become useful again and again. Most of the proof sits in the first race; most of the upside sits in the second.
This market map, featured in our robotics market deck, highlights top companies and startups in the robotics market
Why is robotics startup funding exploding right now?
Robotics startup funding is exploding because investors now believe AI can make robots improve fast enough to justify the painful cost of building hardware.
Crunchbase had already counted $18.8 billion of global robotics startup funding by late June 2026. That was more than the $15 billion invested during the whole of 2025 and the $14.1 billion peak reached in 2021. Its broader physical-AI category, which also includes autonomous vehicles, drones and other machines operating in the physical world, attracted $47.4 billion across 521 deals in the first half of 2026. The same category had raised $26.4 billion a year earlier.
The pace has stayed high. XPeng's robotics business raised more than $900 million in August at a valuation above $6.3 billion. Generalist added almost $200 million shortly after announcing a $400 million round in June. Skild AI raised $1.4 billion earlier in the year at a valuation above $14 billion.
The interesting part is where those giant checks are going. Investors are funding companies such as Figure and XPeng that want to manufacture large numbers of general-purpose robots, while companies such as Skild and Generalist are raising comparable sums to build the intelligence that could run many different robots.
That is a meaningful change from the previous robotics cycle. A hardware startup used to carry all the disadvantages of manufacturing while still selling a machine with fairly fixed capabilities. Today's bet is that the same robot can keep gaining skills through software, data and better AI models after it has already been built.
The bet remains expensive. Still, the amount committed before most general-purpose robots have generated serious revenue shows how much the perceived upside has changed.
| Funding measure | Amount |
|---|---|
| Robotics startup funding in all of 2021 | $14.1B |
| Robotics startup funding in all of 2025 | $15.0B |
| Robotics startup funding by late June 2026 | $18.8B |
| Physical-AI funding in H1 2025 | $26.4B |
| Physical-AI funding in H1 2026 | $47.4B |
If you want more recent data on this point, please see our latest robotics market report.
Are humanoid robots actually becoming a real business now?
Humanoid robotics has crossed into early commercial business, although today's revenue still looks tiny beside the order books, valuations and promised production volumes.
Agility Robotics has just given us an unusually clear look at that gap. In its recent SEC filing, Agility reported only $1.78 million of net sales for 2025. At the same time, the company says it has secured more than $300 million of multi-year contracted orders for its next-generation Digit v5, subject to contractual milestones.
The same filing finally puts numbers around what a humanoid worker might cost. Agility models Digit v5 at roughly $200,000 to buy, plus about $20,000 for deployment and around $36,000 a year for software and maintenance. Using Agility's assumed five-year life, that comes to roughly $400,000 per robot. The company models customer payback at about 1.1 years, although those are planning assumptions rather than quoted prices for every customer.
That is much more useful than another video of a humanoid walking around a factory. We can finally ask whether a robot costing around $400,000 over five years produces enough work to earn that money back.
Figure is also starting to look like a manufacturer rather than a research lab. The company said in April that it had built more than 350 Figure 03 robots and had shortened its production cycle from one robot per day to one per hour in less than four months. Its published figures still do not tell us how many of those machines have been sold to external customers, since Figure also needs robots for training, testing and internal development.
China is already operating at a different production scale. AgiBot said its 15,000th robot came off the line in June.
So yes, humanoids are becoming commercial products. The stronger test now is what happens after those first purchases: how many robots stay busy, how quickly customers buy more, and how much recurring revenue the manufacturers actually collect.
| Company | Strongest commercial evidence so far | What remains unclear |
|---|---|---|
| Agility Robotics | $300M+ of contracted Digit v5 orders | Revenue recognition and large-fleet economics |
| Figure | 350+ Figure 03 units built; BMW production work | Number of external paid deployments |
| AgiBot | 15,000 cumulative robots produced | Mix of paid, repeat and large-volume customers |
| Apptronik | Apollo fleets operating at Robot Parks and customer sites | Commercial fleet size and revenue |
As this chart shows, and as featured in our robotics market deck, search interest in robot costs has increased significantly
What jobs are humanoid robots really doing today?
Humanoid robots today are doing their most convincing work in material handling: moving totes, manipulating parts, sequencing components and pushing or pulling things around factories and warehouses.
