This week’s podcast is about the robot-native network Huawei and China Unicom built for the World Humanoid Robot Games (WHRG) in Beijing.
You can listen to this podcast here, which has the slides and graphics mentioned. Also available at iTunes and Google Podcasts.
Here is the link to TechMoat Consulting.
Here is the link to our Tech Tours.
Here are some slides from the presentation by China Unicom on 5G-A large uplink network.
Here some photos from the WHRG.
Disclosure. I have had a paid consulting relationship with Huawei in the past twelve months.
———-
Related articles:
- AgiBot’s Race to Embodied AI Robots at Scale (1 of 2) (Tech Strategy)
- Why I Like AgiBot (Podcast 290)
From the Concept Library, concepts for this article are:
- Robotics
- Embodied AI
- Robot-Native Networks
- 5G-A
From the Company Library, companies for this article are:
- China Unicom
- Huawei
———-transcribed below
00:05
Welcome, welcome, everybody. My name is Jeff Towson and this is the Tech Strategy Podcast from Techmoat Consulting. And the topic for today, Huawei and China Unicom have basically pioneered robot native networks at something that’s all over the news, which is the World Humanoid Robot Competition. It’s been in Beijing right now. Videos everywhere, robots running, jumping. It’s crazy. know, underneath that,
00:35
They’ve actually built something pretty important is mobile networks that can basically handle robots, which have very different requirements than humans and agents. So I’m going to talk about kind of what they did. I was out there at the competition. I was at the sort of briefing by a Huawei and China Unicom. I’m going to kind of lay out what they did because it’s really a good real world validation of something that’s going to be built within cities, starting with Beijing, I think.
01:04
So this was actually a really interesting use case of something I think we’re going to see everywhere, which would be robot native networks. Anyways, that’ll be the topic for today and why those are important, which I didn’t really appreciate until I was there and then I kind of got it. So anyways, that will be the topic for today. Let’s see, standard disclaimer, nothing in this podcast, my writing or website is investment advice.
01:26
The numbers and information from you and any guests may be incorrect. If used in an opinion express, may no longer be relevant or accurate. Overall, investing is risky. This is not investment, legal or tax advice. Do your own research. And with that, let’s get into the topic. Okay, so I was in Beijing this weekend going to the World Robot Conference, which was amazing, by the way. I’m going to write a lot about it, a lot of good videos.
01:52
Man, this is the year robots have arrived. I thought it was going to be the year of agents, which I guess it is. But in China, this is the year robots have arrived. Maybe not other parts of the world, but they were everywhere. It is stunning what’s happening. I’m going to write it up. Lots of videos, lots of pictures. Super important. And that was Saturday, Sunday. And Sunday afternoon, I bugged out over to the humanoid robot competition, which is north side of the city there where the old ice skating rink.
02:22
venue. That was amazing too. And I was there for sort of a briefing by Hua-Wen Unicom on what they were doing. Really fascinating. So I’m going to go through that and I’ll go through the overall conference in articles in an upcoming podcast. Well, let me talk about sort of what’s going on the network, the mobile network side. A lot of concepts for today. We’ll put these in the concept library. Number one, robot native networks. A couple months ago, I was out in
02:52
in Barcelona at the Mobile World Conference there and I talked a lot about agent networks, agent native networks. Really this idea that like look mobile networks they were built for humans, eight billion of us, we don’t use them that often, we stream a lot of video. Once you move to agents you’re talking, I’ve been using the number at 800 billion. So from eight billion humans to 800 billion agents using these networks all the time and these are intangible creatures that are always gathering
03:21
perception information, uploading it, doing copy-tube, blah, all of that. Well, it turns out robot native networks are a new thing as well because there’s let’s say 100,000 robots in the world right now, but there are 70 million cars. It’s reasonable to assume we’re going to have 100 million robots at some point in the near future. They have very different requirements for what the mobile network needs to do.
03:50
In particular, they need a lot of uplink capacity and speed. know, robots are sensing. They got video cameras. They’re seeing things. They’re uploading it. That’s when the computation is happening. What should we do? Then it’s downloaded in motor action. So the uplink capacity is huge. And the end-to-end latency is big as well. General numbers that were thrown around.
