How to Think About Humanoid Robots
Elon claims 80% of Tesla's value will be in Optimus. Nvidia CEO predicts mass robot deployment in five years. But humanoid robots aren't about convenience—they're about harvesting your most data.
How to Think About Humanoid Robots
The current AI world is alive with the idea that soon, AI humanoid robots will provide a meaningful new utility and business automation platform for countless types of human work.
A number of current players in the AI automation space are showing advancements:
Tesla (Optimus): Leading in scale, with plans for high-volume production. Optimus V3 prototypes show improved agility, walking, and task performance like folding clothes or dancing.
Nvidia (Project GR00T): Focuses on AI platforms for robotics, enabling simulation-to-real-world training. Tools like generative AI help robots learn from videos.
Figure AI: Emphasizing generalization, with robots like Figure 02 deployed in trials (e.g., BMW factories).
Boston Dynamics (Atlas): Known for dynamic mobility, now integrating AI for complex maneuvers.
Chinese firms (e.g., Unitree, UBTech): Rapid deployment in industry, with government support aiming for widespread use by 2027.
Others: Apptronik (Apollo for logistics), 1X (NEO for home assistance), Agility Robotics (Digit for warehouses).
And a number of luminaries are all weighing in on the potential of the new technology to drive their business, and user experiences.
Elon Musk has declared that:
~80% of Tesla’s value will be Optimus.
Source: https://x.com/elonmusk/status/1962618811141812475
From Nvidia CEO, Jensen Huang:
Humanoid robots are less than five years away from seeing wide use in manufacturing facilities
Also from Huang:
Everything that moves will be robotic someday, and it will be soon. And every car is going to be robotic. Humanoid robots, the technology necessary to make it possible, is just around the corner.
Source: https://www.laptopmag.com/laptops/nvidia-ceo-jensen-huang-robots-self-driving-cars-
From OpenAI’s Sam Altman:
I’m excited about a future where signing up for ChatGPT’s highest tier gets you a free humanoid robot... The mechanical engineering and AI for humanoids are quite hard but feel within grasp. Making a billion humanoid robots will take a while... Humanoid robots walking around will feel like the future.
Source: https://x.com/TheHumanoidHub/status/1936494676770803837
What People Really Want:
Humanoid robots hold the answer to the widely memed need that people want AIs to do real tasks around the house, so that people can work on art and writing, not the other way round:
Source: https://x.com/AuthorJMac/status/1773679197631701238
What People Will Get:
Buried in all of this ideation is the reality that AI models have already digested the entire corpus of publicly available books, papers, videos, and internet traffic from the dawn of human civilization.
The amount of energy needed to produce new insights from existing training resources is growing larger. And in the US (at least) the reality of competing energy demands are causing AI companies to look for other options.
Various companies all have private data troves that are being digested in a piecemeal fashion based on individual deals between AI providers and the various data trove owning companies.
Separately, existing robot-specific datasets are far smaller. Scaling these sets up is slow because real-world actions are time-consuming to produce. Simulations and human video provide some benefit, but transferring results to physical robots—especially for intricate tasks—remains challenging.
The better option for AI companies is to digest new large sources of data.
Humanoid robots answer that need for data if they collect video and audio while operating, which is one of the expected ways that the new technology will defer costs for new buyers.
How they will work:
Likely Humanoid Robots will be divided into different use case groups catering to different risk/reward environments, which will service different users’ needs.
Some will function in home environments as butlers and maids, providing automated simple functions like doing laundry, cleaning, cooking and serving food, and handling groceries.
Protocols will be developed to handle work in programmed sequences, and to coordinate with approved deliveries of packages, mail, and groceries to residences.
Information from a given family member’s phone and calendar will be provided to alert the robots to arrival times, and the best times to prepare meals.
Other specialized robots will be used for high risk human work where the potential for human casualties is high. The obvious example is types of conflict jobs. War-fighting drones, police robots. and security robots will all be deployed in areas that would be high risk to humans.