Agility's Digit is the cleanest example. At GXO's Flowery Branch logistics facility, Digit has completed more than 100,000 tote moves. The robot transfers totes between autonomous mobile robots, conveyors and floor locations. One successful pick tells us very little; 100,000 repetitions tell us the machine can survive a repetitive production workflow.
Figure's work at BMW is becoming more complex. Figure 02 previously handled sheet-metal parts and contributed to production involving 30,000 BMW vehicles during 2025. Figure 03 has since returned to the Spartanburg plant for a sequencing job where the robot manipulates parts, repositions its body and moves a wheeled cart.
Apptronik is taking a slightly different route with Apollo 2. Fleets now operate at its Robot Park facilities and customer sites, continuously collecting demonstrations for the models Apptronik is developing with Google DeepMind. Apollo 2 comes in both a walking version and a wheeled version.
That wheeled option tells us something useful about the market. Customers care about hands, reach, perception and the ability to learn new jobs. Walking on two legs only earns its extra complexity when the environment actually requires it.
The strongest humanoid jobs today are therefore fairly mundane. They involve bins, carts, parts, shelves and repetitive movement around workplaces designed for humans. A bit boring, maybe, but that is exactly where a new machine starts becoming economically useful.
| Robot | Real work being tested or deployed |
|---|---|
| Agility Digit | Moving and stacking totes in logistics |
| Figure 03 | Part sequencing, manipulation and cart movement at BMW |
| Apptronik Apollo 2 | Industrial tasks plus continuous training-data collection |
| AgiBot platforms | Manufacturing, electronics and logistics applications |
Are home humanoid robots finally something people can buy?
Home humanoid robots are finally becoming products people can reserve, while verified mass household use is still ahead of us.
1X has gone furthest. NEO can currently be reserved with a $200 refundable deposit. Early Access ownership costs $20,000, and 1X advertises a later $499-per-month subscription. The company says its first-year production capacity of 10,000 robots was spoken for within five days of opening preorders.
Production has started at 1X's Hayward factory in California. The important detail is where the robots have gone so far. In its latest factory updates, 1X described machines coming off the line for internal development and home testing and continued to describe U.S. customer deliveries as starting during 2026. We could not find independent confirmation of a meaningful fleet already operating inside paying customers' homes.
NEO's software also shows how early the category remains. The robot arrives with basic autonomy, while unfamiliar chores can use 1X's scheduled Expert Mode, where a remote human supervises the robot and helps it complete the task. That arrangement may also produce valuable training data, but today's product is still quite different from an appliance that quietly handles the house by itself.
Figure is training its own systems on chores such as tidying rooms and loading dishwashers, and Figure 03 was redesigned with softer materials and hardware choices suited to domestic environments. Figure still has no comparable consumer sales offer.
Home humanoids have reached the preorder-and-production stage. We still need the next piece of evidence: hundreds or thousands of ordinary households using them for months with little human help.
If you want more recent data on this point, please see our latest robotics market report.
This chart, featured in our robotics market deck, shows annual venture capital investment in robotics startups
Why do warehouses keep producing some of the best robotics startups?
Warehouses keep producing strong robotics businesses because they contain repetitive work that is easy to measure and awkward enough that traditional automation still leaves plenty for people to do.
The overall market is already large. The International Federation of Robotics counted 102,900 transportation and logistics service robots sold in 2024, more than half of all professional service-robot units sold that year.
Yet a warehouse still contains dozens of jobs that a conveyor or traditional robot arm handles poorly. Parcels arrive in random shapes. Trailers need to be packed tightly. Pallets need to move between storage levels. Inventory changes constantly. Human workers end up covering the gaps.
Dexterity is going after one of the nastiest gaps: loading trailers. FedEx expanded Dexterity's system at its Hagerstown hub in July 2026 after several years of development and pilot work. The system receives irregular packages and decides in real time how to build stable walls inside the trailer. That move from pilot equipment into a larger production deployment is far more useful evidence than another laboratory benchmark.
Mytra is redesigning pallet movement. Its robots move loads of up to 3,000 pounds horizontally and vertically through a storage grid. During 2025, Mytra signed one deployment around 60 times larger than its previous biggest installation, put another customer system into production and shipped two additional pilots. It then raised $120 million in January 2026.
Humanoids are joining this market too, as Agility's GXO work shows. They are entering a warehouse automation ecosystem that already includes autonomous mobile robots, sorting machines, pallet systems, drones and robotic arms.
