By: Akaash TK, Research Associate, Autonomous Technology & Robotics
In late October 2025, 1X Technologies opened pre-orders for its NEO home robot and ran through its entire first-year production slot, 10,000 units, in five days. A few months later, Figure took its BotQ factory from one robot a day to one robot an hour, a 24-fold jump in under four months. Tesla, meanwhile, announced it is ending Model S and Model X production at Fremont and converting that line to build its Optimus humanoid.
Unlocking labor at scale, across warehouses, assembly lines, and homes, requires robots that can generalize across tasks rather than execute a single routine. That is what the industry is now building toward, and ARK puts the addressable market for this shift at ~$26 trillion, split almost evenly between industrial and household labor.
The Companies Building The Future
ARK follows both the incumbents that dominate today’s specialized robots and the pioneers building generalizable platforms.
Tesla (Optimus) - Tesla is running its robot program on the same vision and compute stack that powers Full Self-Driving (FSD). The Gen 3 hand alone carries 22 degrees of freedom, twice that of its predecessor. The company plans to reveal Optimus V3 in summer 2026, the first version engineered for volume manufacturing, with production on the converted Fremont line to follow.
Figure AI - Figure builds its robots in-house at BotQ, its high-volume manufacturing facility, and powers them with Helix, its proprietary vision-language-action (VLA) model. Every robot in the field continuously sends real-world data back to improve Helix, enabling fleet-wide over-the-air updates that make the entire group smarter with each deployment. Recently, the company live-streamed its Figure 03 robots completing what began as a planned 8-hour logistics shift, which ran autonomously for over 200 hours straight, sorting nearly 250,000 packages with zero hardware failures.
Apptronik (Apollo) - Apptronik’s Apollo is already working in cordoned-off zones at Mercedes-Benz, GXO Logistics, and Jabil, performing real industrial tasks such as sorting, material transport, kitting, and lineside delivery. The company benefits from a major strategic partnership with Google DeepMind, which integrates advanced AI models, including Gemini Robotics, to accelerate Apollo’s embodied intelligence and real-world reasoning. With deep roots in NASA robotics expertise, including work on the Valkyrie platform, Apptronik places strong emphasis on safety, reliability, and industrial readiness.
Physical Intelligence (π) - Physical Intelligence is building the “brain” layer for the robotics industry rather than its own hardware. The company develops general-purpose vision-language-action (VLA) models, with its latest, π0.7, demonstrating impressive open-world generalization, enabling robots to perform tasks they were never explicitly trained on.
Unitree - China’s Unitree has emerged as one of the world’s highest-volume humanoid manufacturers, shipping over 5,500 units in 2025. It has achieved profitability since 2024 as per the prospectus filed for its Shanghai market initial public offering (IPO) in March 2026. The company is targeting up to 20,000 humanoid deliveries in 2026, highlighting China’s significant lead in cost efficiency and manufacturing scale. The units shipped so far skew towards research, development, and entertainment purposes. The breakthrough driving the shift.
What changed is how robots are programmed. Older industrial arms run fixed routines. Hand a part at an unfamiliar angle or an unknown object, and it stalls. The new platforms run on vision-language-action models: they take in camera input alongside a plain-language instruction and produce motion directly. They learn from human demonstrations and fleet data rather than pre-engineered motions. That ability to generalize is what finally opens up messy, unstructured environments for robots.
No ready corpus of this data exists, which makes it a near-term bottleneck, though start-ups around the world are tackling it. The real difficulty is the path to collecting it, and there are two. Field deployment is circular: a robot has to run at low failure rates before it starts logging useful data. Simulation only helps once there is enough real data to close the gap between simulated and real motion, which is why the parallel race is to build world models accurate enough that skills learned in simulation hold up on real hardware. Either way, the compute requirement only matters if one of these paths keeps the data flowing.
Source: ARK Investment Management LLC, 2026, based on data from Google Deepmind 2023, Humanoid 2025, and Figure AI 2025 as of December 18, 2025. In addition to those sources, certain information presented may be the result of ARK’s internal analyses, which draw on various additional sources of information. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of future results. Forecasts are inherently limited and cannot be relied upon.
