Humanoid robots are moving from viral demos to real-world jobs, but who will actually win the race? We have amazing demos from companies like 1X, Figure, Apptronik, Tesla, Agibot, Unitree, and more. But will they all be winners?
In this episode of Humanoid Mega Hub Live, I sit down with Bessemer Venture Partners VP Alexandra Sukin to break down the latest Bessemer robotics report and explore the future of humanoid robots, physical AI, and automation. We discuss the companies making the biggest moves in humanoids, why reliability matters more than impressive demos, the massive opportunity in manufacturing, logistics, healthcare, and the home, and whether advances in AI are happening fast enough to bring humanoid robots into everyday life.
Check it out here:
Alexandra also explains the biggest challenge facing robotics today: data. While large language models benefited from internet-scale training data, robotics companies are racing to build world models, synthetic environments, and new training approaches that can help robots operate safely and reliably in the physical world.
We talk about Boston Dynamics, Figure, 1X, Agility Robotics, Physical Intelligence, Waymo, robotics investing, robotic hands, dexterity, deployment challenges, labor shortages, and the trillion-dollar opportunity that has venture capitalists pouring billions into the space.
And we end with this: will humanoid robots arrive in our homes in the next few years? Or are we still much further away than the demos suggest?
Topics covered:
- The current state of humanoid robotics in 2026
- Why deployment matters more than demos
- Which humanoid companies stand out today
- Manufacturing, logistics, healthcare, and home robotics
- The race to build human-like robotic hands
- Physical AI, foundation models, and world models
- The robotics data bottleneck
- Venture capital’s role in shaping the industry
- Geopolitics and the global humanoid robotics race
- When consumers might finally get a humanoid robot at home
Transcript: who will win the humanoid robot race?
John Koetsier:
Who is going to win in humanoid robotics?
Hello, and welcome to Humanoid Mega Hub Live. My name is John Koetsier. I’m tracking over 400 companies that are making humanoid robots, and another 200-plus that are suppliers. And guess what? They’re probably not all going to win.
To chat about the future of the humanoid robot space, we’re joined by Bessemer VP Alexandra Sukin. She’s an active venture capitalist in the physical AI space, and they just released a massive robotics report that I covered for Forbes.
Welcome, Alexandra. How are you doing?
Alexandra Sukin:
Good. Thank you so much for having me.
John Koetsier:
Super pumped to have this conversation. I want to get into a lot of stuff. We’ll talk about the winners, companies that stand out, verticals, whether we can scale this or not, hands, some of the key functionality and features, and other things like that.
But maybe let’s start with the big picture. As you look at the humanoid robot ecosystem and the space in 2026, what do you see? What’s the state?
Alexandra Sukin:
We’re seeing it evolve really quickly. One thing that I found really interesting was the release of Boston Dynamics’ Atlas demo, which I’m not sure if you saw recently, but it was this amazing video of it lifting a fridge and then very lightly placing it on a desk while an engineer watches it.
We’re just getting these jaw-dropping demos of capabilities for a humanoid to do something like lift a super-heavy object and, conceivably, in the video, there’s not much human intervention in some of the tasks that they’re taking on.
This builds on some of the earlier releases that we’ve seen. The Neo robot from 1X was showing some pretty amazing capabilities doing chores around someone’s home. Even some of the PI demos that have been released with some of their recent models, where, while it’s not a full humanoid, you’ve got a bimanual, or two-arm, setup.
We’re in a kitchen watching a robot cook and cut things, then make things that are hot and moving in a pan. We’re just getting these amazing demos that are emerging from these labs.
John Koetsier:
Absolutely, and it’s exciting.
We’re also seeing robots get jobs, and that started probably two years ago. I talked about how Agility Robotics was the first one that had a paying gig in an actual company, and it’s been followed by many more since then.
I just interviewed Agibot’s president for embodied systems, and he was saying they’ve shipped more robots than anybody globally. He’s saying that’s where we’re going: deployment over demos. That’s very cool to see as well.
The demos are exciting because they show us what’s going to happen in the future, and then we hope the reliability and longevity will be there.
When you look out at the competitive landscape, and you mentioned a couple of names already, which companies stand out for you?
Alexandra Sukin:
There are so many companies now competing in this space, but I would say the ones that stand out to me are the ones actually getting from a demo to the real world. That implies they’ve unlocked some kind of unique technical capability that lets the company operate in a home or in a setting where they can actually perform at super-high reliability.
