Area 2 of 3
Machines at work
A machine that works at one site is a result. At the next site, with different equipment, people and shifts, it is a business.
By Alessandro Singh and Rohith Varma, Last reviewed , 9-minute read
What we mean
We mean robots, sensors and autonomous systems that do a real job in a factory, a field, a warehouse, a plant or a network, and keep doing it at the next site. Others call it robotics, industrial automation or industrial AI.
Think of a robot cell that learns a new part, an inspection system that reads a pipeline or a power line, a vehicle that moves goods around a yard without a driver, or a system that runs irrigation or water treatment. Around them sit the machines, sensors, power equipment, data and control software that run plants, grids, infrastructure and supply chains.
The hard part is the same everywhere. Technology that works in a lab has to work on someone else's site, with their equipment, their people and their safety rules, on every shift.
These machines work in manufacturing and logistics, energy and infrastructure, food and water, and defence. In defence, we never work on systems that choose and engage targets without meaningful human control.
This page is for you if your machine runs at one customer's site, and the second site has not happened yet or happened only with you there.
What we look for
Our thesis is that a technology company is proven at the second deployment. These are its five tests, as they apply here.
- Proof in use. Field hours on a paying customer's site, long enough to show uptime, interventions and how the system fails. Conditions change results: in a study published by the robots' own developers, 23 waste-sorting robots in three office buildings cut contamination by 53% at first, and by 39% once more people were in the buildings. Innovation Endeavors, which invests in robotics companies, gives the same sign of fit: the product is used in production, and customers use it again without the company's involvement.Source 1
Source 2 - A buyer who pays. A budget holder in an operating unit pays for it, at a price that covers the cost of delivering it. A pilot fee is a start, not a pass: in 2024 the same investor warned that pilot revenue can exceed $1 million and still mislead. In Capgemini's 2026 survey, fewer than half of executives (42%) said the business case for AI-enabled robots was clear. Public buyers add formal tenders, and they buy at scale: the European Commission says EU public authorities spend around €2.6 trillion a year on services, works and supplies.Source 2
Source 3 Source 4 - A way to deploy again. The company knows what the second, tenth and hundredth site need: integration with existing systems, installation time, training, and the partners who do the work. In May 2026 the consultancy BCG put system integration, setup and engineering at about three-quarters of the cost of conventional robotics; it gives no source for the figure.Source 5
- A way to support it. Someone maintains the system, answers the call at night and keeps it within the uptime the customer was promised, at a cost the price covers. What the customer buys is uptime.
- A route through the rules. Certification, the safety case, data rules and, where they apply, export controls are mapped, with time and money set aside for them. In the EU, the Machinery Regulation applies from 20 January 2027, and safety components whose behaviour evolves through machine learning need a third-party conformity assessment, a check by an independent approved body. AI safety requirements will then be added by delegated acts that apply by 2 August 2028.Source 6
Source 7 Source 8 Source 9
Why now
The buyer's first reason is a shortage of people, ahead of their cost. In Capgemini's survey of 1,678 executives at large organisations in early 2026, labour shortages (74%) came ahead of rising labour costs (69%) among the reasons to invest in AI-enabled robots. Falling technology costs came last of thirteen reasons, at 7%. A machine bought to cover work nobody can be hired for has to run every shift.Source 10
Buyers have started, and most have not scaled. In the same survey, 79% of organisations were engaging with AI-enabled robots, 27% were deploying or scaling them, and nearly eight in ten executives said scaling was a challenge. The gap between trying and scaling is where the second deployment sits.Source 10
New machines join old ones. The International Federation of Robotics counted more than 600,000 industrial robots installed in factories in 2025, up 11% on the year before, and a record 5 million in operation, more than double the number seven years before. New AI has to work with the machines already on the floor, which is why integration decides so much.Source 11
Capital has followed, narrowly. Crunchbase counted $9.9 billion of venture funding for robotics startups in the third quarter of 2026, excluding defence, against $4.3 billion a year earlier. In European robotics, the ten largest rounds between January and 11 November 2026 took more than nine-tenths of the money on the tech.eu tracker. The tracker counts loans as well as equity, and includes drone makers that supply armed forces. So the money sits in a few large rounds. We look past them, at machines that already work on customers' sites.Source 12
The evidence
What the counts, the surveys and the research found. Each line is one finding.
