00:00Humanoid robots are moving out of the lab and into the real world, with deployments expected to surge in the
00:05years ahead.
00:05Our next guest thinks the industry is entering a major scale-up phase, but says a shortage of real-world
00:11training data remains a key hurdle.
00:14Zunita Todorova is the head of thematic thick research at Barclays, back on the show.
00:18I think let's start by, Zunita, you reminding us what your kind of longer-term forecast is for the addressable
00:26market or size of this humanoid robotics market globally.
00:31Hi, it's great to be on the show again.
00:34Right, so humanoid robotics is scaling really fast.
00:37We estimate that by 2035 it could become a $200 billion market and it could be part of a much
00:43bigger physical AI market worth a trillion dollars.
00:47Now, if we want to put some numbers in terms of deployment units behind these forecasts, our forecasts look like
00:53this.
00:54So in 2026, we expect that we're going to see about 60,000 new humanoid robots deployed worldwide.
01:00That's a massive scale-up compared to what we saw last year when we just had 15,000 robots.
01:06But I think more importantly, we're just scratching the surface of what's possible and we're going to see much more.
01:11And according to our latest research, by 2035, we could see about 13 million new humanoids deployed each year across
01:19two major markets, the U.S. and China.
01:23And I think that if we continue to see the same technological progress as we have seen so far in
01:292026 and if investor appetite for the theme continues to hold strong, I think we could easily see about 30
01:37million humanoids in operation by 2035, which is quite a sizable fleet if you think about it.
01:44Okay, right now we're looking at a video of Tesla's Optimus.
01:47And what I find so interesting about this is the number that you shared of 60,000 units of humanoid
01:53robotics globally in this year.
01:55Don't most of those come out of China?
01:58Well, about 85 to 90 percent of us will likely come out of China.
02:03That's for sure.
02:04China seems to have the edge right now.
02:05But I think there are small ecosystems popping up worldwide.
02:10And I think the U.S. is very quickly narrowing the gap in terms of maybe not so much in
02:16terms of the number of units that are deployed out there, but certainly in terms of the expertise of the
02:21robots and what they can do in the real world.
02:24The limiting factor that I'm being told on a near daily basis by industry is real world data.
02:31So what we mean by that is the ability of the robot to interact with the humanoid robot, to interact
02:37with its immediate physical environment, be that a factory, other industrial setting, whatever it is.
02:43One solution seems to be synthetic data.
02:46Could you just quantify the challenge and how you see the market trying to address it?
02:51Right.
02:52I absolutely agree.
02:53I think that the number one challenge right now is not really demand for humanoids.
02:56I think that there is strong demand for humanoids in all kinds of applications.
03:00The biggest challenge is the lack of physical AI data.
03:03And I think that this is something that people find very surprising because the reality is we have tons of
03:08digital data, but it's just not useful for training humanoid robots.
03:12And I do think that the data necessary to train and to scale, more importantly, a truly general purpose humanoid
03:19robot does not exist on the Internet.
03:22It needs to be retrieved in the real world, in the physical world.
03:25And that could mean something like recording humans doing everyday tasks, from doing the laundry to household work and even
03:32deliveries, and this way generating data that is required to train these very complicated general purpose models.
03:39Now, you mentioned simulation, and I think this is definitely one solution, one bet to this challenge.
03:45However, I do feel that us humans, we live in a very fuzzy, very unstructured world, and simulation has its
03:54advantages.
03:54It's definitely a cost-effective option, but I think ultimately simulation will hit a wall in terms of how much
04:00of that fuzziness, how much of that messiness it can capture.
04:04So, ultimately, I think we'll have to lean on the real world and the physical data that's out there.
04:10And we definitely see more and more companies popping up, which reflects a general industry-wide push towards assembling this
04:18data that is going to feed models for the next five to ten years.
04:22You know, the most bullish on the humanoid industry put a lot of faith in the digital twin,
04:28the idea that a digital twin can ground the humanoid and the real-world physics around it.
04:33That's on the software side.
04:34On the hardware side, is there even a supply chain for the humanoid robot anywhere on this planet?
04:43It's beginning to be put together.
04:46It does not exist at the scale that we would think when it comes to other technological products, such as
04:52EVs, for example, other types of industrial robots.
04:55I think that the reason is that this is going to come all the time.
04:58We just need to find the skill application for humanoid robots.
05:01They are the most efficient functional form, and I think I'm confident that the supply chain is going to emerge
05:06around it.
05:07I think that if you think about it at a high level, we don't need to reinvent everything from scratch.
05:12I think that there are lots of complementarities between, I would say, electric vehicles and humanoids,
05:19especially when it comes to actuators and motion systems.
05:24So I think we can borrow a lot from that.
05:27It's definitely not going to be plug and play, but I think that the basics of that exist.
05:33And I think as the technology improves, the supply chain is definitely going to scale accordingly.
Comments