Why does this matter? If you think about talking to an LLM and it gets something wrong, let’s say when LLMs were first released, it gets a marketing email wrong 70% of the time. That’s okay because you can edit the marketing email.
But if you think about robotics, the bar for reliability is so high. You can’t have a robot come into your house, break all your dishes, or step on your cat.
You can imagine all the things that would happen when robots actually get into the real world and why reliability matters so much.
Some of the companies I think are interesting are starting with narrower tasks and really focusing on reliability issues. What’s interesting about some of Physical Intelligence’s research is that they’re really focused on benchmarking their models and making sure they can demonstrate that once they get out of the lab and into the real world, they’ll be able to perform at high levels of reliability.
I would also say there are some interesting companies that are not trying to go into the home first. Instead, they’re saying, “We’re going to go into other verticals, like manufacturing, perfect our process there, and then move into the home.”
My guess is that some of Boston Dynamics’ demos are showing capabilities related to these non-home use cases, and then they’ll eventually get to the home.
John Koetsier:
That’s such an interesting point because there are vastly diverging opinions on that.
A week and a half ago, I talked to Sanctuary AI CEO James Wells, and he was saying the home is tricky. You can step on a pet, break something, and you need even higher reliability.
Manufacturing lines want five nines of reliability, but if you don’t have that, you can actually be dangerous in the home.
On the other hand, I talked to Dar Sleeper from 1X two days ago, and he’s had one in his home for months. He’s one of a very small subset of people on Earth who have experience having a humanoid robot in their homes.
The interesting thing is that he says it’s instantly the weirdest thing in your home, and you get used to it in about five minutes. He experiences that every time new visitors come over.
That’s going to change our world a little bit.
Alexandra Sukin:
Absolutely. I totally agree. Once you get used to it, you won’t go back.
Once we get laundry-folding robots in our houses, we’ll never go back to folding laundry or putting laundry in the washer and dryer ourselves.
Part of what’s driving excitement about robotics is that it’s easy to imagine how transformational it’s going to be. If it works, and I’m excited about the potential for it to work, hopefully sooner rather than later, it will have a huge impact on how we live our lives.
It’s not just doing chores around the house. It’s taking care of people as they get older or filling jobs where humans don’t want to do the work or where it’s difficult to find people who want to do those jobs.
John Koetsier:
I just did an interview with Gadi Schwartz for NBC News Now, and he was asking about having a humanoid in the house and whether that would be good for parents.
I said absolutely. I’ve been a parent of a young child, and there’s so much to do. There’s cooking, cleaning, childcare, feeding, washing, bathing, dressing.
I wouldn’t want the robot to take care of the baby or the children, but if the robot could take care of everything else, that would free me up to spend more time with my child. I could be more human in my interactions and less rushed.
That’s at least what I told him. I don’t know if that resonates with you.
Alexandra Sukin:
It absolutely does.
When I’ve told people about potential applications for robotics, they really seize on these kinds of things that are difficult, repetitive, or physically demanding.
From a manufacturing perspective, humans often have to lift very heavy things repetitively. If that task could be done by a humanoid, that would be pretty impactful.
I think those are going to be the places where we see humanoid robots first: really high-need use cases.
John Koetsier:
Let’s talk about verticals.
Home is one. Logistics is another. We see a lot of logistics robots. Figure recently demonstrated 80 to 90 hours of continuous operation. It was a battle between their humanoid, Figure 03, and a human. The human won, but just barely.
Brent Adcock said, “Last time a human wins.”
We’ll see if that’s true.
Logistics is big. Manufacturing is another. That’s where Agility Robotics is, where Apptronik is, and where several others are.
You mentioned Boston Dynamics’ Atlas. There’s healthcare and eldercare. We haven’t seen many humanoid platforms there yet, but we’ve seen wheeled platforms being deployed in hospitals.
Eldercare will come. Maybe even farm work. Farm workers are incredibly scarce in some regions.
What verticals stand out to you?
Alexandra Sukin:
Part of what’s interesting about these vertical applications is that they don’t necessarily require a humanoid.
Many of these applications may be solved by non-humanoid robots first, and maybe by humanoids later.
As investors, what’s interesting is that you don’t necessarily need a humanoid robot that comes with all sorts of complexity. It has fingers that can break. It has to balance on two legs and carry heavy things.
