Tesla, Figure, Boston Dynamics, or Unitree
Three years ago, a humanoid robot on a factory line seemed like a futuristic promise aimed more at investors than at real production. By the summer of 2026, the picture looks different: four companies Tesla, Figure AI, Boston Dynamics (owned by Hyundai), and China's Unitree are either already deploying humanoids on assembly lines or ramping up mass production.
In short:
each company has taken a different path. Tesla is betting on vertical integration and automotive scale. Figure is betting on automaker partnerships and real world operating experience. Boston Dynamics is betting on physical capability. Unitree is betting on price and volume. There is no clear leader yet, it depends entirely on what counts as winning.
Why this isn't a single race
It helps to separate two very different competitions:
- The hardware race - who can build physical bodies, with motors, actuators, and batteries, at scale fastest.
- The intelligence race - who can build an AI model capable of controlling an arbitrary body in an arbitrary environment.
Google DeepMind, for example, doesn't build robots at all. The company released a model called Gemini Robotics 2, which it installs on third party platforms such as Apollo 2 from the startup Apptronik. This is closer to what Tesla does with Full Self Driving for cars, just applied to robots.
Tesla, Figure, Boston Dynamics, and Unitree, on the other hand, build both the body and the brain, which is why it makes sense to compare them directly.
Tesla Optimus: betting on automotive scale
The core idea: turn robot production into the same kind of assembly line process Tesla once used to turn electric cars into a mass market product.
To make room for it, the company is even freeing up part of its Fremont factory floor, previously used for Model S and Model X assembly, for an Optimus production line.
The key upgrade in the current generation, which Tesla itself calls Gen 3 (some sources refer to it as Optimus V3):
- A hand with 22 degrees of freedom and 50 actuators housed in the forearm, modeled on human anatomy double the dexterity of Gen 2
- A new AI5 chip, sharing its computer vision architecture with Tesla's Autopilot
- Training on the Cortex 2.0 supercomputer at Giga Texas
There is some confusion around the naming. Some sources use "Gen 3" specifically for the upgraded hands mounted on the existing Gen 2 body, while a fully new body designed for mass production is called V3, with a reveal expected closer to summer 2026.
Tesla has pointed to summer 2026 for the start of low volume production, with an ambitious long term target of a million robots a year. For now, the robots are used only internally at Tesla's own factories. Limited sales to outside companies are tied to late 2026, and a mass consumer launch to 2027, with a target price around $20,000 to $30,000.
Tesla's main strength is vertical integration: its own motors, batteries, chips, and automotive scale manufacturing. Its main risk is the same one that has dogged Autopilot: the company routinely names timelines that later slip.
Figure AI: a humanoid already working on BMW's line
While Tesla is still preparing for mass production, the startup Figure AI (backed among others by OpenAI, Microsoft, Nvidia, and Jeff Bezos's fund) has already accumulated nearly a year of real world operation.
Its previous model, Figure 02, spent almost a year at BMW's plant in Spartanburg, inserting sheet metal parts into welding fixtures in the body shop, and according to BMW contributed to the production of more than 30,000 X3 vehicles.
In the summer of 2026, BMW moved to the next generation, Figure 03. What's new:
- Soft exterior components for safety when working near people
- Wireless charging
- Speech to speech communication with workers
- Reworked hands with tactile sensors and palm cameras
This time the task shifted from the body shop to logistics: the robot sorts previously unsorted parts into sequencing carts for delivery to the assembly line. All of it runs on Figure's own model, Helix, which combines perception, planning, and motor control into a single system.
The key difference between Figure's approach and Tesla's is the partnership model. Instead of building its own factories, the company embeds itself into existing automaker production lines and builds up real world operating experience rather than demo experience.
Boston Dynamics Atlas: the physical ceiling
Boston Dynamics, owned by Hyundai since 2021, unveiled a fully electric production version of Atlas at CES 2026, a fundamentally different machine from the famous hydraulic Atlas known for backflips and parkour (that version was retired in 2024 and was never sold).
Specs of the new Atlas:
- 56 degrees of freedom with full 360 degree rotation at several joints
- Stands about 1.9 meters tall, weighs about 90 kilograms
- Can instantly lift up to 50 kilograms (around 110 pounds)
- Rated IP67 - withstands dust and water jets
- Autonomously swaps one of its two batteries in about 3 minutes, with roughly 4 hours of runtime between swaps
For its intelligence layer, Boston Dynamics has partnered with Google DeepMind, integrating models from the Gemini Robotics family into Atlas.
The entire first production run of Atlas for 2026 is already committed to two customers: Hyundai's own research arm (the Robotics Metaplant Application Center) and Google DeepMind. Outside companies cannot order one yet. Actual deployment on Hyundai's assembly lines in Georgia is planned much later, in 2028, starting with logistics tasks, with more complex assembly work targeted for 2030. Hyundai has said it plans to reach a production capacity of 30,000 robots a year by 2028.
In purely physical terms, Atlas currently looks like the most technically advanced humanoid. But its real industrial deployment starts later than its competitors'.
Unitree G1: scale through affordability
While American companies compete on degrees of freedom and automaker partnerships, China's Unitree has taken a third path: maximum affordability and shipping volume.
By the company's own figures, more than 5,500 G1 robots were sold in 2025, giving Unitree, according to market analysts, a notable share of the global humanoid market. For 2026, the company has stated plans for a multiple fold increase in shipments.