04:16
Uplink speed needs to be about 1 gigabit per second. I it could be 500, but let’s say 1 gigabit per second. Okay, that requires 5G advanced. And end-to-end latency under 30 milliseconds, which is actually not super fast. You don’t need ultra low latency for the robot competition, which I’ll talk about. You do need ultra low latency, but for this you don’t maybe need it that fast. So maybe it’s less.
04:46
The robot latency for the competition for at least the first day was clocked in at about 18 milliseconds. So for normal usage around town, you probably don’t need that speed. No one’s competing in a soccer match. okay, latency has to be pretty fast. The upload capacity has to be huge. That’s a very different network than humans watching videos.
05:13
I’ll go into that. So one, robot native networks, big idea. And then really this idea of AI infrastructure. When we talk about AI infrastructure, we talk a lot about GPUs, memory, data centers. We really need to include networks into that discussion. And we talk about power and batteries a lot too. Networks are a big part of that. For agents and for robots, yeah, that’s part of AI infrastructure. Okay, let me get into, let me just go through the…
05:42
the competition first before I get into the network stuff. The competition, man, I think this was the breakout year. This was the second annual humanoid robot competition. So it’s only humanoids, right? Not dogs, not drones, all that. And they are generally small or medium size. Medium size tends to, I don’t know the exact number they use, medium size tends to be about 170 to 180 centimeters, basically my height, smaller down in the 150 range. There are
06:12
Other categories of ones, that’s the height, but then they also have different versions that can do weight lifting as opposed to running. But humanoids, two size categories, and they all have to be autonomous, which means you can’t robot control them anymore. OK, so the network is super important. This is the second one of these they’ve done. Last year was a year ago.
06:37
And I think this was the breakout year. Nobody really paid too much attention to last year. This one’s everywhere. The videos are everywhere. They broke world record by humans. They broke them left and right, which hadn’t really happened that much before. So, you know, let’s say vertical leap, just jumping straight up. They just broke the human record. The robot that jumped, I think it hit 2.7 meters higher than any human can jump.
07:07
Just standing jump. Long jump. Running. We see this in the Olympics. Running, long jump. Almost 8 meters. I don’t think any human’s even close to that. Running, they pretty much started crushing all the human. Usain Bolt, his records got beat. 400 meter, 100 meter. I don’t know if they broke all of them. They broke a lot of human records in running. Okay, that was impressive. And the videos are funny because when they run, they run kind of funny.
07:36
Other areas, not so much. Tennis, Kung Fu, no, they’re not even close to human levels, but other levels, pretty impressive. So let’s say definitely the breakout year. I had sort of four takeaways. Number one from the competition. Number one, robots run really funny. I’ll put the video in the show notes. They lean way forward and they swing their hands back and forth.
08:07
People are calling it shy running. It’s like they’re shy and they’re Apparently, and this lets them go faster. And apparently this wasn’t trained. They sort of, the embodied AI discovered this posture on its own to let it go faster. Anyway, pretty, turns out robots run real funny. And when they go around the corners, they also look pretty funny. So a lot of interesting stuff. That’s sort of takeaway number one.
08:35
Number two, this thing is going to grow dramatically, this competition. mean, year two was definitely breakout. Year three, I think, is going to be off the charts. This was five days this time. It was 51 different disciplines. Up from 26 last year. So they basically doubled the number of types of activities. Overall, the events, disciplines would be a category like Kung Fu, tennis, dancing.
09:05
Within those 51 disciplines, had 1,300 different events. They’ve all basically doubled from last year, all of them autonomous. Now, some of the events, there’s an interesting breakdown. A lot of the events are sort of individual scenarios like how high can you jump? How well can you do street dancing, which is one. And then others are based on competition, soccer.
09:33
There’s three on three players and there’s five on five. Table tennis, kickboxing. So there’s sort of competition based and then there’s just individual scenarios. The participants also jumped. had 666 different teams brought robots. The total number of robots, 2,056 from 16 countries, like mostly China at this point. A lot of universities, a lot of students.
