Humanoid Robots Are Getting Real Jobs
13:20 Share

I can say to my new Samsung Galaxy S25 Ultra, hey, find a keto-friendly restaurant nearby and text it to Beth and Steve. And it does without me lifting a finger. So I can get in more squats anywhere I can. One, two, three. Will that be cash or credit? Credit. Galaxy S25 Ultra, the AI companion that does the heavy lifting so you can do you. Get yours at Samsung.com. Compatible with select apps requires Google Gemini account results may vary based on input check responses for accuracy.

Welcome to Tech News Briefing. It's Tuesday, March 4th. I'm Charlotte Gartenberg for The Wall Street Journal. Could your gaming PC help train artificial intelligence models? A handful of startups are looking to stitch together virtual networks of graphics processing units, or GPUs, to compete with massive AI data centers.

We'll hear how underused GPUs, the advanced computer chips that power AI, might open the door to new AI players. Then, Star Wars' C-3PO and the Jetsons' Rosie the Robot are still the stuff of science fiction. But humanoid robots might soon be helping us do more everyday tasks. WSJ tech columnist Christopher Mims tells us what the latest humanoid robots are doing and the tech that could bring them into our homes.

But first, tech CEOs like Elon Musk, Mark Zuckerberg, and Sam Altman think that dominating an AI will come from amassing as many GPUs as possible and networking them together in massive data centers. But what if there's another way? For more on that, we're joined now by our tech reporter Deepa Sitharaman.

Deepa, you spoke with some startup founders who believe that underused graphics processing units could bring AI success to companies that are not quite as well resourced as the open AIs and metas of the world. Who are some of these people and what are they proposing?

Right now, they are tiny companies that just started last year, but they have this broader, bigger vision of finding a way to fight against the man is sort of how they view the big tech companies. In their view, there's a lot of resource hoarding, the big open AIs of the world, the

XAIs of the world, like the Elon Musks, Sam Altmans, Mark Zuckerbergs, they're all pulling together and buying up what are already these crazy expensive GPUs, ensuring that they're the ones that can train gigantic models or machines.

make models work. And here you have these companies that say it doesn't need to be like that. Having said that, it's a huge risk. There is absolutely zero guarantee that something like this will actually function in the real world. What devices use GPUs?

that might theoretically be stitched together. GPUs were originally built for gaming, and there are a lot of gaming rigs out there that have GPUs where somebody isn't gaming

24 hours a day and so it lies idle at least some of the time. Additionally, GPUs were used for things like crypto mining. You can also just utilize those things. There's not like 100% utilization necessarily on these types of GPUs.

Plus, there's a lot of companies that for whatever reason they might be dealing with sensitive data, they're deciding to have their GPUs in-house so that if they need to use these big AI systems, they can rely on their own GPUs and not go elsewhere, just from a data protection standpoint.

Those guys aren't using GPUs 24-7 necessarily either. And that GPU is networked and hooked up and online and gobbling up power. That's a revenue opportunity missed. What risks does renting out your GPUs expose your company to or maybe a particular individual? It isn't yet clear what the broader risks

would look like. Like how do you ensure privacy and security in this kind of distributed system? The people that are trying to build these models and trying to operate within these data centers, a lot of them are risk averse already given that they're already taking this big risk by paying for all these GPUs and trying to do these models. The risks of distribution are maybe something that a lot of entrepreneurs may not even want to take on.

So there's the privacy issue. How do you protect data? Those are things that people are going to have to work out. You talk to the entrepreneurs, they think all of this can be sorted out in time. Lowering the barrier to entry to building AI tools seems fairly meaningful. And that is something that really could change the way we think about AI systems currently and the way that we engage with the technology overall. That was our tech reporter, Deepa Sitaraman.

Coming up, where's my robot made? Tech has not yet made Rosie the Robot, but humanoid robots are finally getting real jobs. We'll find out where after the break.

Your data is like gold to hackers. They're selling your passwords, bank details, and private messages. McAfee helps stop them. SecureVPN keeps your online activity private. AI-powered text scam detector spots phishing attempts instantly. And with award-winning antivirus, you get top-tier hacker protection. Plus, you'll get up to $2 million in identity theft coverage, all for just $39.99 for your first year. Visit McAfee.com. Cancel any time. Terms apply.

Newly powered by AI brains, humanoid robots are becoming increasingly capable. Currently, they're mostly used in what's known as materials handling, basically moving things around in warehouses and manufacturing facilities. And while these are generally low-stakes tasks, humanoid robots could save money for companies that use them. Here to tell us about the tech powering these bots and what's next for the industry is our tech columnist Christopher Mims.

