The "Physical AI" Turning Point at CES: Robotaxi Moves Towards Scaling, Humanoid Robot Supply Chain Quietly Forms

2026 may mark the beginning of AI's large-scale entry into the physical world—from walking robots to autonomous vehicles, AI is accumulating ecological hardware.

According to news from the Wind Trading Desk, a research report released by Deutsche Bank on January 13 shows that the bank's analyst team attended the CES exhibition in Las Vegas last week and felt a significant surge in market enthusiasm and relevance. The bank pointed out that autonomous driving vehicles (Robotaxi + consumer-grade L4) and the most eye-catching humanoid robots occupied the center stage of the exhibition.

Deutsche Bank summarized in the report: “Overall, we predict that 2026 will be a year when autonomous vehicles increasingly transition from testing/validation to scaling, while humanoid robots will shift from laboratory experiments to small-scale deployments.”

The report emphasizes that the humanoid robot field is nurturing a brand new supply chain, with suppliers attempting to transition to this field in hopes of achieving large-scale production in the future. Meanwhile, the deployment momentum of Robotaxi in the autonomous driving field is strong, with chip giants like NVIDIA reshaping the competitive landscape by launching new platforms.

Deutsche Bank listed 10 core observations in the report:

1. Humanoid Robot Supply Chain Taking Shape: Actuators Become the "Muscle" Entry Point

Deutsche Bank believes that although still in the early stages, suppliers are already transitioning to the humanoid robot supply chain, with a path similar to electric drive systems: providing both integrated solutions and underlying components.

• Schaeffler is attempting to become the main "muscle" for humanoid robots, providing linear and rotary actuators. It showcased an integrated planetary gear actuator for humanoid robots at CES: a compact integration of a two-stage planetary gearbox + motor + encoder + controller. This unit has high thermal stability, a torque range of 60–250 Nm, and very low backdrive capability, able to withstand external forces and avoid unintended reversal of drive components, suitable for continuous operating conditions. Deutsche Bank mentioned that NEURA has agreed to use Schaeffler actuators in its humanoid robots, and it seems that other customers are already using (at least some components) or will use them in the future.
• Hyundai Mobis also announced that it will supply actuators for Boston Dynamics' Atlas, aiming to enable the robot to leverage the automotive supply chain for manufacturing.

As the supply chain begins to "automobilize," what gets priced first is often not the concept, but the penetration of key components and the ability for large-scale manufacturing.

2. Onboard Chip Landscape: NVIDIA Remains the Preferred Choice, but Differentiation Begins to Appear

Deutsche Bank observed that NVIDIA still dominates onboard processors for humanoid robots, mainly due to performance and ease of use. Companies using Jetson Orin or Thor include: 1X, Agility, Apptronik, Boston Dynamics, Figure AI, Mentee, (currently) NEURA, UBTECH, Unitree, and others.

In contrast:

• Tesla and XPeng use self-developed inference chips.
• Qualcomm launched the next-generation "full-stack architecture" solution for robots (Dragonwing IQ10 Series) at CES, but Deutsche Bank stated it is unclear whether it can achieve large-scale adoption by customers; meanwhile, VinMotion's Motion 2 humanoid robots use the IQ9 Series, while the IQ10 is initially aimed at industrial AMRs and more advanced full-size humanoid robots.

3. "Physical AI" moves from scripts to Agentic: VLA becomes the main line

One of the most significant paradigm shifts on-site is the transition from "pre-programmed/scripted actions" to Visual-Language-Action (VLA), enabling robots to "reason" to complete tasks.

• Boston Dynamics replaced traditional MPC (Model Predictive Control) with Google DeepMind Gemini Robotics' VLA model, allowing Atlas to understand previously unseen environments (such as unstructured factory chaotic scenes).
• Its action execution is supplemented by TRI's Large Behavior Model (LBM), similar to Figure's Helix dual-system model: System 1 for high-frequency rapid responses, System 2 for low-frequency high-level reasoning and language; Deutsche Bank also pointed out that Figure seems to be developing two sets of models in-house.

4. The training competition escalates: Real-world data and simulation "closed loop" are the focus

Deutsche Bank assesses that the industry debate has shifted from "simulation vs. real, which is better" to "how to efficiently close the loop."

