Humanoid Robots Are on the Horizon—But When?

This is The Stepback, a weekly newsletter diving into a key story from the tech landscape. For insights on robotics and AI, keep an eye on the latest updates. The Stepback lands in subscribers’ inboxes at 8 AM ET.

We’re taking a winter break! The Stepback will return on January 11th, 2026. In the meantime, feel free to catch up on previous issues.

How it started
I have a soft spot for robot fail videos. I can watch them on repeat, laughing to myself as a form of therapy. Maybe I’m a bit of a sadist, or maybe I just need to get out more — you can judge me later — but they always get me. Recently, I’ve been captivated by a clip of Tesla’s Optimus robot taking a tumble at the company’s Autonomy Visualized event in Miami, which has been making the rounds on social media.

In the footage, Elon Musk’s much-hyped humanoid robot was handing out water bottles from behind a table when it knocked several over, flailed its arms, and collapsed backward like a marionette with its strings cut. If you watch closely, you’ll catch two amusing moments: a small spray of water as its arm crushes a bottle mid-fall (which made me chuckle) and a motion that looks eerily like someone removing a VR headset.

This isn’t the first time Tesla has exaggerated the capabilities of its autonomous robot, which Musk claims is vital to the company’s future. An early demo featured a dancer in a tight bodysuit to showcase what the Tesla Bot, now known as Optimus, could potentially do. Later demonstrations were revealed to be humans in disguise, remotely controlling the robots with something akin to a VR headset, a method Tesla is known to use in development.

Humans have been fascinated by robots for centuries, from ancient myths of stone golems to modern sci-fi, driverless cars, and Roombas. The idea of breathing life into the inanimate has always intrigued us. Much of the current excitement around humanoids can be traced back to Musk, so it’s reasonable to approach his promises of revolutionizing the world with a healthy dose of skepticism. Musk, who has pledged to create a “robot army” of 1 million humanoids, has a reputation for making outlandish predictions, and the field of robotics has seen its fair share of hype cycles. Historically, technology has lagged behind the enthusiasm of those eager to bring AI into reality, but today, we’re being told that the tech is finally ready to deliver.

So, what does “ready to deliver” actually mean in 2025?

How it’s going
Right now, there’s a gold rush for humanoid robots. Every major tech player has them on their agenda, with companies like Nvidia, Meta, SoftBank, Google, Amazon, Microsoft, Intel, and Tesla all investing heavily in this next big tech frontier. They’re not alone; a growing number of challengers, including Apptronik, Boston Dynamics, Figure AI, and 1X, are eager to join the fray.

China is also in the mix. The government has identified embodied AI — which includes non-humanoid robots like drones and quadrupeds — as crucial for future economic growth. In typical fashion, it’s positioning itself as a global leader in robotics through massive investments, government initiatives, and state subsidies. Tech giants like Ant Group and Baidu, along with startups like Unitree and AgiBot, are all getting involved.

If you look at the demos coming out of China and elsewhere, you might think the humanoid robot future is already here. This summer, robots competed in dance, combat, and track events at the inaugural World Humanoid Robot Games in China. A similar event took place in Greece — the International Humanoid Olympiad — at the birthplace of the Olympics. Humanoid fights have gained unexpected popularity, occurring everywhere from organized contests to underground fight clubs.

Companies are also eager to transition bots from factories to homes. Advocates argue that humanoids are worth pursuing over other robot types because they can operate in human-centric environments. Figure recently showcased its Figure 03 robot, which can handle household chores, releasing a video of it doing dishes and folding laundry. 1X introduced Neo, claiming it’s the world’s first consumer-ready humanoid robot. There are even videos of Neo shakily completing basic tasks. While I’m not sure I can afford it, 1X is selling units for $20,000, with deliveries starting in the US next year.