Can you talk to me about some of the startups and companies and some specific humanoid robots that you've seen lately? Okay, this is Christopher Mims' top humanoid robot companies in order of likelihood that you'll actually hear about them again five years from now and they won't just be gone. All right, number one, Aptronic. Aptronic is an Austin, Texas-based company. Their focus is on making humanoid robot bodies as affordable as possible. So that's good in terms of making them viable. Agility Robotics.

They are the first in the world to really be used in production in a Spanx warehouse outside of Atlanta, Georgia. They're moving shapewear. Reflex Robotics. They're doing cool stuff with a logistics company called GXO. Picture a robot picking stuff out of bins and putting them in boxes to be shipped to you. Can't count out Tesla's Optimus robots. There's Figure. They just got a boatload of money and they claim to be building kind of start to finish an AI powered robot. Then

There's Boston Dynamics. They've been at it forever. They make those terrifying big dog robots that you've seen videos of.

on the internet. And now they make humanoid robots. And they don't have any tests with customers that they're willing to tell me about yet. But they're good at what they do. And they've had more experience at it than anybody. You talked about some of the job tasks that they're taking on a lot of logistics stuff and warehouses. What other tasks are you seeing those robots take on? Eventually, these robots are going to be flexible enough

that they will probably show up in the service industry. You could picture them working in the back of the smoothie shop, again, doing materials handling mostly. In the future, one of the goals is to make them part of elder care. So a lot of elder care is just helping people move around, helping people shift in and out of bed or go to the bathroom or feed themselves. These are all applications where these robots could run

really help tackle the labor crisis in that industry. What's the tech that's helping current robots do more than before? So robots, they're different than other computers in that they have bodies. So one big thing that's helping them is all the parts that make up their bodies are getting cheaper, they're getting more capable, their camera eyes, that all comes out of the

cell phone supply chain we've gotten really good at making that kind of thing their joints their limbs their actuators their motors their equivalent of our muscles people have done a lot of work in terms of making those stronger more precise lighter use less power that's really key because these robots unless you want them plugged in all the time they got to run on batteries and they apparently really chew through batteries pretty quick then there are their brains

And there have been a lot of breakthroughs lately in terms of using the same kind of underlying AI that drives things like chat GPT to learn how to animate a robot body. And pretty much all these companies are betting that that innovation is going to continue at a pretty swift pace in order to make these robots more versatile.

You spoke with Ayanna Howard, who's dean at the Ohio State University College of Engineering. She's a former robotics researcher at NASA's Jet Propulsion Lab. She's a startup founder. And here she is on our Bold Names podcast. Humanoid robots is the next shiny penny that's out there that tech folks and VCs can invest in.

Those who aren't in robotics, it seems like the next logical step around artificial intelligence and generative AI is, oh, let's think about generative AI with the physical embodiment. Christopher, how close are we to truly useful humanoid robots that say, think with AI? So we have truly useful humanoid robots today that are able to do really basic stuff, just like moving things around inside of warehouses and factories.

The challenge, honestly, is how are we going to make them affordable? The companies I talked to claim that the effective hourly wage of their robots, so they don't sell these robots, they rent them out for a wage like a human, is such that from day one, they say they're saving companies money. That doesn't mean that these startups are making money. If you could make one cheap enough, it could have...

pretty limited utility and you might still want to use it. But if it's really capable, maybe at that point the cost of the robot compares favorably with a human worker who has to take breaks and gets benefits and labor protections and everything else. How is particularly the AI going to be leveraged to get us where some of these companies are hoping to go? It's really unclear.

So the CEO of NVIDIA, Jensen Huang, has said he thinks we're soon going to have what he called a chat GPT moment for robotics. That's a bold claim because moving a robot body is 100, 1000 times more complicated than making a chat bot.

And when it comes to what makes modern AIs work, it's how much data you have available to train them. We had an internet's worth of text to train chatbots on. There does not exist an internet's worth of instructions for how to move robot limbs. So all these companies are having to come up with their own data. So they got to

test the robot in the lab. A lot of the training is done in simulation, and then they have to translate that learning from simulation to the real world. And just the real world is monstrously complicated. And frankly, it's full of fall hazards as well. So that's going to be a problem if you're walking around in your robot. That was WSJ tech columnist Christopher Mims. You can hear more about humanoid robots and AI on the latest episode of our Bold Names podcast. Find it wherever you get your podcasts.

And that's it for Tech News Briefing. Today's show was produced by Jess Jupiter with supervising producer Catherine Millsap. I'm Charlotte Gartenberg for The Wall Street Journal. We'll be back this afternoon with TNB Tech Minute. Thanks for listening.