The G1 is noticeably smaller than its competitors, standing about 1.3 meters tall and weighing around 35 kilograms, making it more of a compact research and demo platform than a full sized industrial robot like Atlas or Optimus.
Pricing varies sharply by configuration:
- Base version, no developer SDK - $13,500 to $16,000
- Developer oriented EDU versions (up to 43 degrees of freedom, full SDK access) - $40,000 to $70,000+ for top configurations
This layered lineup, an entry level "showroom" model and a much pricier "development" line, is often the source of confusion in coverage that calls it "the $16,000 robot."
Reported real world uses include pilot programs for baggage handling at airports, cleaning tasks, and use as a research platform at universities and companies. The G1 isn't designed for industrial assembly at the level of BMW or Hyundai. It's better described as a mass market tool for those who need a working platform today rather than maximum performance.
Robot
Company
Key strength
Currently deployed at
Status as of mid 2026
Optimus (Gen 3 / V3)
Tesla
Vertical integration, automotive scale manufacturing
Inside Tesla's own factories
Low volume production starting summer 2026
Figure 03
Figure AI
Own VLA model, Helix; real operating experience
BMW's Spartanburg plant (logistics)
Already working on the line
Atlas
Boston Dynamics / Hyundai
Maximum physical mobility (56 DoF), autonomous battery swap
Hyundai and Google DeepMind research centers
First batch committed; assembly line deployment from 2028
G1
Unitree
Low price, shipping volume
Pilot projects, research, demos
Selling in volume, thousands of units shipped
Investment: who has put in how much
Money flows through this industry via very different channels, so comparing the raw numbers needs some care. In one case it is venture rounds, in another it is a corporate acquisition, and in a third it is simply a line item in a public company's capital budget.
Tesla doesn't raise separate funding for Optimus at all. The robot is paid for out of the company's own capital expenditures. For 2026, Tesla has guided to more than $25 billion in total capex, but that figure covers several things at once - Optimus, the Robotaxi service, AI infrastructure, and chip manufacturing and the company has not broken out a separate number for the robot alone.
Figure AI, by contrast, is a classic venture backed startup. Since its founding, the company has raised roughly $1.9 billion in total, with more than $1 billion of that coming in a single Series C round in September 2025. That round valued Figure AI at $39 billion post money. Investors include Parkway Venture Capital, Brookfield, Nvidia, Intel Capital, and Microsoft, with the OpenAI Startup Fund having backed the company in earlier rounds.
Boston Dynamics took a different path entirely, changing owners rather than raising venture rounds. In 2021, Hyundai bought an 80 percent stake from SoftBank for $1.1 billion, then in 2026 bought out the remaining stake for roughly $325 million, a deal that values Boston Dynamics at around $3.3 billion, though some analysts have floated much higher potential valuations, into the tens of billions of dollars. Hyundai Motor Group now owns the company outright.
Unitree chose a third route: going public. In August 2026, the company listed on Shanghai's STAR Market, raising about $904 million in the offering, a deal that valued Unitree at roughly $9 billion. DeepSeek came in as a strategic investor in the listing, and Alibaba, Tencent, and automaker Geely had earlier been reported as interested backers.
Finally, Gemini Robotics 2 doesn't have a funding history of its own. Development happens inside Alphabet's internal R&D budget, and Google DeepMind doesn't raise separate investment for the project. Apptronik, whose Apollo 2 robot serves as the demonstration platform for the model, does have a fairly typical venture story: the company has raised more than $935 million in its Series A round, including a February 2026 extension, and is now valued at roughly $5.5 billion.
The main takeaway:
the companies are competing in entirely different financial weight classes. Tesla is spending billions out of its automotive profits, startups like Figure and Apptronik are raising venture money against a future share of the market, Unitree just went public, and Boston Dynamics has spent this whole period simply as an asset inside Hyundai.
Where Gemini Robotics 2 fits in
Google DeepMind's Gemini Robotics 2 deserves a separate mention, since it's sometimes mistaken for "yet another robot." In reality it isn't a physical machine at all, but a multimodal neural network that can be installed on someone else's hardware.
In Google's demo videos, the model controls, for example, the Apollo 2 humanoid from the startup Apptronik, someone else's "body." Notably, Google has also partnered with Boston Dynamics: Gemini Robotics is being integrated into Atlas as well.
This highlights the main split in the market. Some companies -Tesla, Figure, Unitree build the full stack, body plus brain, while others, like Google DeepMind, specialize purely in intelligence and are willing to work with any hardware partner.
Bottom line
It's hard to name a single leader in mid 2026, because the companies are competing on different criteria:
- Real work on an assembly line → Figure AI is ahead: its robots have logged nearly two years combined at BMW's plant
- Physical capability → Atlas from Boston Dynamics leads with its 56 degrees of freedom and autonomous battery swapping
- Affordability and volume → Unitree's G1 is unmatched
- Long term bet on scale → Tesla promises not just individual robots but assembly line economics at automotive scale and that part of the plan still has to be proven in the months ahead
This article draws on materials and statements from Tesla, BMW Group, Boston Dynamics, Hyundai Motor Group, Google DeepMind, and Unitree Robotics, as well as coverage from The Robot Report, Interesting Engineering, TechRepublic, Forbes, Notebookcheck, InsideEVs, Bloomberg, Caixin Global, CNBC, and Yahoo Finance, as of August 2026.
The industry moves quickly, and the production timelines, prices, and financial figures cited here are company stated plans or market estimates at the time of publication and may change.