Key Challenges:
Technical Limitations:
The tradeoff for robots put in high risk environments will be that risky environments will often come with complex tasks, and task switching.
Complex tasks and time constrained task switching will likely be some of the last bastions of human cognition as more capable robots are developed.
Currently robotics and AI research shows an 'uncanny valley' problem of small tasks where robots can perform either extremely simple operations (like lifting objects) or highly complex specialized functions (like precision surgery), but fail spectacularly at mundane middle-ground activities that humans consider trivial—such as folding a fitted sheet, mopping around furniture, or troubleshooting when the internet connection drops during a task.
Employment Impact:
Humanoid robots will simultaneously destroy and create jobs at different speeds, creating a painful transition period. While they'll rapidly eliminate manual labor positions—from warehouse workers to cleaners—the new roles they generate in robot maintenance, programming, and interaction design will emerge slowly. A displaced factory worker can't immediately become a robot technician without years of retraining. This timing mismatch means decades of economic disruption as one generation of workers faces obsolescence while new robot-adjacent careers gradually develop. The tradeoff: immediate job losses for eventual productivity gains that may take a generation to fully realize.
Business Models:
Robot financing could take on many flavors, from outright purchases and maintenance like a car, to the leasing model of Xerox machines, or even as "Robot-as-a-Service" subscriptions where you pay monthly for capabilities rather than owning hardware—meaning your home robot could lose features if you miss payments or gain new abilities through software updates, fundamentally shifting from tool ownership to renting access to AI functionality.
Economics measurement challenge:
GDP famously does not account for home improvements, so GDP measures of robot labor in the home will likely undercount the true economic value these machines provide.
Some value will be captured in the sale or lease of the robot, and when a robot does laundry, cleans floors, or prepares meals, it's performing work that was previously unpaid household labor—work that economists estimate would add trillions to GDP if properly valued. This creates a measurement paradox: as robots take over more domestic tasks, official economic statistics may show some productivity growth even as people's quality of life dramatically improves through freed-up time and reduced household drudgery.
Implementation Considerations:
Since the dawn of computing, data storage has moved pendulum-like from a server-terminal construction, to personal PCs, to the cloud, moving from remote storage to local storage and back to remote storage.
The question of where a company’s or family’s robot instruction and knowledge base are stored would likely align with the current market preference for data location.
Currently, the majority of data is stored in cloud architectures, but as privacy concerns weigh on the calculation of how best to integrate a robot into a business workplace, or residence, that pendulum might swing again towards local storage, to provide some level of privacy for robot purchasers.
AI Agents are already using knowledge bases and instruction sets that are designed to be separate from the backend LLMs that support the Agents themselves. So the ease of doing this might be simple, but the implementation by the robot selling companies will dictate terms.
Key takeaways:
The humanoid robotics industry sits at a critical juncture where ambitious timelines from tech leaders clash with significant technical, economic, and social realities that will likely determine the pace and scope of actual deployment.
Reality vs. Hype: The humanoid robotics revolution promises to reshape how we live and work, but the reality will likely be more nuanced than the bold predictions from industry leaders who envision seamless integration within five years.
Data-Driven Business Model: Success will fundamentally depend on data collection capabilities, making privacy and data ownership critical considerations for consumers and businesses.
Technical Hurdles Remain: Real-world manipulation, task generalization, and reliable operation present significant challenges that must be overcome before widespread adoption.
Specialized Deployment First: The most pragmatic path involves robots excelling in controlled environments—warehouses, factories, and high-risk scenarios—before gradually expanding to domestic applications.
Trade-offs Are Inevitable: While robots may free us from dangerous or repetitive tasks, they'll introduce new dependencies, privacy concerns, and economic disruptions.
Integration Over Innovation: Success depends not just on technological advancement, but on thoughtful integration that balances automation benefits with human agency and choice.