- Few are at scale. In Capgemini's 2026 survey, 4% of executives said they operated AI-enabled robots at scale, and nearly two-thirds expected them to reach scale within five years.Source 10
- More machines are coming. The International Federation of Robotics forecasts that 655,000 industrial robots will be installed in 2026, up 9%, and 806,000 in 2029. Each has to be installed, integrated and kept running at a site.Source 11
- Half the new jobs could go unfilled. US manufacturers could need as many as 3.8 million more employees between 2024 and 2033, and 1.9 million of those jobs could go unfilled (The Manufacturing Institute and Deloitte, a projection, 2024).Source 14
- The lab is not the site. What a robot learns in simulation works less well in the real world, and closing that gap is a field of research of its own (Zhao and others, IEEE SSCI, 2020).Source 15
- How often a person steps in can be counted. Crop robots drove themselves for 25 kilometres in real fields, with an average of 767 metres between the times a person had to step in (Sivakumar and others, ICRA workshop, 2024).Source 16
- Humanoids are early. In the same Capgemini survey, 72% of executives named technical immaturity as an obstacle to humanoid robots, and 63% named cost.Source 10
Taken together: buyers want these machines and few run them at scale. What stands between is the site: its equipment, its people and its conditions.
Where this could be wrong
Margins can stay thin, even at scale. One listed warehouse automation company reported a gross margin of 18.8% on revenue of $2.2 billion in its 2025 financial year, when the services it runs for customers cost more than they earned. In the first nine months of its 2026 financial year the margin was 21.9%, and those services earned more than they cost. So we look at what one site costs to install and to keep running, and what is left.Source 17
A second site can be as bespoke as the first, and an integrator may have done it, not the company. So we ask who did the second deployment and what it cost.
General-purpose robots could make the second deployment easy. If one trained system works at any site, the second site stops being the proof. That has not happened yet. The best method succeeds 89% of the time on short handling tasks in simulation, but in the household benchmark in Stanford's 2026 AI Index the best team's system completed 12% of tasks in full, and the homes were simulated. So we watch how well these systems work at sites they have not seen.Source 19
Robots change the work around them. In US workplaces, a one-standard-deviation rise in exposure to robots cut work injuries by about 1.2 a year for every 100 workers. US areas more exposed to robots also saw more deaths from drugs and alcohol and more mental health problems (Gihleb and others, Labour Economics, 2022). So we ask what a deployment does to the people around it: they decide whether the next site says yes.Source 20
What would change our mind: if by our review in May 2027 trained systems work at sites they have not seen, without integration site by site, the second deployment stops being the test, and we will say so here.
Last reviewed We review this page every six months. When we change our view, the old view stays here, with the date and the reason.
What we bring
For companies, we do the commercial work, in three parts. The buyers: manufacturers, utilities, the public sector and defence, in the countries you are entering. The route in: procurement, local partners and compliance. The people who run it: senior leaders, part-time until your permanent team is in place, and a sales process your team can run. For a way to deploy again, the engineering is yours. We bring the commercial half: the integrators and channel partners, the contract and the price for the next site.
From the founders' careersSix markets opened across the Middle East and Asia-Pacific, and four global systems integrator partnerships signed.
Partnerships the founders have built
Global systems integrators, consultancies and an in‑country joint venture.
- Accenture
- PwC
- Deloitte
- Jamjoom Group
The founders have closed deals with teams at
Enterprise and government deals closed by the two founders.
- Airbus
- Hitachi
- Panasonic
- Samsung
- Moventia
- NEOM
When it is time to raise, we help you raise: the story, the numbers, the data room, an investor strategy and support through the round. Introductions and transaction mandates are arranged through regulated partners.