10:02
Pretty fantastic. So yeah, it jumped from last year to this year. I think next year is going to be crazy. Now, one of the complaints you’ll hear online is, well, who cares about this stuff? Who cares about robots dancing? Who cares about them doing backflips? OK, this is takeaway number three. This is all about industry and real world robot performance and standard setting.
10:31
Now they’re starting with the idea we’re going to replicate human activities with humanoid robots. And okay, they have sports, soccer, but they also have arts. And they have practical tools like cutting things. They are clearly doing scenario-based competitions that are directly applicable to real-world settings. Factory work.
10:58
working in hotels, working in homes, doing logistics. So even though they say, hey, we have 1,300 different events, they’re very directly leading to real world applications. So how do you get better at real world applications? I mean, you could put people into homes and see how they do and maybe robot one is better than robot B. But the best way to do it is to have a national
11:25
international competition and let them compete in that specific cell skill so that we know robot one is better than robot two at whatever. So the scenario based ones are kind of individual skills and performance. The competition based ones and really what they’re doing is they’re getting synchronized movements across lots of robots. That’s a different thing. The soccer teams, three member teams, five member teams.
11:55
They have to synchronize together. That’s interesting. that is also a real world situation, working together in a factory or something like this. So all of this is directly tied. And I heard a talk by the one of the guys who works at the competition, the group that runs the competition. They are clearly focused on, it’s not about who can do a better backflip. It’s all about driving forward real world performance that’s directly applicable.
12:26
and they break it into sports, arts, and basically practical skills. So one, it’s to drive performance. Two, it’s to create standards. So we all know we are competing in this area with these basic metrics. And then everyone competes against those skills. That’s very different than, let’s say, AI competing in leaderboards, where you’re trying to do some benchmark on an IQ test online. No, this is
12:53
Practical real-world performance and we can see the standards It also helps all the developers the robot manufacturers kind of know You know, this is what I’m going to be judged on so they can all sort of build according to these standards You basically want industry standardization, but really at the industry vertical level. So that’s kind of what they’re doing They’re driving industry performance and standardization That’s really
13:23
Pretty impressive. Okay, last one takeaway, number four.
13:29
I don’t think the US has any real chance of winning in robots globally. I think that’s just gone. That is a possibility is not realistic. Their best strategy should be to focus on being number two. How can we be a fast follower to China, Chinese businesses, because they’re just way out front.
13:55
I looked it up, there are 150 to 200 humanoid robot companies in China right now. There’s probably 300 robot dog companies in China right now. It is just, and when I went to the big conference, it was five different halls just packed with robot companies, 90 % of which I’d never heard of before. And it’s not just the robot, it’s the supply chain, it’s the actuators, it’s all of that. This is
14:23
They’re too far ahead. I mean, they’re already deploying at scale. Companies like Unitree, they’re putting out thousands of robots right now. So is Agibot. What U.S. company has moved beyond the pilot stage? Has anyone doing mass production? Optimus at Tesla, they aren’t. Who else is selling these things and deploying them in the thousands? And soon to be tens of thousands. So one, they’re way out front.
14:53
Two, they have certain advantages that I don’t think anybody can beat. I’ll give you my short list.
15:02
We could call these capabilities as advantages. They have a manufacturing base, which is the supply chain, the Chinese manufacturing ecosystem that has been built over 30 years. That one, you can’t compete with this. Nobody else has it. We’ve seen this play out in EVs. Chinese EVs just went everywhere. Nobody can match their price points. And not only do you have the production base,
15:32
What’s the best use case for robots? Well, it’s in manufacturing. So they have the largest production base and the largest proven demand for these things. That’s point number one. Number two, the part I didn’t appreciate, I’m going to do a talk on this.
15:49
They are combining robots with infrastructure. Two things China does incredibly well. One is manufacturing. People talk about that. The other is they are very good at deploying infrastructure at scale, building bridges, building roads, building airports. On average, China builds two Chicago’s every single year. Well, all of that capability, that activity, they’re putting robots into it. I’m going to do a whole talk on this because
16:19
I was stunned to see how the state-owned enterprises are using robots in shipbuilding, in train building, not just building, but operations, maintenance, in construction, in state grid, in the telco business. It is huge. That was literally my biggest takeaway from the big conference was that the state-owned enterprises are all in on robotics.