That competition is healthy. A warehouse operator can compare each robot against a specific labor cost and throughput target. Startups that cannot beat the existing workflow get exposed quickly.
Are specialized robots still beating general-purpose robots?
Specialized robots are still comfortably ahead on proven commercial use today, especially when we measure years in service, cumulative jobs completed and obvious customer ROI.
Dusty Robotics says its construction robots have printed more than 300 million square feet of building layouts across over 1,000 projects. Carbon Robotics is used by more than 200 growers in 15 countries. Diligent Robotics' Moxi fleet had already completed more than 1.25 million hospital deliveries with roughly 100 robots across more than 25 facilities when Serve Robotics agreed to acquire the company.
Those machines have narrow jobs. Dusty's FieldPrinter transfers building plans onto construction floors. Carbon's LaserWeeder finds and kills weeds. Moxi carries hospital supplies.
That narrow scope makes the commercial question much easier. A hospital can count how many staff trips Moxi removes. A contractor can compare layout hours before and after FieldPrinter. A farmer can measure labor, herbicide use and crop output.
General-purpose robots have a harder job because flexibility itself has to become valuable. A humanoid costing hundreds of thousands of dollars makes sense when one machine can eventually move between several jobs, survive changes in the workplace and learn new tasks without a major engineering project.
We have early evidence of that flexibility, especially in factories and warehouses, but the cumulative operating record remains much stronger for specialized machines.
The gap could shrink quickly if robot learning keeps improving. For now, startups selling one excellent physical skill have the clearer business.
If you want more recent data on this point, please see our latest robotics market report.
This chart, featured in our robotics market deck, breaks down Figure’s playbook in robotics
Is the biggest robotics bet now the robot brain?
Some of the biggest startup bets in robotics now sit inside the robot's brain, with investors spending billions on models designed to control many different machines.
Skild AI raised $1.4 billion in January 2026 at a valuation above $14 billion. Generalist announced $400 million in June and then raised almost another $200 million in August, according to regulatory filings and reporting from TechCrunch and Axios. Those two model-first robotics companies alone have attracted roughly $2 billion this year.
Their pitch is unusually ambitious. Skild wants one “omni-bodied” model that can work across different robot shapes. Its S1 model, introduced in August, is designed to watch a single video demonstration and attempt a new task through in-context learning. Generalist is also building models that transfer skills across robot hardware and learn from short demonstrations.
Skild's strategy has already moved beyond research models. In April it acquired Zebra Technologies' robotics division, formerly Fetch Robotics, giving it access to established warehouse robots and customer environments. It has also announced work with ABB Robotics and Universal Robots.
Physical Intelligence is developing general robot policies such as π0, while FieldAI builds models intended to run on wheeled robots, quadrupeds, humanoids and larger industrial vehicles.
The strategic question is simple and huge: does the winning company own the whole stack, or does robotics develop a common intelligence layer that many hardware makers plug into?
We have seen that pattern before in computing, where a large amount of value accumulated in software shared across different hardware. Robotics is currently spending billions to see whether physical machines evolve the same way.
Where are robotics startups getting enough data to train all these robots?
Robotics startups are building their own data factories because internet-scale text and images cannot teach a robot exactly how force, contact and movement work in the physical world.
Figure has taken the most aggressive public approach with Index. The company recruits people to record physical activities such as cooking, cleaning, laundry and stocking shelves. Figure recently reported more than 16 million uploaded videos, over 44,000 weekly active contributors and $15 million paid to contributors. Its data pipeline was receiving roughly 30 minutes of new video every second.
Apptronik is collecting data directly through robots. Its Robot Parks run fleets of Apollo 2 machines repeatedly performing physical tasks, with the resulting data feeding work with Google DeepMind's Gemini Robotics models.
Generalist and Skild are pushing hard on video because watching humans is far cheaper than collecting every example with a robot. Their latest models are designed to learn increasingly useful behavior from demonstrations and then transfer that knowledge onto physical machines.
Hardware scale also feeds the data race. Figure said earlier this year that it had produced more than 350 Figure 03 units, over 9,000 actuators and more than 500 battery packs. A bigger fleet gives the company more machines that can train, fail, generate edge cases and produce new data.
That can become a compounding advantage: better software makes the fleet more useful, a bigger fleet generates more experience, and the next software update improves from that experience.
The companies that solve data collection cheaply may gain an advantage well before anyone solves general-purpose robotics itself.