• NEURA adopts a more "physics-first" approach, building a large physical training center called NEURA Gym, believing that simulation is "approximate" and will distort in complex contacts (e.g., "threading a needle"); it collects high-fidelity data through hundreds of robots performing sorting, assembly, and other real tasks, then inputs it into the "Neuraverse" to generate real failure "synthetic twins" for training in simulation, ultimately pushing repair solutions back to real robots.
• Another company mentioned the inability to simulate the "tactile" feel of objects, requiring a human to demonstrate first: through remote operation, a person in a VR suit controls a humanoid robot to perform actions like "picking up grapes"; after a few "perfect examples," it then uses NVIDIA GR00T-Mimic to generate "100,000+" action variants in simulation and employs reinforcement learning to make the actions smoother.
• In contrast, Mobileye emphasizes that its Mentee will primarily train using simulation.

5. "General" gives way to "specific roles": Commercial proof takes priority

Deutsche Bank believes that in the short term, "general humanoid robots" will more likely be introduced into specific scenarios to prove commercial viability before discussing entry into households.

• Keenon Robotics (China): Already holds 40% of the global service robot market share, with cumulative exports of approximately 100,000 units overseas; product prices range from below 10,000 to about 100,000, focusing on strong task customization. The highlight of CES 2026 is its flagship humanoid robot XMAN-R1, which can make popcorn, pour drinks, and interact with human-like gestures; its "Brain" is the Keenon Operator Model 2.0, with a VLA model aimed at the service industry, capable of understanding commands like "find the guest at table 4 and give him candy." Keenon also mentioned building a collaborative ecosystem at the Shangri-La Hotel in Shanghai: the MAN-R1 serves as the human-machine interaction "face," the W3 delivers items to rooms, the S100 carries heavy luggage, and the C40/C55 handles cleaning. Additionally, in high labor cost markets like Japan, the lifespan of its robots reaches 8 years, significantly higher than the industry norm of 3–5 years.
• Deep Robotics focuses on industrial inspections: measuring coverage distance (up to 63km), capable of performing 24/7 autonomous patrol monitoring in hazardous areas such as substations, power plants, and oil and gas facilities; used in emergency scenarios for disaster relief, firefighting, and toxic gas detection, employing replaceable batteries to reduce charging friction.

6. The cost reduction formula is straightforward: Scale = the premise for cost decline

On the humanoid robot side, Deutsche Bank attributes the main driver of cost reduction to: increased volume diluting expenses + improved supplier bargaining.

• Some companies claim costs have dropped from “$200,000 to $100,000” and plan to reduce it to “$50,000” in the “next few years,” provided sales reach several thousand units.
• Boston Dynamics and Hyundai Motor announced a goal of achieving an annual production capacity of 30,000 units by 2028; and its 2026 production has already been fully allocated to Hyundai's automotive factories. The company also noted that actuators account for about ~60% of the BoM, and this part will be manufactured by Hyundai Mobis within the Hyundai system to accelerate scaling.
• Mobileye disclosed under the backdrop of acquiring Mentee: if the annual production reaches 50,000 units, the manufacturing cost of a simplified design (without tendon drive system) is about “$20,000/unit”; if the annual production reaches “100,000 units,” the cost can be halved to “$10,000/unit,” aiming for a ramp-up by 2028, with production handled by Aumovio.

7. Robotaxi momentum is stacking up: 2026 looks more like a "commercialization acceleration year"

Deutsche Bank believes that with Tesla launching Robotaxi in 2025, the commercialization momentum of multiple players will be stronger in 2026, as evidenced by the large presence of Waymo and Zoox at CES:

• Waymo: Since its establishment, it has provided over 10 million paid rides; the latest disclosure shows it will reach 450,000 paid rides per week by December 2025, expanding to Houston, Miami, and international markets such as Tokyo and London.
• Amazon's Zoox: Transitioning from public testing in Las Vegas to showcasing a "market-ready product," focusing on a "cab-style" Robotaxi for dense urban areas, completely devoid of a traditional cockpit.
• Mobileye and Volkswagen: This year, they will launch L4-level Robotaxi services in Los Angeles using specially prepared ID. Buzz electric vans.
• Additionally, a self-driving vehicle program based on Lucid Gravity, jointly promoted by partners Nuro, Lucid, and Uber, is set to launch in the San Francisco Bay Area by the end of 2026, with plans to expand to more cities