16:46
And what they’re building just blew my mind. hadn’t even thought about it very much. It dwarfs what we see coming out of these robot dogs and whatever. I’ll show you a ton of this in an article. So yeah, they’ve got massive infrastructure building and maintenance and operating capabilities that’s all going to be robot native. Mining, min metals, bow steel, tunneling, robots in all of it. Okay, number three.
17:15
which I’ll talk about in the network part, the intelligence of a robot, especially a group of robots, is directly related to your network. And the basic argument which I’ll talk about is, for robots to get smarter and smarter, you need more intelligence, which means more GPUs. You either put those GPUs in the cloud or you put them in the robot on device. If you put them in the robot, they
17:44
soak up a huge amount of power, which means you need bigger batteries in the robot. That’s a problem. The bottleneck for robot deployment and really intelligence, how many robots and how smart they are. The two bottlenecks are batteries in the robot and GPU on robot. So you have to move the intelligence into the cloud, which means you need a network that’s robot native. Okay.
18:14
Who’s really good at building telecommunications networks? China. They pioneered 5G mostly. They are absolutely leading in 5G advanced. And 6G is already moving. They’re already deploying this in China. You get 5GA, which is 5G is one gigabyte down, 500 gigabyte up. OK, 5GA is about double that, pretty much.
18:44
OK, you can go all over China, turn on your phone, and you’ll see at the top of your phone 5GA. It’s not everywhere, but it was deployed at scale last year. So they’ve got the network that will let them deploy at scale in a way that few other countries can. Maybe Korea. Thailand has a very good mobile network.
19:08
That’s it. Then they have the SOEs, the state-owned enterprises. One of the reasons China can build bridges and roads and rail and trains and satellite networks, all of this infrastructure so effectively is because they have a whole network of state-owned enterprises that all work together. They go into a local province. The state-owned bank basically issues loans that go to China construction and
19:36
You know, maybe the steel comes from BOW Steel and they all work together to deploy infrastructure at scale very effectively. Well, if robots are being plugged into all of that, yeah, you’re going to have a massive demand for robots coming from the SOEs who are going to be very well funded and who are not going to be buying robots with a purely commercial mindset. They’re going to view them as a strategic priority and if we lose money on them, that’s okay.
20:06
That’s how they can build so many airports. They don’t need to build them on a purely commercial basis. They can build whole districts of cities, lose money for 10 years, and it’s okay because the bottom line is not a PLL, it’s an overall strategic priority for the country or longer term development plan. We call that state directed development, which I used to give lectures on this at Peking University all the time.
20:31
Anyways, you put all those things together and you tell me how any other country beats China in robots. I don’t think it can be done. They’re going to be number one globally. Number one in China, number one globally. So the US’s best strategy is to be number one in the US, which means protectionism, and then try to be number two outside of the US, which means picking your spots like Tesla with the optimist. But you’re not going to win across the board. You’re going to have to sort of
21:01
be a specialist and a fast follower. And then maybe you can catch up over time as you build up your manufacturing base and some other things. But today, no way. I walked away totally convinced that this is going to make the sort of massive deployment of EVs around the world over the last five years out of China. This is going to dwarf that because when they deployed EVs, there were already existing car companies they had to compete with.
21:29
There’s no existing competition for this, the robot deployment. It’s open field. And that was kind of takeaway. With that, let me switch over to sort of the China Unicom Huawei topic. But yeah, it was kind of stunning. I highly encourage you to do this. If you get a chance, well, it’s going to be here now. But go to the World Conference for Robots and then stop by the World Robot Competition. It’s pretty amazing.
21:57
Okay, let me get on to the main topic. Okay, so I went up to the north of Beijing, basically where they have the bird’s nest and all the buildings they built for the opening of the Olympics back in 2008, the Beijing Olympics. I was actually at the opening ceremony where they had the runner and the torch and the president of China was there, president of Russia was there. It was pretty cool. I ended up being in the audience, which was kind of like amazing in retrospect. They built a bunch of these coliseums up there.
