Humanoid AI’s robots in operation
John Koetsier
Two years ago Humanoid did not exist. In July it closed a $152 million Series A at a $1.35 billion post-money valuation, bringing the London-based company’s total funding to about $270 million. I had a chance to catch up with the founder and CEO Artem Sokolov, who started the company in May 2024 with roughly $30 million of his own money, to ask about Humanoid’s plans, go to market strategy, order book as well as its relatively unique path to scaling production and how the company will approach competition.
Sokolov’s grandparents worked in a jewelry factory. He later ran one himself and watched his own employees do repetitive, physically punishing work: one of his motivations for building humanoid robots that can automate what humans should not have to do.
Humanoid now employs more than 250 people across London, Boston, Vancouver and San Diego, and it moved fast to get to this point, building the wheeled version of its HMND 01 platform in seven months and the bipedal version in five. That’s impressive: the industry norm, even now when you can get so many components just by ordering them, is closer to 18 to 24 months.
Part of the reason: hiring people from other humanoid robot companies. Chief product officer Sotirios Stasinopoulos told me Humanoid has hired more than 50 veterans from Boston Dynamics, Sanctuary AI, Apptronik and 1X, supplying what he called “second-mover advantage.”
Humanoid leads with a wheeled bimanual robot — humanoid torso, head, two arms, wheels instead of feet — saying it covers roughly 90% of industrial use cases without the cost, complexity and failure modes of bipedal locomotion. The Alpha units handle about 10 kilograms total, five per arm; Beta is targeted at double that when it arrives in Q4 2026. Humanoid’s AI stack, a four-layer framework the company calls KinetIQ, is developed in house, and Sokolov says its reinforcement learning layer, KinetIQ Ascend, already runs some tasks at about 1.5x human speed with success rates up to 99%.
A key difference with many other humanoid robot companies: Humanoid doesn’t intend to build robots in-house at scale.
Robert Bosch Robotics GmbH is its contract manufacturer for Europe, contributing design-for-manufacturing oversight and supply chain muscle, with the option of putting Bosch actuators, drives and sensors into future versions. Schaeffler is simultaneously an investor, a supplier committed to more than half of Humanoid’s actuator demand through 2031 — a seven-digit number of units — and its largest publicly announced customer, with a four-digit robot deployment across its global plants targeted by 2032. Humanoid has also run pilots with Siemens, SAP and Ford, says it has more 34,000 pre-orders representing a massive $2.4 billion pipeline.
That, of course, is the critical part: transitioning pre-orders into actual orders and then cold hard cash. Humanoid’s roadmap — several hundred units with Bosch in 2027 scaling to a massive 150,000 in 2030 — is aggressive even against Bank of America’s estimate that the global market will approach a million units a year in the early 2030s.
So I asked Sokolov about all of it: pricing, pre-orders, the partner-heavy manufacturing model, and what winning actually looks like from London.
John Koetsier: First, congratulations on the raise. Now the real job begins?
Artem Sokolov: Absolutely. Raising capital is an important milestone, but it’s not the destination. It gives us the resources to execute.
This investment is significant for several reasons. It makes Humanoid the largest humanoid-first robotics Series A in Europe, reaching a $1.35 billion valuation just two years after founding. More importantly, it allows us to accelerate the next phase of the company: expanding our proprietary AI platform, scaling commercial deployments with global industrial leaders, and bringing our next generation of industrial robots (Beta) to market faster.
I also think it sends an important message about Europe. For years, people have questioned whether globally competitive technology companies could be built here. We believe Physical AI is an area where Europe has genuine strengths, combining world-class engineering with one of the world’s strongest industrial ecosystems.
For me, though, this mission has always been personal. My grandparents spent their lives working in a jewelry factory. Growing up, I hardly saw them because they were almost always at work, and they never really had the chance to experience the world outside their walls. Years later, after running my own jewelry manufacturing business, I found myself watching my employees face many of the same realities. It reinforced something I’d believed for a long time. If technology can take over repetitive, physically demanding work, people should have the opportunity to spend more of their lives doing something else. That’s ultimately why I founded Humanoid.
Now it’s about turning that vision into reality. We’re continuing to advance reinforcement learning, expand KinetIQ, our AI brain, increase real-world training, introduce Beta in Q4 2026, and build the team that will support the next stage of commercial deployment. Everything we’re doing is aimed at delivering measurable value for industrial customers while freeing people from repetitive and physically demanding work.
John Koetsier: I’ve seen that you’ve got 34,000 pre-orders. First, is that accurate? Second, what is the timeline for these deployments?
Artem Sokolov: Yes, that’s correct. We’ve secured more than 34,000 non-binding pre-orders for our wheeled industrial robot, representing approximately $2.4 billion in our commercial pipeline. While these are not committed orders, they are strong indicators of customer interest and reflect the strongest pipeline in the market at this stage of development. That commercial traction is also reflected in our landmark deployment agreement with Schaeffler to deploy a four-digit number of wheeled robots across its global manufacturing facilities.