This chart, featured in our robotics market deck, shows annual funding in robotics startups
Why are construction robots becoming quietly commercial?
Construction robotics is becoming one of the more convincing startup markets because founders are automating individual jobs where crews still spend huge amounts of time measuring, moving and repeating the same actions.
Dusty Robotics has built one of the clearest examples. Its FieldPrinter reads BIM or CAD drawings and prints the layout directly onto the construction floor. Dusty says its robots have now printed more than 300 million square feet across more than 1,000 buildings.
Monumental has taken on bricklaying. The European startup recently said it had more than 100 robots laying actual bricks on building sites and raised another $32 million to expand the fleet and enter the United States.
Gecko Robotics focuses on infrastructure that already exists. Its climbing robots inspect boilers, tanks, power facilities and other large industrial assets, then turn those scans into detailed condition data. A deal announced with power-plant operator NAES started above $100 million and can expand beyond $250 million.
August Robotics offers another example of how specific these products can get. The company raised $30 million this year to expand robots built for jobs such as downward drilling on industrial and construction sites.
Customers are already paying for robots that save survey time, lay bricks repeatedly, drill the same kind of hole or inspect dangerous equipment.
Trying to automate the whole construction site can wait. There is already plenty of money in automating one miserable task properly.
Why does agriculture look like one of robotics' strongest business cases?
Agricultural robotics works best when the robot removes a large recurring cost from every acre, and Carbon Robotics is showing how powerful that equation can become.
Carbon's LaserWeeder uses computer vision to identify weeds and lasers to destroy them. The company now says more than 200 growers across 15 countries use its systems.
The customer stories are more useful than the machine's raw specifications. Hungenberg Produce says LaserWeeder cut its weed-control costs by more than 80% and improved yields by roughly 20% to 30%. Rio Fresh bought three machines. Other growers describe cutting herbicide use, reducing hand-weeding crews and improving crop consistency. These are company-published case studies, so the exact percentages are customer claims rather than neutral trials, but repeat purchases carry more weight than a single testimonial.
Carbon is now expanding into tractor autonomy through Carbon Autonomy. That move makes sense. Once a startup already has cameras, perception models and farm data, the same technology can start taking over other repetitive passes through a field.
Agriculture gives robotics startups something many consumer markets cannot: a customer who can calculate the economic value in dollars per acre.
If a robot saves enough labor or chemicals during every growing season, farmers have a reason to keep using it even if the machine looks nothing like the science-fiction version of robotics.
If you want more recent data on this point, please see our latest robotics market report.
This chart, featured in our robotics market deck, compares the main business model options for warehouse AMR robotics providers
What is actually working in medical robotics today?
Medical robotics is already working at meaningful scale when startups focus on a precise clinical task or remove repetitive work from hospital staff.
CMR Surgical's Versius system has now been used to treat more than 45,000 patients globally. The company is expanding into the U.S. after receiving FDA clearance for its newer Versius Plus system in an initial surgical indication.
Mendaera is tackling a completely different procedure. Its handheld Focalist system combines ultrasound imaging, robotic alignment and guidance for needle placement. A peer-reviewed study published this summer reported technical success in 60 of 61 procedures across urology, anesthesiology and nephrology. The participating doctors were first-time Focalist users who received a single 45-minute training session.
Then there is Diligent Robotics. Moxi does not operate on anyone. The robot moves medications, laboratory samples and other supplies through hospitals. By the time Serve Robotics announced its acquisition of Diligent, almost 100 Moxi robots across more than 25 hospital facilities had completed more than 1.25 million autonomous deliveries.
These three examples span surgery, needle guidance and hospital logistics. They also share one useful trait: the robot has a clearly defined job.
Medical robotics gives us some of the strongest evidence that the startup market can grow without waiting for a machine with human-level general intelligence. Better precision, easier procedures and fewer wasted staff trips are already enough.
Are delivery robots actually a real business yet?
Delivery robots are a real operating business today, but Serve Robotics shows that fleet size alone tells us very little about whether the economics work.
Serve ended the second quarter of 2026 with $3.2 million of quarterly revenue, up 404% from a year earlier. Its fleet had expanded to more than 2,000 deployed robots, and the company has been adding food, grocery, healthcare and laundry delivery.