However, actual use cases remain limited, and impressive demos don’t equate to functional products. Despite notable advancements, many demos are staged, scripted, or even teleoperated. For instance, while it’s technically true that Ant Group’s R1 was cooking at a trade show, it moved at a pace that would make Meryl Streep’s Miranda Priestly proud. Similarly, the idea of owning 1X’s Neo becomes less appealing when you realize it requires random humans to telecommute into your home to control it. And the excitement around robot sports stems from the contestants being wobbly and unpredictable, not from their athletic prowess.

This raises an important question: If the tech doesn’t live up to the hype, why is there so much excitement around humanoids? For the first time in a while, it’s not entirely unreasonable to think the tech might be catching up.

What happens next
Historically, getting robots to perform even simple tasks that humans do instinctively, like walking or picking up a glass of water, has been a monumental challenge. Hardware — but mostly software — has been a limiting factor, confining machines to highly structured environments and narrow tasks. However, advances in AI are changing the game and accelerating robot development.

The remarkable breakthroughs in AI image and text generation from various companies are largely due to large language models (LLMs) that power chatbots. These models use complex networks to recognize and reproduce patterns from vast amounts of training data. Larger AI models are trained on nearly everything available online, enabling them to act in a generalized manner rather than following rigid, hand-coded rules. For robots to function in the real world, they need this kind of generalization, and roboticists are working to give machines the flexible understanding of the physical world necessary for survival.

The challenge lies in data. The material used to train LLMs is readily available online, but the data needed for training robots is much harder to come by. Robots require detailed examples of real-world movement, which aren’t easily accessible at scale. Companies are striving to change this, with significant efforts underway to create the data necessary for training robot models. On the surface, it may seem absurd — Tesla has workers donning cameras and sensors to teach Optimus to behave like a human — but it’s proving effective. This is also why companies like 1X are deploying less-than-autonomous bots into homes; it allows them to gather the data needed to develop autonomous bots by remotely conducting complex tasks like loading dishwashers or cleaning tricky objects.

Humanoids are also becoming more affordable, especially in China, as hardware costs decrease and economies of scale kick in. Prices vary widely — entry-level units can be as low as $1,400 for the Chinese model Bumi, while models from Unitree or 1X range from $13,500 to $20,000, with industrial models costing more than houses. As prices approach levels that some consumers can afford, more bots are being deployed in the real world. This creates a feedback loop: as companies gather more data, they can build better models and robots that more people want, assuming they’re willing to welcome them into their homes.

Yet, despite all this progress, it may still be hype. In November, China’s leading economic planning agency cautioned that a humanoid robot bubble could be forming, highlighting the disparity between the number of companies and the scale of investment versus the lack of viable use cases. The reality is, they’re not truly autonomous yet, so aside from hobbyists and researchers, who in their right mind would buy one? If I wanted someone to clean my home, I could hire a cleaner without the hefty upfront cost and save my money until a robot capable of actually performing the task becomes available.

Until companies stop relying on flashy promo videos and teleoperating supposedly autonomous robots, it will be challenging to gauge our current position. Perhaps the robots are finally on their way, or maybe I’ll just have a steady stream of fail videos to enjoy. Time will tell, but I’ll be ready with the popcorn.

By the way
* A fascinating data creation and labeling industry is emerging from the robotics boom. Around the globe, people are being paid to generate the kind of data needed to train robot models.
* While companies are racing to gather as much real-world data as possible, it doesn’t necessarily have to be real. Google DeepMind claims its AI world model can generate 3D environments to help train robots.
* If you enjoy robot fails like I do, check out this Russian humanoid faceplanting during its big stage debut last month.

Read this
* An alum explores the humanoid hype machine, revealing that kicking robots is a common practice among roboticists.
* A report details the team of workers striving to teach Optimus how to act more human, a role that sounds demanding and a bit absurd.
* It’s not a humanoid, but there’s an interesting story about how robots are being used to pack groceries in one of Ocado’s warehouses.
* An explanation of why humanoid robots need their own safety regulations.
* A case for looking beyond the human form in robotics.