From the beginning, we’ve focused on industrial applications because that’s where we see the strongest demand today. Manufacturing, logistics, warehousing and energy face real labor shortages, and customers already understand the value of automation. That clears the path for Physical AI to deliver meaningful outcomes.
Our approach is phased. We typically begin with a short proof of concept lasting two to four weeks on-site, followed by a longer pilot lasting three to six months and, ultimately, commercial deployment. For a company building entirely new technology, like ours, those early pilots are incredibly important. They let us validate our hardware, software and AI in real operating environments and understand how customers use the robots. Also, they allow us to collect the data that helps improve every new generation of our platform.
I think one of the biggest challenges for the industry now is bridging the gap between a successful pilot and deploying reliable fleets across multiple facilities. Our customers need confidence that the robots can integrate into their day-to-day operations and keep performing reliably as deployments grow, as this is what gives them a clear return on their investment.
As for our next milestone, we’re planning the rollout of our Beta robots in Q4 2026, followed by broader commercial deployments across our industrial customer base as we continue scaling production with our contract manufacturing partners.
John Koetsier: You have a different R&D and production and servicing model than many other humanoid companies, with partners like Bosch and Schaeffler taking on big roles. Why? How do you see this evolving going forward? Is this model better/faster/scalable compared to keeping things in-house?
Artem Sokolov: For us, being a full-stack company means owning the technology that defines the robot. We develop the core AI, hardware and software ourselves because that’s where we believe we can create the most value. At the same time, companies like Bosch and Schaeffler have spent decades building world-class manufacturing capabilities and deep industrial expertise. Working together allows each of us to focus on what we know best.
The biggest advantage is how quickly we can learn and improve. Because we own the core technology, we can respond rapidly to what customers tell us and continuously improve the robot as it moves into more factories.
These partnerships also strengthen the supply chain around the robot. Humanoid systems depend on specialized components such as actuators, and working closely with established manufacturing partners gives us greater confidence in quality and availability. I think that will become increasingly important as the industry moves from pilot programs to large-scale deployment.
More broadly, I think this reflects one of Europe’s biggest strengths. As I mentioned earlier, the continent has an extraordinary industrial base and generations of engineering expertise. Building within that ecosystem gives us access to trusted supplier relationships and partners who understand what it takes to deploy technology at industrial scale.
As the market grows, I believe this approach becomes even more valuable. Building a great robot is one challenge. Manufacturing thousands of them and deploying them across multiple sites over the long term is a very different one. Combining our in-house technology with experienced industrial partners gives us the strongest foundation for that next stage.
John Koetsier: Where do you plan to fall in terms of pricing? There’s pressure on the low-end from Chinese competitors, and some very capable robots in the US, not to mention Neura … where do you see Humanoid sitting?
Artem Sokolov: Cost is obviously important, so we think about it from the very beginning. Every engineering decision we make considers how to reduce the bill of materials and the total cost of ownership as we scale production.
That said, industrial customers don’t choose robots based on price alone. They want a robot that integrates into their existing operations, performs reliably, and delivers a clear return on investment over its lifetime. That’s the equation we’re focused on.
We’ve also learned that customers don’t all want to adopt robotics in the same way. Some want to own the robots as part of their long-term industrial infrastructure. Others prefer Robot-as-a-Service because it lowers the upfront investment and gives them more flexibility as they expand deployments. We think both approaches will play an important role in the market, so we’re supporting both from the beginning.
Our Robot-as-a-Service offering includes deployment, operational support, and access to teleoperators who can step in remotely whenever needed. It’s particularly well-suited to retailers and e-commerce operators. Direct purchases tend to make more sense for customers with stable, repetitive workflows, such as automotive manufacturers, where robots become part of their long-term operations. Across both models, we’re targeting a payback period of around one year.
As manufacturing scales, hardware costs will naturally come down. Our focus is making sure customers see value from the moment the robot goes into operation and throughout its lifetime.
John Koetsier: Talking about competition … there’s a ton of it, and a lot of it is very well-funded. What’s your path to winning, and what does winning mean for Humanoid?
Artem Sokolov: Competition is a good thing. It shows that humanoid robotics is moving from research into a real industry.
From our perspective, the biggest opportunity over the next several years is industrial. Manufacturing, logistics, warehousing, and energy already face labor shortages, and companies are actively looking for practical ways to automate repetitive work. That’s why we’ve focused on B2B from the beginning. We think that’s where Physical AI can deliver the fastest and most measurable impact.
The same thinking led us to start with a wheeled robot, which can address around 90% of industrial use cases. It gives customers a practical way to introduce Physical AI into existing operations while allowing us to keep improving the platform through real-world experience.