Then we get to the number that makes the story much more interesting. Serve reported 792 daily active robots during the quarter, down slightly from 812 in the previous quarter. Compared with a deployed fleet above 2,000, that means the equivalent of only about 40% of deployed robots were active on an average day under Serve's metric.
Demand also disappointed. Serve cut its full-year revenue guidance from around $26 million to $9 million–$10 million after Uber Eats volumes came in below expectations. Using the midpoint of the new range, expected revenue dropped by roughly 63%.
There are encouraging parts. DoorDash-related revenue grew nearly 50% sequentially during the quarter, and higher-margin recurring revenue rose above half of total revenue. Serve is also moving beyond sidewalks through its acquisition of hospital-robot company Diligent.
Still, the latest numbers are a useful warning for the whole robotics sector. Building thousands of autonomous machines is only half the job. Those machines need enough work every day to justify the capital tied up in them.
This chart, featured in our robotics market deck, breaks down revenue across customer segments in the robotics market
Is defense becoming one of the fastest routes to robotics revenue?
Defense has become one of the fastest routes from robotics prototype to serious contract because governments are already buying autonomous systems in quantities that commercial customers often approach more slowly.
Forterra gives us an unusually concrete example. The company designed, built and delivered 105 autonomous Lancer ground vehicles to support Ukraine in less than six months under a U.S. government program.
Forterra has also received two ROGUE-Fires Block 2 delivery orders worth a combined $92 million with Oshkosh Defense. The system puts autonomous driving technology onto military ground vehicles for dangerous missions.
Shield AI is pushing autonomy into aircraft. Its Hivemind software was selected for the U.S. LUCAS program, where autonomous aircraft are expected to operate collaboratively rather than requiring one human pilot for every vehicle.
Anduril has accumulated a different kind of operating history underwater. Before its selection for a U.S. Navy and Defense Innovation Unit extra-large autonomous underwater vehicle program, the company said its autonomous underwater vehicles had traveled more than 42,000 kilometers and logged more than 6,700 mission hours.
Across air, land and sea, startups increasingly want the autonomy software to move between vehicles.
Defense also changes the economics. Governments will pay heavily for machines that reduce risk to soldiers, increase the number of systems one person can control or perform missions that would otherwise require expensive crews.
That makes defense one of the clearest places where general autonomy can generate large contracts before general-purpose household or workplace robots are mature.
Is China actually ahead in humanoid robotics now?
China is ahead on humanoid manufacturing volume today, and the harder question is whether that production lead turns into repeat commercial demand and durable profits.
AgiBot reached 10,000 cumulative robots in March 2026 and 15,000 by June. Moving from 10,000 to 15,000 in roughly three months is far beyond the published production totals of the leading U.S. humanoid startups.
TrendForce currently expects China's humanoid market to reach around RMB 15 billion in 2026 and grow at least another 60% in 2027. It says deployments are spreading through automotive manufacturing, electronics, aerospace, logistics and energy.
Orders are becoming more concrete too. Galbot has secured an RMB 236 million procurement order for embodied-AI equipment. XPeng spun its robotics operation into a heavily funded business and raised more than $900 million in August at a valuation above $6.3 billion as it prepares to scale its IRON humanoid.
Unitree gives us another useful benchmark even though it has now moved into the public market. TrendForce reported that Unitree's 2025 revenue more than tripled to RMB 1.699 billion and that humanoids accounted for more than half of revenue for the first time. Profit has been under pressure this year even as sales grow.
That is the awkward bit. Recent reporting from The Wall Street Journal says Chinese regulators are asking for stronger evidence of financial health and revenue potential from humanoid companies seeking public listings after Unitree's volatile debut.
China has moved furthest on the industrialization side of the problem: suppliers, production lines, lower-cost hardware and thousands of physical machines.
The next race is already starting. Customers need to reorder those robots because the economics work, rather than because every large company wants a humanoid pilot on its innovation roadmap.
If you want more recent data on this point, please see our latest robotics market report.
This chart, featured in our robotics market deck, shows how home cleaning robot technology has evolved over time
So what are robotics startups actually building now?
Robotics startups are building practical machines for narrow jobs today while pouring enormous amounts of capital into making future robots easier to teach and redeploy.
The commercially strongest group remains specialized robotics. Warehouse machines load trailers and move pallets. Dusty's robots print construction layouts. Carbon removes weeds. Moxi delivers hospital supplies. These companies can point to millions of completed tasks, hundreds of customers or large production contracts because the job is tightly defined.