22:27
the National Speed Skating Oval, which is referred to as the ice ribbon. That’s where they had the robot competition. think that’s going to stay, as far as I can tell, that’s going to end up being the home for this because they had to build some pretty impressive tech on the mobile network side and some other things to make this work. I mean, you need the mobile network for connectivity. You also need tons of battery charging. Like the robots all have to be charged.
22:56
There’s really interesting requirements to building an event for 2,000 competing robots. the tech of this place is actually pretty interesting to think about. But what this ended up being, which I didn’t appreciate until I got up there, was this ended up being a really cool real world test and validation of building a robot native network.
23:23
which was basically built by China Unicom and Huawei. China Unicom, which I’ve never really talked about China Unicom on this podcast, really pretty cool company, actually. The mobile networks of China, China Unicom, China Mobile, both state-owned entities, which people that often confuses people because when you hear state-owned entities, like if you’re from the US, you think the post office. Well, no, the post office is kind of a monopoly.
23:52
And it’s a bureaucratic monstrosity with really terrible service that’s been around for 130 years. People don’t equate high performance with state-owned entities, especially when they’re a monopoly or a duopoly. Okay, China, that’s not how things work in China, which is pretty surprising. If there is a state-owned entity, like in this case, a duopoly, China Mobile, Unicom, there’s also China Telecom, but I won’t talk about that.
24:21
They’ve been given licenses to do this. They are highly competitive against each other and the operating performance is really impressive. And I’ve never been quite sure how China has achieved this. They’ve done what very few countries have done. I can’t think of any off the top of my head really, where they have state owned entities that are also highly competitive with each other and high performing. Usually you don’t see that.
24:47
In this case, another thing to think about is when you hear the term state-owned entity, that just kind of means who owns it. That’s not necessarily what you care about. What you care about is does the enterprise operate with state, with sort of towards state goals, state strategic goals, or do they operate with more of a commercial mindset? I need to maximize profits and operate like a business.
25:17
State-owned entities can sometimes operate with sort of state directives and objectives. Like we’re going to build airports across the country over 20 years and we don’t necessarily care if they make money in the short term because we know the benefit to the whole country in the long term. Okay, that’s more of a top-down strategic objective from the state. Other times they can operate pretty much like regular businesses and over the years they can sort of go back and forth between those two hats.
25:48
I’ll give you an example. Like, back around 2000, the major state-owned banks had pretty good NPLs, non-performing loans ratios, on their books. Because for 10 to 15 to 20 years, they had sort of operated with a strategic mindset, where they were making a lot of loans without necessarily caring if they got paid back, because they were financing infrastructure build-out and other things. So when they got ready to go public, their non-performing loan ratio was quite high.
26:17
And they had to clean up their books before they could do that. They moved the NPLs off onto separate vehicles and then sort of sat them there forever and they outgrew them and they disappeared. But as they went public, they kind of switched to a more commercial mindset for, you know, eight to 10 years. And then they switched back over time after the financial crisis 2008. They switched back from a commercial mindset. Let’s only make loans if they’re commercially viable.
26:45
they switch back to sort of a strategic mindset where we’re going to support the country because that’s the state objective. So these companies can go back in between those two things. They can go back and forth depending on what they’re talking about. And it’s not just state-owned enterprises that operate that way. You can see private businesses operate the same way, especially if they’re in tech. Maybe they’re making aircraft. Well, that’s mostly SOA. Private businesses can operate the same way.
27:13
So the SOE thing doesn’t necessarily mean what people think it means. Anyways, China Unicom, very cool company. They’re making robots. They have smart home devices. They have 5GA everywhere. Well, not everywhere. They have 5G everywhere in China. And they have 5GA. I haven’t checked recently what it is. But if you’re in China, you go into any major city, you turn on your phone, you’re going to see a 5GA on your phone screen, which basically means…
27:41
Instead of 5G is typically 1 gigabyte download, depending what you’re talking about, 500 megabyte upload or 100 megabyte upload. Once you go to 5GA, you basically times 10. And the one people care about is the upload. So 5GA, also called 5G Advanced, you’re talking about 1 gigabit per second upload speed, which is amazing. And it turns out.