Being a full-stack company is a big part of that. We own the core technology, so every deployment feeds back into the platform. The more robots we deploy, the more we learn, and every new deployment benefits from that experience.
We’ve already completed successful pilots with companies including Bosch, Schaeffler, Siemens, SAP, and Ford, and our partnership with Schaeffler continues to expand commercially. Every project gives us a better understanding of how customers use the robots in real production environments and helps us prepare for larger deployments.
For me, winning is about becoming the company industrial customers trust when they’re ready to deploy Physical AI at scale. If we keep solving real problems for our customers, I think everything else follows.
John Koetsier: What are the biggest jumps in performance you’re expecting from your next models?
Artem Sokolov: I think the biggest improvements will come from AI. Hardware will continue to evolve, but reinforcement learning is where we’re seeing the fastest progress.
We’re investing heavily in KinetIQ Ascend, our reinforcement learning approach, and in fleet learning, where every deployment helps improve the next one. We want robots to learn new tasks more quickly and adapt naturally as their environment changes.
Every deployment teaches us something new. As more robots enter real industrial environments, they generate the experience that helps improve the platform for the entire fleet. That’s one of the biggest advantages of deploying commercially from an early stage.
I think we’ll see major progress in manipulation, dexterity and autonomy over the next few years, but the biggest change will be how quickly robots learn. Once they’re able to improve continuously from real-world experience, progress becomes much faster.
John Koetsier: For go-to-market, are you planning to sell robots outright, offer robots as a service, or some hybrid model?
Artem Sokolov: We’re planning to support both because that’s what we’re hearing from customers. After speaking extensively with companies across our commercial pipeline, it’s clear there isn’t a one-size-fits-all solution. The decision often comes down to how each organization prefers to invest in new technology and how quickly they want to scale deployments.
Some customers want to own humanoid robots as part of their long-term industrial infrastructure. Others prefer Robot-as-a-Service because it lowers the upfront investment and gives them more flexibility as they expand deployments. We see value in both approaches because they remove different barriers to adoption.
More broadly, I think offering both models will help accelerate the market. Some companies are ready to make humanoid robots part of their capital investment today. Others want to gain operational experience before making that commitment. Supporting both gives customers the flexibility to adopt Physical AI at a pace that fits their business.
Our focus is making deployment commercially viable regardless of how customers invest. If the robots deliver reliable performance and a clear return on investment, I think both models will become a natural part of industrial automation.
John Koetsier: I chatted with your CPO about speed, and he mentioned that you’re aiming for better-than-human on many tasks. When will you achieve that?
Artem Sokolov: We expect our robots to reach human-parity speed and move beyond it, but the timing will vary by task. Humans are incredibly adaptable, and that remains a huge advantage in many industrial environments.
What reinforcement learning changes is how robots improve. Instead of programming every movement by hand, we train them in simulation and then continue refining them through real-world deployments. KinetIQ Ascend is a big part of that. It allows robots to learn directly from experience, so they become faster and more reliable over time.
We’ve already seen that in our own systems. In some reinforcement learning tasks, our robots are operating at around 1.5 times human speed while achieving success rates of up to 99%. Those results come from running thousands of training iterations before validating performance in real industrial environments.
I think the first areas where robots consistently outperform people will be repetitive industrial tasks that require the same movement over long periods of time. Reinforcement learning has already shown strong results there, and every deployment helps us improve the platform even further. Each robot contributes new data, and the entire fleet benefits from what it learns.
That’s how we move towards robots that can consistently exceed human performance on industrial tasks. Those are exactly the kinds of repetitive and physically demanding jobs people shouldn’t have to do, and that’s been our mission from the very beginning.
John Koetsier: How many humanoid robots will the industry have shipped by 2030? How many will be Humanoid humanoids?
Artem Sokolov: The industry is still at a very early stage, but the pace of development has been remarkable. We’re seeing customers move from exploring the technology to planning real deployments, and I think that’s what will define the next few years.
Bank of America estimates the global humanoid robotics market could approach annual production of around one million units in the early 2030s. This shows that adoption is accelerating as the technology matures and more companies begin deploying robots in real industrial environments.
For us, the roadmap is clear. We’re planning to manufacture 150,000 units in 2030, working with multiple contract manufacturing partners to reach that scale. Before that, together with Bosch, we’re already planning to manufacture several hundred robots in 2027. Each step gives us the opportunity to scale manufacturing, learn from deployments and prepare for the next stage of growth.
I’ve always believed this industry would be built through execution. You have to build the robots, deploy them successfully, learn from customers and keep improving the platform. That’s what we’re focused on every day, and I think that’s how we’ll earn our place in the market.
John Koetsier: Thank you for your time.
Be the first to comment