The Problem With Humanoid Robots Was Never the Technology. It Was the Price.
How Tesla, Hyundai, and Figure AI Are Racing to Make Robots Cheaper Than Cars
Elon Musk says the Optimus robot costs Tesla roughly $10,000 to manufacture and will sell for around $20,000. That means for every 100,000 units sold, Tesla collects a billion dollars. The number sounds absurd until you realize that the company has already begun converting a factory to build them.
But here’s the thing. The production cost is not the story. The story is that right now, in June 2026, if you actually wanted to buy a functional humanoid robot, you couldn’t.
Not really. Boston Dynamics’ Atlas is estimated at around $145,000 per unit. Many other available machines are priced above $50,000 and offer functionality akin to a high-tech toy rather than a real helper. There are cheaper Chinese options, but the base models aren’t programmable for custom tasks. No reasonable consumer is going to spend that kind of money on a device that walks from one corner of a room to another.
What Tesla is doing, whether you believe in the timeline or not, is attempting to make the humanoid robot as accessible as a car. Except this time, the car cleans your house.
Why Tesla’s Price Advantage Is Structural?
When Musk gives a number, you apply a healthy skepticism multiplier. The man is not known for hitting deadlines. The price aspect presents structural reasons to give this claim genuine consideration on this occasion.
When a robotics startup wants to build a humanoid, it has to source every component from different suppliers: motors, sensors, batteries, chips, and production tooling. Each layer comes with its own margin. Tesla manufactures nearly all of these in-house. Batteries have been their core business for fifteen years. Electric motors: millions produced annually. AI inference chips will be fabricated at Terafab, the $25 billion semiconductor joint venture with SpaceX and xAI. And the factories that can build all of this at an industrial scale already exist and are already running.
Andrej Karpathy, one of the most respected names in AI, articulated this clearly when he was still at Tesla during the Optimus project launch.
All the research accumulated on autonomous driving — neural networks trained on billions of kilometers of visual data, the ability to interpret an environment in real time — transfers directly to a robot. Functionally, a humanoid can be considered an autonomous vehicle that uses two legs for locomotion instead of the four wheels typical of wheeled vehicles. That technological foundation took Tesla more than a decade to build. It cannot be reproduced in two years by a startup working from scratch.
From Model S to Optimus: The Fremont Conversion
On May 9, 2026, the last Model S and Model X ever produced at Tesla’s Fremont factory rolled off the line. Fourteen years of production for the Model S. Eleven for the Model X. Over 610,000 vehicles combined. The page is turned.
The entire production line is now being converted to manufacture Optimus robots, with a target capacity of one million units per year from that single site. A second factory is already under construction at Giga Texas, designed to produce 10 million robots per year. For context: Tesla currently produces about 1.8 million cars annually worldwide. The robot production target is five to ten times that.
On May 21, Tesla released the first images of the Optimus pilot production line at Fremont. That same week, a video surfaced showing Optimus distributing water bottles to people in what appeared to be a completely unscripted setting. We are a long way from the December 2025 demonstration in Miami, where the robot famously dropped its bottles and fell backward on stage. In six months, the improvement in object manipulation is visible and measurable. Not yet fluid, but functional.
In this industry, a functional robot at $20,000 will always beat a perfect robot at $150,000.
The Competition Is Not Sleeping
Tesla has the manufacturing story. It does not have the field to itself.
On May 19, Hyundai announced at a JPMorgan Chase investor session in Boston that it plans to deploy more than 25,000 Atlas robots across Hyundai and Kia manufacturing facilities, with a production capacity target of 30,000 robots per year by 2028. The initial deployment will begin at the Metaplant America facility in Savannah, Georgia. Atlas will not be sold to consumers. It will be integrated directly into automotive assembly lines, which is a different strategy from Tesla’s. Hyundai isn’t trying to sell robots to your household. Hyundai is trying to replace stations in its own factories. A recent Deutsche Bank analysis estimated that replacing just 10 percent of workers in a factory with robots could save $141 million per year. At 20 percent, the savings rise to $510 million. Hyundai is running exactly that calculation, with the advantage of being both the customer and the owner of the robot manufacturer.
Then there’s Figure AI, and that is a different story entirely. On May 13, the company launched a public livestream of its Figure 03 robots sorting parcels in a warehouse. Three robots, which viewers quickly named Bob, Frank, and Gary, rotated autonomously through shifts, running on the company’s Helix-02 neural system. The original plan was an eight-hour endurance test. The stream ran for 200 continuous hours. In that time, the robots processed 249,560 packages with zero human intervention, zero teleoperation, and zero hardware failures. When one robot’s battery ran low, it autonomously walked to the charging station, and another stepped in.
This Robot Just Did Something No Other Humanoid Has Ever Done — And It Worked in a BMW Factory for 11 Months
Watch this video. It’s probably the most impressive robotic demonstration ever performed.