28:09
That’s what you need for robots, which I’ll talk about. Anyways, you can get 5G everywhere. That’s China Unicom building that and operating it. Where does the equipment come? It comes from Huawei and others, but mostly Huawei. So what are the problems? Why was this robot competition cool? Because when you start to have robots, 2,000 of them competing in real time, you need a couple…
28:37
There’s a couple problems and I’m summarizing a presentation by China Unicom. They basically said, look, first of all, you got a scale problem. You got 2000 robots and you got 6 to 12,000 humans in an arena, all using the same mobile network. And the humans are actually doing a lot of uploading a video because they like to take videos and upload it. But the robots more or less need a dedicated network. How do they do that? Well, talk about how they solved it.
29:06
The other thing you need is you need ultra low latency, under 30 milliseconds end to end for a robot. It senses something, it sees something with its eyes, it goes up to the cloud, it processes when it’s through the higher cognitive function, it comes down and gives a motor command, know, lift your leg. So that’s got to be 30 milliseconds. So we’ll call that ultra low latency. You have massive bandwidth.
29:35
especially in terms of uplink. And I think that’s actually the key thing that matters. The two things that I think matter are, well, high speed uplink with big capacity. You need the one gigabyte per second going up, which means you need a 5GA network. And then you need reliability as well. You can’t have these robots going down the street or competing a competition if it’s glitchy. So you need reliability. You need the massive
30:04
fast capacity uplink. Second to that, you need fairly low latency. In this case, it’s a competition, so you need under 30 milliseconds. Outside of the competition on the streets, you probably don’t need it that low. Okay, so those are your problems, and here’s how they solved it. I’ll give you the solution. Basically, their solution was something that Huawei and China Unicom have been working on for about six years.
30:33
In these two companies founded something called 5G Capital, which was basically the idea of building the world’s greatest mobile network in Beijing. And the idea is we’re going to put 10,000 base stations across Beijing, which are capable of, in this case, one gigabyte per second uplink. OK.
31:01
That’s what they started out to build. They call it the 5G, a giga uplink network. And I’ll put the slide in the notes of what 5G Capital has been building over the last five years. This is not the first thing they’ve built. is like they’ve been rolling stuff out every single year. This is why I argue that, I think Beijing is going to be the robot capital of the world very quickly, because they’re the only ones who have the robots and they’ve got the mobile network to support it.
31:30
And the competition was really a real world use case, test case for this. You could consider it a stress test for doing this. So what is their giga uplink? They also call it the 5GA large uplink network. What does it have? Well, it has a peak uplink speed of one gigabyte per second. This is in the competition. They have end-to-end latency for humanoid robots within 30 milliseconds.
32:00
They actually have an integration with the Bay Do positioning system, which is pretty interesting. They’ve done this by deploying two separate bands. They have a dedicated 100 megahertz band, which is what the robots are using. And then they have a 300 megahertz band, which is what the humans are using in the stadium.
32:27
The guy that really got my attention was the president of Huawei’s wireless product line, David Lee. He had a quote which was, quote, the second world humanoid robot games have provided a compelling real world validation of the 5GA 100 megahertz giga uplink network, blah, blah. Basically four embodied AI scenarios. This is a first real world validation stress test.
32:57
for a robot native network. That’s how I see it. They used the 3.5 GHz range and for the robots it was 100 MHz band. Okay. Now, within the competition themselves, they said they were averaging about 397 Mbps uplink the first night of the competition. So they weren’t hitting peak 1 Gbps, but they were pretty good and they could handle
33:27
1 gigabit per second. OK, so that’s kind what they built. They had a couple other sort of technological solutions, which I won’t go through here too much, but I’ll put the JPEGs in the notes if you’re curious. They basically did carrier isolation, which is they separated the human and the robot network. They also did cell merging, so they built the network into one integrated network, all the towers act as one. They did network slicing, which I mentioned.
33:57
break the usage into various activities, then they sort of queue and route different activities differently. they give priorities for the robot scene competition, obviously. Communication and other things would be different. So OK, they did some network slicing, and then they had some other stuff. Those are kind of the main things they did. But basically, they built a network designed for robots embodied AI.