8. NVIDIA Alpamayo: Packaging "Brain + Skull" for Automakers, but Validation is Still on the Way

NVIDIA announced the launch of Alpamayo ("Brain") for autonomous driving, paired with Thor ("Skull"), attempting to lower the threshold for automakers to deploy advanced capabilities: companies like Lucid and Mercedes can directly "plug in" NVIDIA's solution without investing billions from scratch to build AI infrastructure.

Deutsche Bank remains cautious: this indeed raises discussions about Tesla's moat, but concerns are premature at this stage; NVIDIA still needs traditional OEMs to deliver on their promises, and whether its model can cover real-world boundary cases remains to be seen. Deutsche Bank pointed out that its training data volume is only a fraction of what Tesla has collected.

Even if Alpamayo performs ideally, Deutsche Bank still believes Tesla has a structural cost advantage due to vertical integration (complete vehicles, chips, AI infrastructure, networks, etc.); if autonomous driving/Robotaxi becomes commoditized, cost will be the biggest differentiator.

9. Aptiv: End-to-End AI ADAS + Connectivity and Software Platform, Focused on "Cross-Industry"

Aptiv's core showcase is the next-generation end-to-end (E2E) AI-driven ADAS platform: achieving L2++ hands-free driving "human-like logic" in complex urban environments with the newly released Gen 8 radar and PULSE sensors.

On the software side, it launched the cloud-native middleware platform LINC co-built with Wind River, achieving truly software-defined vehicles through 5G and C-V2X; and demonstrated with Verizon how vehicles can "see pedestrians/cyclists around the corner" by sharing real-time data. Aptiv also emphasized the expansion of sensors into aerospace and collaborative robotics—Deutsche Bank believes this is the narrative "New Aptiv" needs to prove to seek a revaluation of its multiples.

10. Visteon: 700 TOPS Domain Controller, Plug-and-Play Upgrades, Focused on "Execution"

Visteon unveiled the SmartCore HPC domain controller at CES, with a computing power of 700 TOPS, capable of integrating up to 14 cameras and multiple high-speed data connections into a single "central brain." It also expanded its collaboration with Mahindra, launching the SmartCore Pro (triple screens + 360-degree view) for the upcoming XUV7X0.

To address the "legacy platform" barrier, Visteon also introduced a plug-and-play solution powered by NVIDIA DRIVE AGX Orin AI-ADAS Compute Module, allowing automakers to add AI assistants or safety features without completely restructuring their architecture; Deutsche Bank mentioned that this product has already been adopted in China's ZEEKR models.

Additionally, Visteon launched the "Entry Cockpit" for screens under 7 inches, bringing mobile screen projection and digital navigation to two-wheeled vehicles and entry-level models. Deutsche Bank commented that its "surgical" vertical integration helps cost competitiveness and promotes further expansion among automakers (especially Asian OEMs) where penetration has been relatively low According to Deutsche Bank, the message conveyed by CES 2026 is very direct: autonomous driving and humanoid robots are shifting from "can it be done" to "can it be scaled, and can costs be reduced."

When Boston Dynamics pointed out that the cost of actuators accounts for about 60% and announced the pre-allocation of production for 2026, the industry has begun to price using manufacturing language; meanwhile, Waymo's 10m+ paid rides and 450,000 rides per week rhythm have pushed Robotaxi from concept to more concrete operational data. For investors, the next phase to track is not flashier demos, but supply chain binding, capacity ramp-up, and unit cost curves.

The above exciting content comes from [ZhuiFeng Trading Platform](https://mp.weixin.qq.com/s/uua05g5qk-N2J7h91pyqxQ).

For more detailed interpretations, including real-time analysis and frontline research, please join the【 [ZhuiFeng Trading Platform ▪ Annual Membership](https://wallstreetcn.com/shop/item/1000309)】

[![](https://wpimg-wscn.awtmt.com/3c4a713c-7a38-4582-9850-d0eabaf0e7ad.png)](https://wallstreetcn.com/shop/item/1000309)