Author: Rinat Mirzaitov

Abstract This paper examines the race to build general-purpose physical AI, meaning learned systems that can do open-ended physical work in unseen places and on different robot bodies. Using filings, research, standards and market data available in September 2026, it proposes a six-level scale of physical generality, P0 to P5, and places the field at the boundary between P2 and P3. The best systems reach 56% zero-shot success on household tasks in unseen homes, and about 19,100 humanoids shipped worldwide in the first half of 2026, more than 97% of them from Chinese vendors. The race is harder than the…

Useful physical data remains the better-evidenced bottleneck in robot learning. Figure’s Helix 2.5 results strengthen that assessment: the company reports that pretraining on its Index dataset raised full-task success in unfamiliar homes from 9% to 56%. [1] For investors, the infrastructure thesis is becoming more concrete. Developers are paying to acquire experience and committing to larger computing systems. The evidence supports a market for collecting, processing and learning from physical data. It does not yet establish that general-purpose robotics requires more training compute than frontier language models. The distinction matters. Demand for infrastructure can become commercially significant before compute becomes…

The humanoid robot market is broader than factory automation, but its major segments are developing at different speeds. Current evidence supports five distinct buyer markets: research, development and experience; data production and model training; industrial and commercial work; household and personal assistance; and military applications. The important market signal in 2026 is the separation between shipment volume and operating proof. Chinese manufacturers dominate reported unit shipments, yet researchers disagree sharply about what those robots are being used for. Research, education, performances and AI data production account for a substantial share under some estimates. Meanwhile, several of the strongest customer-confirmed examples…

Humanoid robotics may be getting close to its ChatGPT moment, but not for the reason another polished robot video has gone viral. The more important signal is underneath Figure’s new Helix 2.5 demonstration. Figure says a humanoid pretrained on a large dataset of human behavior was able to take three learned household behaviors into 30 homes it had never seen, using unfamiliar objects and without collecting new training data in those homes. In a controlled comparison disclosed by Figure, Index pretraining increased full-task zero-shot success from 9% to 56%. That is still far from a dependable household robot. A 56%…

XPENG’s IRON humanoid already has enough public evidence to make one unconventional use case worth taking seriously: fashion. That does not mean IRON is ready to replace professional models. It means the physical characteristics XPENG has chosen to emphasize, human-like proportions, smooth walking, customizable body forms and deliberately model-like movement, align unusually well with fashion shows, luxury retail and experiential marketing. XPENG itself has not announced fashion modeling as an IRON commercial program. Its stated near-term plans focus on company stores, campuses and shopping-guide roles. Fashion is therefore an analytical extension of demonstrated capability, not a confirmed customer market. The…

The humanoid robotics race has a new class of competitor. For the past several years, the industry has largely been framed as a contest between companies building robot bodies, actuators, hands, control systems and manufacturing capacity. Now two of the most powerful frontier AI laboratories are moving directly into the physical layer, but they are approaching it from very different directions. OpenAI CEO Sam Altman made the clearest declaration yet in a September 1 interview with Alex Heath. Asked whether OpenAI was building a humanoid robot, Altman replied: “We will definitely do a humanoid.” He added that OpenAI expects to…

Anthropic’s August 27, 2026 research preview of the Model Hardware Standard, or MHS, puts a useful question in front of the robotics industry: how much of Physical AI’s difficulty comes from intelligence itself, and how much comes from the fact that every machine exposes the physical world differently? MHS proposes a shared driver layer that lets AI agents discover programmable devices, inspect their state and capabilities, and operate them through common primitives such as read and write. Anthropic says device descriptions can also expose characteristics and enforced safety limits. Universal Robots subsequently said it tested the preview on four cobots,…

Humanoid robotics has an unusually important characteristic for startups: many of the industry’s biggest problems are not confined to the companies building complete robots. They sit underneath them. Actuators remain expensive and difficult to scale. Dexterous hands still struggle with the combination of strength, compactness, sensing, and reliability. Robot learning requires large amounts of physical-world data. Safety requirements become harder when robots leave cages and work around people. Factories still need robots connected to existing equipment, software, networks, and production processes. Once deployed, somebody also has to measure reliability, diagnose failures, recover from exceptions, and keep fleets productive. That creates…

Smart Analytics Global (SAG) estimates that 19,100 humanoid robots shipped worldwide in the first half of 2026, up 272% from 5,100 a year earlier. AGIBOT accounted for approximately 8,400 units, Unitree 5,900, Galbot 900, UBTECH 700 and Leju 600. AGIBOT and Unitree alone represented about 75% of reported shipments. Those numbers are significant. They show an industry moving from prototypes toward manufacturing volume. But they do not answer the more commercially important question: Where are those robots actually working? That question matters because a robot can be produced, sold, shipped, delivered, accepted and operated at different points in its commercial…

Humanoid robotics is usually framed as a race among companies building complete robots. Humanoid Analytics’ July 2026 market snapshot shows a broader competitive structure. Of 351 tracked Core and Adjacent entities, 225 are core humanoid builders or substantial corporate programs. Another 126 sit in Adjacent segments including dexterous manipulation, sensing, components, AI, data infrastructure and system integration. Adjacent companies therefore represent 35.9% of the tracked Core and Adjacent universe. That number is the useful finding. It is not market share. It does not measure revenue, adoption, maturity, technical quality or market value. It shows that more than one-third of the…