Humanoid companies have moved forward quickly. Agility has a genuine order book and live deployments. Figure is manufacturing hundreds of machines and doing increasingly complicated work at BMW. AgiBot has already produced robots in the tens of thousands. Apptronik has operating fleets collecting real-world data. The newest evidence makes it increasingly difficult to treat humanoids as a collection of research demos.
Their current jobs remain concentrated around material handling, logistics and manufacturing. Home robotics sits earlier in the curve: 1X has real reservations and a real factory, while large-scale customer use still needs to be demonstrated.
Then we have the part of robotics attracting some of the wildest capital. Skild, Generalist, Physical Intelligence and FieldAI are trying to build intelligence that can move across tasks and robot bodies. Figure and Apptronik are creating enormous data-collection systems for the same reason. Much of the long-term race now comes down to how quickly a robot can learn something new.
The startup market falls into three fairly clear stages.
| What startups are building | Where it stands today | Examples |
|---|---|---|
| Robots for one valuable job | Already commercially proven in several markets | Carbon Robotics, Dusty Robotics, Dexterity, Diligent |
| Flexible humanoid workers | Moving into early production and paid deployments | Agility, Figure, Apptronik, AgiBot |
| General robot intelligence | Huge funding and fast technical progress; commercial model still forming | Skild AI, Generalist, Physical Intelligence, FieldAI |
| Home humanoids | Production and reservations have started; mass household use remains unproven | 1X, Figure |
| Defense autonomy | Already reaching meaningful government contracts and deployments | Forterra, Shield AI, Anduril |
The clearest answer to “what are robotics startups building now?” is practical. They are automating physical work one job at a time while simultaneously trying to invent the AI that could stop every new job from requiring a new robot.
Most of the upside sits in that second part. Most of the proof sits in the first.
OUR METHODOLOGY
This analysis asks what robotics startups are actually building now and which parts of the market are becoming real businesses. We separated the evidence into capital, production, commercial adoption, real-world deployment, operating economics, data and learning, and the types of physical work robots are already performing.
We prioritized the freshest evidence closest to the underlying event: regulatory filings, company disclosures, customer and partner announcements, regulator records, peer-reviewed research and specialist industry data. We gave more weight to concrete activity such as capital committed, orders signed, machines produced or deployed, tasks completed, active usage, repeat purchases, regulatory clearance and measured performance.
Each type of evidence was used for what it can actually establish. Production volume tells us about manufacturing scale. Orders reveal demand. Active deployments and repeated tasks tell us more about operational maturity. Revenue, utilization and repeat purchases give a clearer view of whether that activity is turning into a sustainable business.
Company-reported figures are used to describe what those companies say they have built, sold, deployed or achieved. Broader conclusions come from patterns that appear across several companies and sectors, so a large funding round, factory milestone or customer announcement does not carry more weight than the underlying evidence supports.
Key sources used for this analysis include Crunchbase News on global robotics startup funding, Crunchbase News on physical-AI funding, Agility Robotics' SEC filing on orders, deployments and commercial economics, Figure on Figure 03 production scale, BMW Group on Figure deployments in Spartanburg, 1X on NEO pricing, reservations and Expert Mode, the International Federation of Robotics on professional service-robot volumes, Apptronik on Robot Parks and Apollo 2 data collection, and Figure on its Index physical-data network.
Additional key sources include Skild AI on its funding and omni-bodied robotics strategy, Generalist on its funding and physical-AI strategy, Mytra on deployment growth and its Series C, Dusty Robotics on FieldPrinter deployment scale, Carbon Robotics on customer adoption and LaserWeeder economics, Monumental on construction deployments, CMR Surgical on Versius patient volumes, the FDA on Versius Plus clearance, PubMed on the Mendaera Focalist feasibility study, and Serve Robotics' SEC filing on revenue, daily active robots and guidance.
For defense and Chinese humanoid manufacturing, we also used Forterra on its Lancer deployment, Forterra on ROGUE-Fires production orders, Shield AI on the LUCAS program, and AgiBot on its 15,000-robot production milestone.
In our robotics market deck, we identify pain points entrepreneurs should prioritize
Related blog posts
- The most recent funding news in the robotics market
- What are the top startups in the robotics market?
- What are the key fundraising trends in the robotics market?
- How strong is fundraising in the robotics market right now?