The question now is how fast. Humanoid work is significantly harder than driving. ARK estimates the complexity ratio at around 200,000 times that of a robotaxi, accounting for degrees of freedom, balance, perception, and task variety. Tesla’s FSD shows what sustained compute investment can do for a hard perception-and-control problem. By mapping the compute behind FSD’s performance gains and extending the curve, ARK projects that Optimus could reach human-level task performance by the end of the decade, assuming compute infrastructure and robot hardware keep improving.
Note: *Cumulative AI Compute Units are defined as the total number of NVIDIA H100-equivalent compute units, benchmarked to H100 performance at launch, required to solve a given task. “MW”: Megawatts. Source: ARK Investment Management LLC, 2026, based on data Tesla 2025a, Tesla 2025b, and Tesla 2024, as of December 18, 2025. In addition to those sources, certain information presented may be the result of ARK’s internal analyses, which draw on various additional sources of information. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of future results. Forecasts are inherently limited and cannot be relied upon
What We Believe The Skeptics Get Wrong
Our view differs from the popular concern that robots this capable will displace large numbers of workers. From about 1950 to 2000, automation and productivity climbed while the US labor market kept expanding. Labor force participation only began to diverge after 2000, and that is mostly attributable to aging demographics and globalization rather than to machines replacing workers. As productivity rose, each hour of human work became more valuable, and people produced more while working fewer hours. ARK’s broader bet is that general-purpose robots will spin up entirely new job categories, the way earlier technologies did, even if we can’t picture them yet. The displacement risk is real and worth taking seriously, especially through short transitions, but ARK’s research points to augmentation as the dominant effect over time.
Source: ARK Investment Management LLC, 2026, based on data from U.S. Bureau of Labor Statistics 2025a, 2025b, and Organization for Economic Co-operation and Development 2023, as of December 18, 2025. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of future results. Forecasts are inherently limited and cannot be relied upon.
What To Watch Next
A few markers to watch over the next eighteen months:
Tesla’s V3 ramp - The reveal is the easy part. The open question is whether Tesla can move from a deliberately slow first run to real volume on a brand-new line.
Commercial conversion - A number of companies have pilots running inside partner facilities. Watch whether 2026 pilots turn into scaled, paid orders in 2027.
The first home data - NEO deliveries this year will be the first real test of a consumer humanoid in actual houses, including how often it still needs a human in the loop.
China’s output - Nearly 90% of humanoid robots sold globally in 2025 were Chinese. Unitree is already selling robots at price points well below Western competitors, the same cost dynamic that reshaped elective vehicles (EVs).
The Bottom Line
For sixty years, getting more out of a robot meant narrowing its environment. The bet now is the reverse: one platform that learns many jobs, in any setting, and improves with every unit deployed. If that holds, robotics stops being a line item in a handful of factories and starts looking like infrastructure across the economy.
Important Information
ARK Investment Management, LLC (“ARK”) may hold a financial interest in the companies discussed in this article through various strategies and investment vehicles it manages. Readers are urged to use caution when considering the forecasts and other forward-looking information provided in this article, as it is inherently subjective and reflects ARK’s inherent bias toward positive expected results. There is no guarantee that actual results will align with the forecasts, and they might not be predictive. The information provided in this article is for informational purposes only and does not constitute investment advice. All statements made regarding companies or securities are strictly beliefs and points of view held by ARK and are not endorsements by ARK of any company or security or recommendations by ARK to buy, sell or hold any security. Historical results are not indications of future results.
Certain of the statements contained in this material may be statements of future expectations and other forward-looking statements that are based on ARK’s current views and assumptions and involve known and unknown risks and uncertainties that could cause actual results, performance or events to differ materially from those expressed or implied in such statements. ARK assumes no obligation to update any forward-looking information contained in this material. Certain information was obtained from sources that ARK believes to be reliable; however, ARK does not guarantee the accuracy or completeness of any information obtained from any third party.