There’s a tremendous amount of engineering complexity that comes with a humanoid.
If you’re doing picking and packing in a warehouse, moving things in manufacturing, or driving an autonomous vehicle, you don’t necessarily need a humanoid.
While I think humanoids are going to be impactful, they’re not the only robotics form factor we’ll see. In some verticals, the humanoid may not be the best form factor.
Manufacturing is exciting because there’s a big need and labor shortages that have to be addressed. Warehousing is another.
At Bessemer, we’re investors in Waymo, which hopefully you’ve used.
John Koetsier:
I have.
Alexandra Sukin:
It’s a magical experience.
John Koetsier:
There were no unicorns, but it was nice.
Alexandra Sukin:
Exactly. It’s incredible. The steering wheel moves while you’re sitting in the passenger seat watching the car drive itself.
There will be lots of other spaces as well.
You mentioned hospitals. Today it’s primarily wheeled robots moving supplies from one floor to another. Hopefully, eventually, we’ll get humanoids in there that can do things like moving people around and performing more complex tasks.
John Koetsier:
Hospitals are interesting.
One of the early use cases I’ve seen is pill delivery, where you’ve got a robot carrying things around. That’s a relatively defined use case.
Others are heavier robots lifting people. Nurses may weigh 130 pounds, and they’re expected to help move a 250-pound linebacker who’s blown out his knee.
That’s challenging.
Healthcare costs are also way too high globally, so if robotics can help and eventually reduce costs, I think that’s great.
One thing I wanted to discuss is hands because we’ve seen tremendous development there lately.
I wrote about Kybra Labs. Genesis AI had a cool demo cooking eggs and slicing tomatoes. Then I chatted with 1X. I couldn’t record it, but they showed me the hand they’ll be shipping.
It blew my mind.
One critique of humanoid robots right now is that they’re slow. They reach out to shake your hand and everything happens at half speed or quarter speed.
This hand was unbelievably fast. Twenty-two degrees of freedom, but not just degrees of freedom. Twenty-two actuated degrees of freedom, including adduction and abduction.
It’s an incredible hand, and I don’t think they’ll exhaust its functionality with the software they initially ship. I think there’s a long runway ahead for making it even more useful.
Alexandra Sukin:
That’s amazing.
I agree with you that a lot of robotics demos are very cool, but they’re often shown at accelerated speed. If you go see them in the real world, which I have, they don’t always perform at the speed you’re hoping for.
Part of that is safety. Part is the speed and accuracy of the models themselves.
That goes back to the deployment gap and how much still needs to be solved before robots actually enter our homes.
If a robot can do your dishes but takes twenty times longer than you do, that’s probably not enough value to justify the cost.
They have to solve that problem.
To your point about hands, that’s a super exciting area.
Part of the reason is that some foundation model companies believe that to develop truly granular and dexterous hand movement, you need to closely couple a training glove, the robotic hand, and the model.
You might perform tasks such as picking something up, turning it over, or zipping something. That data gets tied directly into the robotic hand and the model controlling it.
The goal is to close the loop between training, real-world performance, and model development.
If you don’t have all three tightly integrated, it’s difficult to know whether a problem is in the hardware, the training paradigm, or the model itself.
I think that’s part of why companies like Genesis have been able to raise large rounds based on that hypothesis.
John Koetsier:
I agree.
One other thing about that hand: it solves the mannequin-hand problem.
You know how robotic hands are often stuck in a rigid position and only close a little bit? This one has a much more natural resting posture.
That should help reduce the uncanny valley effect and make interactions feel more natural.
Another thing that’s important is noise.
In a factory, you don’t care if it’s noisy. But if you’re going to have it in your home, a quiet humanoid is much better.
Hospitals are another example. They’re already noisy, and you want to minimize that and let people rest.
We’ve talked about foundation models a little. We’ve come a long way with LLM-based approaches.
Early on, they were probably wrong more often than they were right. Now they’re probably right more often than they’re wrong, but they’re still probabilistic rather than deterministic.
That means you can’t guarantee outcomes.
Do we need fundamentally different approaches in physical AI?
Alexandra Sukin:
In the early days, there was a lot of excitement around taking the paradigms from LLMs and applying them directly to robotics.
The first roadblock was data.
With LLMs, we had internet-scale datasets. In robotics, that doesn’t exist.
There’s no massive dataset of robots performing tasks across different form factors, along with reward signals indicating whether outcomes were correct or incorrect.