For funds and corporate venture teams, we scout and screen companies in this area against your thesis, and write up where each stands on the five tests. The note is factual and commercial, with no recommendation to invest or pass. The sample scouting note shows one, on a fictional company that inspects storage tanks with robots.
We also invest our own money when the fit is right, and can take part of a fee in equity. Size and terms are agreed case by case, in writing. Every stake is disclosed in writing to every client involved.
If we have an interest, we say so. If it conflicts, we step back.
One site running and the second not yet? Tell us where you are.
Newsletter
One email a month
Our newest notes, and what we are seeing in AI, machines at work and deep technology.
Sources
In the order we cite them, with the day we read each one.
Alexander Herzog, Kanishka Rao, Karol Hausman, Yao Lu and 36 others, Robotics: Science and Systems 2023 (arXiv:2305.03270), , Accessed , Back to the text, where source 1 is cited
2. Lessons learned the hard way: robotics (opens in a new tab)
Sam Smith-Eppsteiner, Innovation Endeavors (an investor's essay), , Accessed , Back to the text: citation 1 of source 2, a citation 2 of source 2, b
3. Physical AI: Taking human-robot collaboration to the next level (opens in a new tab)
Capgemini Research Institute (the report behind the survey), , Accessed , Back to the text: citation 1 of source 3, a citation 2 of source 3, b
4. Public procurement (opens in a new tab)
European Commission, Accessed , Back to the text, where source 4 is cited
5. The CEO's guide to physical AI (opens in a new tab)
Boston Consulting Group (a consultancy's analysis; it gives no source for the figure), , Accessed , Back to the text, where source 5 is cited
6. Regulation (EU) 2023/1230 on machinery (opens in a new tab)
European Agency for Safety and Health at Work, , Accessed , Back to the text, where source 6 is cited
9. Regulation (EU) 2026/1744 (Digital Omnibus on AI) (opens in a new tab)
EUR-Lex, Official Journal of the European Union, , Accessed , Back to the text, where source 9 is cited
Capgemini (a consultancy's survey of 1,678 executives), , Accessed , Back to the text: citation 1 of source 10, a citation 2 of source 10, b citation 3 of source 10, c citation 4 of source 10, d
11. Five million robots now operate in factories globally (World Robotics 2026) (opens in a new tab)
International Federation of Robotics (an industry body's count), , Accessed , Back to the text: citation 1 of source 11, a citation 2 of source 11, b
Crunchbase News (a data provider's count), , Accessed , Back to the text, where source 12 is cited
13. Robotics funding in Europe, 2026 (data to 11 November 2026) (opens in a new tab)
tech.eu Funding Explorer, , Accessed , Back to the text, where source 13 is cited
14. Manufacturers Need as Many as 3.8 Million New Employees by 2033 (opens in a new tab)
The Manufacturing Institute and Deloitte (an industry body and a consultancy's projection), , Accessed , Back to the text, where source 14 is cited
15. Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: a Survey (opens in a new tab)
Wenshuai Zhao, Jorge Peña Queralta and Tomi Westerlund, 2020 IEEE Symposium Series on Computational Intelligence (arXiv:2009.13303), , Accessed , Back to the text, where source 15 is cited
Arun N. Sivakumar and five others, IEEE ICRA Workshop on Field Robotics 2024 (arXiv:2404.17718, a workshop paper), , Accessed , Back to the text, where source 16 is cited
Symbotic (the company's own results, as filed with the US Securities and Exchange Commission), , Accessed , Back to the text, where source 17 is cited
Symbotic (the company's own results, as filed with the US Securities and Exchange Commission), , Accessed , Back to the text, where source 18 is cited
Stanford Institute for Human-Centered Artificial Intelligence, , Accessed , Back to the text, where source 19 is cited
20. Industrial Robots, Workers' Safety, and Health (opens in a new tab)
Rania Gihleb, Osea Giuntella, Luca Stella and Tianyi Wang, Labour Economics 78 (first an NBER working paper, 30180), , Accessed , Back to the text, where source 20 is cited