34:25
in large numbers doing very advanced high complexity tasks. And that’s really what you need to build across Beijing to turn this into a robot city. You need a robot native network, more or less. Okay, they’ve got one. All right, last point. Let’s kind of dial up from there from the competition to, okay, based on this, what does a robot native network…
34:50
Need to be or likely will be if we’re going from hundreds of thousands of robots in the world to 70 million sold every year. What does that mean what kind of network do you need this kind of before the same thing when I said look what kind of network do you need to build for a world with 800 active 800 billion active agents not just a billion humans well same question if we have a hundred million robots in this world what kind of network do they need.
35:16
And it’s probably we can talk about it at the city level. So we’ll start with Beijing. Here’s my takeaways, which is really just a couple of things. OK. You got to separate into two different bands, 100 megahertz band. They did it at 3.5 gigahertz. That seems to be the starting point for a robot native network. And that network basically prioritizes for uplink.
35:44
Humans can use a 300 megahertz band that is more used for downlink and watching videos and stuff like that. Okay, that’s pretty much what they did. The mobile network apparently is the bottleneck for how smart these robots can be and how many of them you can have. So they’re the bottleneck for embodied intelligence. Turns out to be in many cases network.
36:13
because you’re going to run out of capability on the robot itself because of on-robot commute and the amount of batteries you can put in a robot because these GPUs take a lot of energy. And the more batteries you jam into a robot, the heavier it becomes. There’s not a lot of room for batteries in robots. So the mobile network becomes more and more important the more you move up the scale in terms of complexity.
36:40
intelligence itself and number of robots, which you could call collective intelligence. Okay. So we’re shifting the compute to the cloud overall, and we’re keeping motor control down on robot, or they call it on terminal, because you can have embodied intelligence and things other than robots. They can be devices you wear on your head. They can be smart glasses. They can be EarPods. So anything terminal based.
37:09
You could call it network versus terminal and robots, starting with humanoid robots, kind of, you know, they’re going to use a lot of that. Okay, so the network really matters. Then David Lee, the Huawei guy, he had some interesting comments, or maybe it was the Unicom guy, I don’t remember who. How much, you’re getting all this data uploading happening, right? The robot sees things, senses things, not just with its eyes.
37:36
and its ears and its sensors in its hands. Robots can smell. You can give them sensors there. They can hear. What they can see with their eyes can go everything from normal vision to night vision to thermal to UV. There’s tremendous sensing that can go on. And that all goes, uploads in basically data streams, data sets. Well, how sophisticated do you want that? And the example mentioned was,
38:05
Okay, if a robot has one hand, we have one data stream from the hand. Typically, if you look at a hand, the basic hands will just have sensors in the joints for pressure, torque, things like that. They’ll have pressure pads in the fingers, things like that, so you can hold something without dropping it. The more sophisticated ones will have much more, the dexterous hands. They can have cameras in the wrist joint so they can see what they’re holding as well.
38:31
They can get very sophisticated in the sensors in the hand. So the amount of data can go way up and that’s just one hand. What if the robots got two hands? Well, we’ve just doubled the data set going up. What if we put cameras in every single finger? Well, now we’ve got 20 data sets going up. So the amount of data that can be uploaded can just for perception and understanding can change dramatically.
39:01
Then that’s all got to go into the GPUs in the cloud. Tremendous amount of compute, depending on what you’re doing. You got to download the motor controls and other things into the robot. So yeah, you can see how the more sophisticated you get, the amount of uplink is going to explode in terms of just speed, capacity, all of it. So the point that was made was the efficiency. And you can’t just keep
39:29
increasing and increasing. You’re going to max out very soon. So he said the efficiency between the terminals and the network really matters. How the network works with these terminals, in this case a robot or a robot hand, how that links together tightly is a really important question. And how you can do that efficiently is a really important question because the amount of uploaded data explodes quite quickly. Network stability.
39:58
also matters. You don’t need super fast latency for most stuff, but you can’t have these robots be glitchy. Not if they’re doing things in the kitchen or doing activities or running down the street, especially when there are drones in the air. have to be there. You probably need ultra fast, ultra low latency.