Figure also staged a separate human-versus-robot competition over ten hours. A human intern sorted 12,924 packages at 2.79 seconds per package. Figure 03 sorted 12,732 at 2.83 seconds per package. The human won. Barely. But the human also needed lunch breaks, legally mandated rest periods under California labor law, and eventually, sleep. The robot needed none of those things. Figure AI CEO Brett Adcock posted the results and added one line:
“This is the last time a human will ever win.”
The Airport That Can’t Wait
The market isn’t waiting for perfection. It’s building with what exists right now.
In May 2026, Japan Airlines launched a two-year trial of Unitree humanoid robots at Tokyo’s Haneda Airport, in partnership with GMO AI & Robotics. The robots — compact G1 models standing about 1.3 meters tall, priced around $13,500 for the base variant — are assigned to baggage loading, cargo transport, and cabin cleaning. Japan recorded over 42 million international visitors in 2025. The country’s working-age population is projected to decline by 31 percent by 2060. Japan Airlines doesn’t have the luxury of waiting for the technology to mature. Neither does South Korea, whose fertility rate has been in freefall for years, nor a growing number of other aging economies, watch the same demographic curve.
The Hands Problem
Optimus still has real limitations, and Tesla is transparent about the most important one. Walking speed caps at 3.2 km/h, which is slow by any measure. The target is 8 km/h, a light jog, but they aren’t there yet. Navigation on complex terrain remains problematic.
But the hardest problem is the hands. According to Musk’s public statements, approximately half of the engineering effort for Optimus is dedicated solely to its hands. When you think about it for a moment, this is perfectly logical. A vast portion of what makes humans intelligent is tied to our capacity for manipulation.
Picking up an egg without crushing it.
Ensuring a shirt is folded in a tidy manner.
Activating a door handle. You perform these gestures a hundred times a day without conscious thought, and each one is a substantial engineering challenge for a machine.
The current Optimus hands feature 22 degrees of freedom and 50 actuators, up from 11 degrees of freedom in the previous generation. That’s a significant jump, but still far from the dexterity of a human hand. Some evolutionary theories suggest that human intelligence itself developed precisely because of our capacity to manipulate tools with our hands. Reproducing that in a machine is a project in its own right.
The Only Thing That Has Ever Actually Changed the World
Here is the point that very few people understand about revolutionary technology, and it’s worth sitting with because it’s probably the single most important factor in whether a technology succeeds or remains a laboratory curiosity.
It is never the technology itself that changes the world. It is the ability to produce it on a mass scale and crash the price. As long as a product stays expensive and low-volume, it remains a toy for researchers or billionaires. The moment someone figures out how to manufacture it by the million for a fraction of the initial cost, everything shifts.
Solar panels have existed since the 1950s. For decades, they were a curiosity reserved for satellites and a handful of enthusiast homes. What changed wasn’t a scientific breakthrough. It was China scaling production so aggressively that prices fell by a factor of 100 in twenty years. Overnight, solar became competitive with coal. The technology was always there. It was the manufacturing scale that unlocked it.
Musk understood this better than anyone because he did it before with SpaceX. The problem with rockets was never that we couldn’t build them. We’d been building them since the 1960s. The problem was that each launch cost hundreds of millions of dollars because the vehicle was discarded after a single flight. Musk inverted the approach: reusable hardware, integrated manufacturing, and iteration speeds that the rest of the industry considered reckless. SpaceX divided the cost of launch by ten and now holds a near-monopoly on the orbital launch market. They are now using the same industrial mindset for humanoid robotics, aiming to keep everything in-house, produce at an unprecedented scale, and make it affordable for the public by cutting costs.
Tesla’s deployment plan is straightforward. First, the robots work inside Tesla’s own factories, learning simple tasks in a controlled environment. Then, sales or leasing to large external companies like Amazon and Walmart, which have been reporting labor shortages for years. And eventually, households. Musk has mentioned a lease-before-purchase model, similar to automotive leasing. A robot in your home that cooks, mows the lawn, cares for elder family members, and handles the full range of household tasks. Essentially, everything a full-time household manager would do, except that it doesn’t require a salary and never takes a vacation.
We’re not there yet. That requires honesty. But the trajectory is clear, it’s accelerating, and the competitive field is building in real time. Boston Dynamics targets 30,000 Atlas units per year by 2028. Figure AI has raised nearly $2 billion from Microsoft, Nvidia, Intel, Amazon, and OpenAI. Tesla targets one million, then ten million, then eventually hundreds of millions of units per year. In three years, Tesla has taken Optimus from a person in a spandex suit on a stage to a pilot production line running at Fremont. Boston Dynamics needed decades to reach comparable capabilities. Figure AI demonstrated multi-day autonomous operation live on YouTube to hundreds of thousands of viewers.
The question is no longer whether humanoid robots will become mainstream. That debate is effectively over. The question is who will be ready when they arrive.
Thanks for reading. Your thoughts are always welcome in the comments.