People have tried to work around that. One approach is using video data as part of the training set.
Others are building data-collection centers where people use gloves, end effectors, or grippers to simulate robot tasks and generate training data.
But we’re finding that these datasets are nowhere near as rich as what we had for LLMs.
There’s simply much more complexity when you’re trying to deploy models in the physical world.
A major focus now is overcoming the data bottleneck.
One strategy comes from autonomous vehicles: world models.
Waymo collected huge amounts of real-world driving data. They used that data to build simulations of the real world and then trained vehicles inside those simulated environments.
That’s one approach robotics companies are trying now.
We’re seeing world-model companies emerge with a focus on robotics. Examples include General Intuition and World Labs, led by Fei-Fei Li.
Another approach is synthetic data generation, often combined with reinforcement learning to make simulated environments more valuable.
That’s probably the number-one challenge model companies are trying to solve today.
John Koetsier:
It really is.
I chatted with Edward Johns, director of the Imperial College Robotics Lab in London, and they’ve developed something that taught a robot a thousand different real-world manipulation tasks in less than a day from human demonstrations.
There’s real progress happening, and it’s super promising, but we clearly need more.
That’s a good segue into the last topic I want to discuss: investment.
This is the business you’re in. You’re deploying capital to achieve outcomes.
Do we need significantly more investment in robotics, specifically humanoid robotics? Have we got enough?
What’s your thinking?
Alexandra Sukin:
It’s hard to say how much money we’ll ultimately need to achieve foolproof, real-world deployments.
One challenge in robotics is that we’re still uncertain about timelines and capital requirements.
We’re figuring that out in real time as we discover which techniques work, how scalable they are, and how long it takes to prove them.
But when you think about the market opportunity, if we get humanoids in the next few years, we’re talking about trillions of dollars in global labor spend.
The best companies in this category could become incredibly large businesses, disrupting labor markets that traditional software simply can’t touch.
They’ll have physical deployments and potentially enormous moats.
That’s driving a tremendous amount of venture excitement and investment.
It’s hard to say whether there’s enough money in robotics, but the size of the opportunity absolutely justifies the excitement.
For the most successful companies, the prize is going to be massive.
I suspect we’re not done seeing robotics companies raise very large financing rounds.
John Koetsier:
It’s always interesting.
Sometimes VCs try to anoint a winner by massively funding them so nobody else can compete.
Other times, companies get flooded with cash and become inefficient.
There’s no single formula for success.
It is interesting to see Humanoid, the UK humanoid robotics company. They recently signed a 1,000-robot deal with Schaeffler, a large German industrial manufacturer and actuator supplier.
Part of that agreement includes Schaeffler supplying actuators.
The minimum volume in that contract suggests Humanoid believes it will ship 100,000 robots by 2031.
That’s not very far away.
We also have countries that see humanoid robotics as critical to their economic future and geopolitical power.
Korea recently invested the equivalent of $30 million in the humanoid space to ensure future labor capacity.
China certainly views humanoids as part of its long-term competitiveness strategy.
Other countries are beginning to recognize the same thing.
The players in the United States certainly understand it. I’m not sure the government does yet, but the industry definitely does.
There’s a lot happening.
Alexandra Sukin:
I definitely agree.
These are going to be enormous categories and markets.
You can imagine the advantage a country would have if it could supplement labor, build more, and provide more services with robots.
You can also imagine the disadvantage if another country couldn’t.
People are waking up to that possibility and realizing that robotics could become both an economic engine and an imperative for future growth.
John Koetsier:
I think I’ll end with one question.
When are you going to get a humanoid in your home?
Alexandra Sukin:
I wish I knew the answer to that question.
I’ll certainly be trying to find the answer because it would completely change our investment strategy.
In some ways, I compare it to Waymo.
The company was founded in the mid-to-late 2000s, and now we’re seeing Waymos on the road in 2026.
I’m hoping the timeline for humanoids is meaningfully shorter than the timeline it took autonomous vehicles to go from concept to large-scale deployment.
We’ve already got LLMs, all the research surrounding them, and the lessons learned from autonomous vehicles.
There are a lot of reasons to believe that progress could happen faster this time.
If it ends up being significantly less than fifteen years, that would be amazing.
John Koetsier:
Very cool. Thank you so much for your time.
Alexandra Sukin:
Yeah. Thank you so much, John.