Figure's Helix vision-language-action (VLA) model lets its robots pick up virtually any small household object, including thousands of items they have never encountered beforeCITE:E1. Figure describes this generalization — following natural-language instructions to manipulate previously unseen items — as a core capability of the Helix system, demonstrated on 2025-02-20CITE:E1. This is the software side of the humanoid robot buildout: a foundation model that generalizes manipulation the way large language models generalize text.
Boston Dynamics has turned to an outside supplier, Hyundai Mobis, to supply actuators for its fully electric Atlas robot, which replaced the earlier hydraulic designCITE:E2. Announcing the move in January 2026, Boston Dynamics and Hyundai Mobis said the two organizations "will work together to build a highly reliable component supply chain and accelerate the pace of actuator development and production"CITE:E2. That a robotics pioneer needs a dedicated external supply chain just for actuators signals that this single component class, not the robot's software, is the harder constraint on scaling production.
Figure says robotics lacks a large-scale training dataset equivalent to those that power vision or language AI — in its words, there is "no 'YouTube for robot behaviors'"CITE:E5. Figure launched Project Go-Big on 2025-09-18 specifically to collect real-world manipulation data at scale to fill that gapCITE:E5. Paired with the actuator constraint above, this points to two separate bottlenecks sitting behind the same headline capability shown by Helix: enough reliable hardware to build robots, and enough real-world data to train them.
NVIDIA has placed humanoid robotics at the center of its "physical AI" strategy, with CEO Jensen Huang declaring on 2026-03-16 that "physical AI has arrived — every industrial company will become a robotics company"CITE:E7. Robot makers including Figure, Boston Dynamics, and Agility have adopted NVIDIA's Cosmos, Isaac, and GR00T foundation models to accelerate developmentCITE:E7. On the manufacturing side, Morgan Stanley's report found that of 100 publicly traded companies worldwide "confirmed to be involved" in humanoid development, 56% are based in China, and China hosts 45% of the world's integrators — firms that customize robots for end usersCITE:E6. The compute and foundation-model layer is anchored by NVIDIA in the U.S., while the identified company base and integration capacity are concentrated in China.
Goldman Sachs Research raised its 2035 total addressable market forecast for humanoid robots more than sixfold, from $6 billion to $38 billion, citing AI progress as the driverCITE:E4. On the ground, Tesla is building toward a one-million-unit annual production line for Optimus in Fremont, but CEO Elon Musk told investors on the company's 2026-01-28 earnings call that meaningful production volume would only arrive "about the end of this year"CITE:E3 — meaning, as of the target he stated, late 2026, not the present.
| Metric | Value | Evidence |
|---|---|---|
| Objects Figure's Helix can grasp | Thousands of previously unseen household items | E1 |
| Goldman Sachs 2035 TAM forecast | $38 billion (up from prior $6 billion projection) | E4 |
| Tesla Optimus target production line | 1 million units/year, Fremont | E3 |
| Tesla's stated timeline for meaningful production | "About the end of this year" (2026), per Musk, Jan 2026 | E3 |
| Humanoid companies based in China | 56% of 100 identified worldwide (Morgan Stanley) | E6 |
| Global robot integrators based in China | 45% | E6 |
The pieces cited above sit in tension with each other. Figure's Helix already generalizes to thousands of unseen objectsCITE:E1, yet Figure itself says robotics still lacks the large-scale dataset that vision and language AI rely onCITE:E5, and Boston Dynamics needed an outside partner just to secure actuators for its electric AtlasCITE:E2. Goldman Sachs' sixfold-higher $38 billion market forecastCITE:E4 and Tesla's one-million-unit Fremont production-line targetCITE:E3 describe where the industry is aiming; Musk's own "about the end of this year" timeline for meaningful outputCITE:E3 marks how far that aim currently sits from delivered volume. Meanwhile, NVIDIA supplies much of the foundation-model layer that companies like Figure, Boston Dynamics, and Agility now build onCITE:E7, while Morgan Stanley's count places the majority of identified humanoid companies and integrators inside ChinaCITE:E6 — two different layers of the same supply chain, sourced from two different bases.
Figure's Helix vision-language-action (VLA) model lets its robots pick up virtually any small household object, including thousands of items they have never enc…
Boston Dynamics has turned to an outside supplier, Hyundai Mobis, to supply actuators for its fully electric Atlas robot, which replaced the earlier hydraulic d…
Figure says robotics lacks a large-scale training dataset equivalent to those that power vision or language AI — in its words, there is "no 'YouTube for robot b…
NVIDIA has placed humanoid robotics at the center of its "physical AI" strategy, with CEO Jensen Huang declaring on 2026-03-16 that "physical AI has arrived — e…
Figure's Helix demo shows the perception-and-language side of physical AI already works — robots generalizing to thousands of unseen objects. But two of the industry's own leaders are the strongest evidence that hardware and data, not software, are the binding constraints: Boston Dynamics had to bring in Hyundai Mobis just to build an actuator supply chain, and Figure had to launch its own Project Go-Big because no large-scale robot-behavior dataset exists. Goldman Sachs' sixfold-higher $38 billion 2035 forecast is effectively a bet that these two bottlenecks get resolved on schedule. The one indicator worth watching is whether Tesla actually reaches Musk's own stated bar — meaningful Optimus production by "about the end of this year" — because a slip there would say more about actuator and data readiness across the whole industry than about any single company's execution.
Utility-scale solar has become one of the cheapest new power sources, with costs down roughly 90% since 2010, while 2024 deployment volume outpaced every other generation technology. Paired with record-low battery prices, solar-plus-storage is already powering an AI data center in Arizona — though it still falls short of full 24-hour dispatchable baseload.
2026 marks the European Union AI Act's real enforcement start: the AI Office gains fining power over general-purpose AI on August 2, while high-risk system deadlines are pushed to 2027 and 2028. California activates two new laws on January 1 covering frontier-developer safety disclosure and training-data transparency. The federal government moves the opposite direction, ordering a Justice Department task force to challenge state AI laws, naming California's SB 53 as a target.
China's EHang (億航) is the only eVTOL maker actually flying paying passengers today, holding a full Chinese type, production, and operating certificate set and delivering 221 aircraft in 2025. US rivals Joby and Archer remain in certification or pre-launch stages, while Germany's Lilium and Volocopter both went insolvent in 2024–2025, with Volocopter absorbed by a Chinese buyer.