40:22
So that’s kind of important. it’s easy to see that when you try to, mean, given what I’ve just said, try to predict network demand. How much demand is there going to be? How much does this mobile robot native network, how much is it going to have to supply to meet future demand? And it becomes almost impossible to predict that. You could see the demand surge.
40:52
like go beyond what anyone expected. So this sort of terminal capabilities is going to be a really difficult question. It’s not just the numbers of robots, and it’s not just how intelligent they are. It’s that times that times how much data do you want to upload? Do you want a camera in every finger?
41:12
It could go through the roof. How much of the robot’s going to do? Yeah. So the demand question, I don’t know how you’d put a number around that today. But that’s going to be a real big challenge for the mobile networks. OK. Anyways, that’s it for this topic. You can tell I’m kind of blabbing on about this because I’m still trying to get my brain about what this would look like.
41:35
But I like the idea of thinking about agent native networks versus human native networks versus robot native networks. You could call it embodied AI native networks. It’s harder to say that. But at least if you say embodied AI, that gets you more into the idea of other types of terminal. So yeah, I’m still trying to get my brain around it. yeah, pretty fantastic. I’m pretty amazed. OK, that’s it for the, I guess, for the topic. Oh, full disclosure. I’m trying to be more diligent about this. I have had a consulting arrangement with
42:05
Huawei in the past 12 months. So if I’m going to talk about them, I should be more explicit about that. Now, this is just me showing up, talking about what I saw. So this is all my content, my opinion, all that. But yeah, they were a client in the past year. So anyways, full disclosure on that. What else? That’s kind of what I wanted to talk about. Any fun stuff? I haven’t really been just been running around. I’m going back to a
42:34
China in a couple days. This is just a quick trip home. Gonna go into Hangzhou, spend some time at Alibaba. I’m going to visit a couple companies there as well, which I’ll talk about. Really, I’m actually really excited about those. I’ll talk about who I’m going to visit in Hangzhou. That should be pretty fantastic. No TV stuff I can recommend, really. Oh, you know what I watched? I watched the Michael Jackson movie on the plane. That was really kind of nice. I didn’t think it was going to be big thing. I’m not a huge Michael Jackson fan or anything.
43:04
But yeah, do kind of, there’s a funny thing about these movies about musicians and stuff. Sometimes you watch them and you become more of a fan or you think more highly of them. That was the case here. Like I hadn’t thought about Michael Jackson. I was never a huge fan. I absolutely knew who he was and he was good. But then you watch, I watched the movie and I’m like, okay, he really was pretty fantastic as an entertainer. You know, my opinion went up. The same thing happened when they did the Queen movie.
43:32
Bohemian Rhapsody or whatever it’s called. I thought, okay, that was pretty neat. Every now and then I watch a biopic like this and my opinion of the person drops. Like I think less of them than beforehand. That happened with the Elton John one, whatever that movie he did. Like I left, I watched that movie and I left liking him less than I did. Eric Clapton too, I read a book about Eric Clapton.
43:59
I always thought he was a pretty cool guitar guy and I read the biography and I’m like, dude, I don’t like this guy. I liked him a lot more before I knew anything about him. There should be a word for that where you like someone less the more you know about them and you’re still a fan but not like you were before. I don’t know what that’s called anyways. Okay, that’s it for me. Have a great week. I’ll talk to you next week. Bye-bye.
——-
I am a consultant & keynote speaker on how to increase digital growth and strengthen digital AI moats.
I am the founder of TechMoat Consulting, a consulting firm specialized in increasing digital growth and strengthening digital AI moats. Get in contact here.
I write (a lot) about digital growth and digital AI strategy (3 best selling books, +2.9M followers on LinkedIn). There is a free book and email newsletter below.
My Moats and Marathons book series is a framework for building and measuring competitive advantages in digital businesses.
Note: This content (articles, podcasts, website info) is not investment advice. The information and opinions from me and any guests may be incorrect. The numbers and information may be wrong. The views expressed may no longer be relevant or accurate. Investing is risky. Do your own research.
