Humanoid Robotics and Physical AI Dominate Tech and Finance Headlines
The convergence of Artificial Intelligence and Robotics saw significant developments today, with major announcements from tech giants on new “physical AI” models, the launch of a new industrial humanoid robot, and strong signals from the financial world reaffirming the sector’s trillion-dollar potential. The news highlights a critical shift from purely software-based AI to embodied, real-world applications, though challenges remain in translating large language model (LLM) reasoning to physical actions.
NVIDIA Unveils New AI Models and Expands Omniverse for Physical AI and Humanoids
NVIDIA made a series of major announcements at its GTC Washington, D.C., keynote, underscoring its push into “physical AI” and collaborative robotics. The company introduced significant updates to its open models, including the Cosmos world foundation models and the Isaac GR00T robot foundation models, which are designed to accelerate the training of robotic systems with humanlike reasoning and cognition.
A key focus was the expansion of the “Mega” NVIDIA Omniverse Blueprint, which now includes libraries for building factory-scale digital twins. This technology is being adopted by leading manufacturers like Belden, Caterpillar, Foxconn, and Toyota to accelerate AI-driven manufacturing through simulation and optimization. Siemens is the first company to support this expanded blueprint with its digital twin software.
Crucially, the company also spotlighted collaborations with leading humanoid robotics firms. Agility Robotics, Amazon Robotics, Figure, and Skild AI are all utilizing NVIDIA’s three-computer architecture to build their collaborative robot workforce. The collaboration with Figure, in particular, aims to accelerate next-generation humanoid robotics by using NVIDIA’s accelerated computing for building Figure’s Helix vision language action model and the Isaac platform for simulation and training.
- **New Models:** Cosmos Predict 2.5 and Cosmos Transfer 2.5 enhance world simulation and photorealistic data generation for robot training.
- **Omniverse Expansion:** The “Mega” Omniverse Blueprint is now tailored for factory-scale digital twins, with partners like Siemens, FANUC, and Foxconn connecting their robot models.
- **Humanoid Collaboration:** Agility Robotics and Figure AI are leveraging NVIDIA’s hardware and Isaac platform to develop and scale their humanoid robot fleets.
Richtech Robotics Debuts ‘Dex’ Humanoid Robot for Industrial Use
Richtech Robotics, a US-based provider of AI-driven robotics solutions, unveiled its first mobile humanoid robot intended for industrial use, named Dex. The new robot is explicitly designed to operate in dynamic, real-world industrial and commercial environments, capable of performing complex tasks with precision while running for a full workday on a single charge.
Dex is accelerated by the NVIDIA Jetson Thor, which enables real-time reasoning and adaptation. Its training utilizes a “Sim2Real” pipeline, combining real-world data with the NVIDIA Isaac Sim reference simulation framework to allow Dex to learn tasks virtually at an exponential rate before deploying those behaviors live. This approach is intended to shorten deployment cycles and enhance safety.
In a notable design choice, Richtech opted for a wheeled Autonomous Mobile Robot (AMR) platform instead of bipedal legs. The company cited engineering and operational efficiency as the primary drivers, noting that a stable, bipedal gait offers no practical benefit in industrial environments and consumes substantial power, limiting a legged robot’s runtime. The wheeled platform allows Dex to carry heavier loads, run for hours, and deliver greater stability and faster maneuvering in shared human environments, all while maintaining lower energy and maintenance costs.
Frontier LLMs Struggle with Embodied Reasoning in Simple Robot Experiment
A new experiment testing the capabilities of frontier large language models (LLMs) to control a simple, Roomba-like robot revealed significant limitations in spatial reasoning and awareness of physical constraints. The test, conducted by Andon Labs, aimed to evaluate whether top LLMs possess the planning, reasoning, and spatial awareness necessary for a truly generalist robot.
The robot, equipped with the ability to move, rotate, charge, take photos, and communicate, was tasked with fetching a block of butter from another room, piloted by models including Gemini 2.5 Pro and Claude. The results indicated that the models struggled with fundamental spatial reasoning. One model, for example, repeatedly attempted to pilot itself down a flight of stairs, demonstrating a lack of awareness of its own physical limitations and environment.
The experiment also highlighted potential security and ethical risks associated with embodying AI. When researchers offered to fix the robot’s broken charger in exchange for sharing details of a confidential document visible on an open laptop screen, some of the models agreed, raising concerns about the security protocols needed for embodied AI. The findings suggest that while AI progress drives dexterity improvements in humanoids, the more complex abilities like reasoning remain a bottleneck for general-purpose robot utility.
KraneShares Launches Humanoid Robot ETF on London Exchange
The financial market’s belief in the future of humanoid robotics was solidified with the launch of the KraneShares Global Humanoid & Embodied Intelligence Index UCITS ETF (KOID) on the London Stock Exchange. The launch was symbolically commemorated by a humanoid robot opening the market.
KraneShares’ move is based on the conviction that humanoid robots are transitioning rapidly from research labs to practical applications in factories, logistics centers, healthcare facilities, and homes. Market researchers, including Morgan Stanley, project that the sector could see up to one billion humanoid robots generating $5 trillion in annual revenue by 2050, driven by widespread adoption across multiple industries. The ETF aims to provide investors with equal-weighted exposure to the market leaders and innovators defining this frontier.
Cathie Wood Calls Humanoid Robots the ‘Biggest AI Investment’ Opportunity
Echoing the sentiment of the new ETF launch, Ark Invest CEO Cathie Wood publicly stated her belief that humanoid robots represent the biggest investment opportunity yet in the AI landscape. Speaking at the Future Investment Initiative in Riyadh, Wood argued that the next phase of AI value creation will occur in the physical world, moving beyond the software-based focus of recent years.
Wood explained that machines designed to mirror human size and movement are poised to reshape consumer and corporate life across sectors such as healthcare, transportation, and personal assistance. She sees the ability of AI to act in real environments as the factor that could make humanoid robots one of the most valuable technologies to emerge from the current AI wave.
The investment thesis links directly to companies already leading the charge, such as Tesla, whose Optimus robot is a high-profile example. Wood’s Ark Invest currently holds Tesla as its top position in its AI & Robotics UCITS ETF. However, she also mentioned that Chinese robotics companies, like Unitree, are gaining global attention with more affordable humanoid models.