The "Physical AI" Turning Point at CES: Robotaxi Moves Toward Scale, Humanoid Robot Supply Chain Quietly Takes Shape

The "Physical AI" Turning Point at CES: Robotaxi Moves Toward Scale, Humanoid Robot Supply Chain Quietly Takes Shape

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

According to Windchase Trading Desk, a research report released by Deutsche Bank on January 13 reveals that its analyst team attended the CES convention in Las Vegas last week and directly felt a significant surge in market heat and relevance. The bank notes: autonomous vehicles (Robotaxi + consumer-grade L4) and the most eye-catching humanoid robots took center stage at the exhibition.

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

The report highlights that the humanoid robot sector is cultivating an entirely new supply chain, with suppliers seeking to pivot into the field in hopes of future large-scale releases. Meanwhile, Robotaxi deployment in autonomous driving is gaining momentum, and chip giants like Nvidia are reshaping the competitive landscape by rolling out new platforms.

Deutsche Bank lists 10 core observations in the report:

1. Humanoid Robot Supply Chain is Taking Shape: Actuators Become the “Muscle” Entry Point

Deutsche Bank believes that while it’s still early, suppliers are already transitioning toward the humanoid robot supply chain, a path similar to electric drive assemblies: offering both integrated solutions and underlying components.

  • Schaeffler is attempting to become the main “muscle” for humanoid robots, providing linear and rotational actuators. At CES, it showcased an integrated planetary gear actuator for humanoid robots: a compact integration of a two-stage planetary gearbox, motor, encoder, and controller. This unit features high thermal stability, a torque range of 60–250 Nm, and very low backdrive capability, allowing it to withstand external forces and prevent unintended reverse motion of drive components, suitable for continuous operation. Deutsche Bank mentions NEURA has agreed to use Schaeffler’s actuators on its humanoid robots, and seemingly other customers are already using (at least some components) or will use them in the future.
  • Hyundai Mobis also announced it will supply actuators for Boston Dynamics’ Atlas, aiming for robots to be manufactured via the automotive-scale supply chain.

When the supply chain starts to “automobilize,” what gets priced first is not the concept, but the penetration and scale manufacturing capability of key parts.

2. Onboard Chip Landscape: Nvidia Is Still the Top Choice, but Divergence Is Emerging

Deutsche Bank observes that Nvidia still dominates humanoid robot onboard processors, 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, etc.

By contrast:

  • Tesla and XPeng use self-developed inference chips.
  • At CES, Qualcomm launched its next-gen “full-stack architecture” for robots (Dragonwing IQ10 Series), but Deutsche Bank says it’s unclear if it will see mass adoption; meanwhile, VinMotion’s Motion 2 humanoid robot uses the IQ9 Series, while IQ10 is initially for industrial AMRs and more advanced full-size humanoids.

3. “Physical AI” Moves from Scripts to Agentic: VLA Takes Center Stage

One of the most notable paradigm shifts onsite is moving from “pre-programmed/scripted actions” to vision-language-action (VLA), enabling robots to “reason” to complete tasks.

  • Boston Dynamics replaced its traditional MPC (Model Predictive Control) with Google DeepMind Gemini Robotics’ VLA model, enabling Atlas to understand environments it hasn’t seen before (such as unstructured, chaotic factory scenes).
  • Its motion execution is supplemented by TRI’s large behavior model (LBM), similar to Figure’s Helix dual-system model: System 1 for high-frequency rapid response, System 2 for low-frequency higher-order reasoning and language; Deutsche Bank also points out Figure seems to be developing two proprietary models.

4. Training Wars Upgrade: Real-World Data and Simulation “Closed Loop” Are Key

Deutsche Bank judges the industry debate has shifted from “simulation vs. real world, which is better” to “how to efficiently close the loop.”

  • NEURA is taking a more “physics-first” approach, building the NEURA Gym large-scale physical training center, arguing that simulation is only “approximate” and loses fidelity in complex contact tasks (e.g., “threading a needle”); it collects high-fidelity data from hundreds of robots doing real-world sorting and assembly, then inputs this into “Neuraverse” to generate synthetic twins of real fails to train in simulation, finally pushing fixes back onto real robots.
  • Another company noted it’s impossible to simulate how objects “feel,” requiring humans to demonstrate first: with remote operation, people wear VR suits to control humanoid robots to perform actions like “picking up grapes”; then, with a small number of “perfect demonstrations,” they generate 100,000+ action variants in simulation using NVIDIA GR00T-Mimic, and use reinforcement learning to further smooth the actions.
  • By contrast, Mobileye stresses its Mentee will be mainly trained in simulation.

5. “General Purpose” Gives Way to “Specific Roles”: Commercial Proof Comes First

Deutsche Bank believes that in the short term, “general purpose humanoid robots” will mostly be deployed in specific scenarios to prove commercial feasibility before moving into homes.

  • Keenon Robotics (China): Already owns 40% of the global service robot market, with about 100,000 units exported overseas; product prices range from below $10,000 to around $100,000, focusing on task customization. At CES 2026, its flagship humanoid robot XMAN-R1 can pop popcorn, pour drinks, and interact with human-like gestures; its “Brain” uses Keenon Operator Model 2.0, a VLA model for service industries, able to understand instructions like “find the guest at table 4 and give him candy.” Keenon also mentioned building a collaborative ecosystem at the Shangri-La Commercial Hotel in Shanghai: MAN-R1 is the human-machine interaction “front,” W3 delivers items to rooms, S100 moves heavy luggage, C40/C55 clean. In high labor cost markets like Japan, their robots last up to 8 years, significantly above the industry’s typical 3–5 years.
  • Deep Robotics focuses on industrial inspection: measured by coverage distance (up to 63km), can autonomously patrol and monitor hazardous areas like substations, power plants, oil & gas sites 24/7; used for emergency rescue, firefighting, and toxic gas detection, with swappable batteries to reduce charging friction.

6. Cost-Reduction Formula Is Straightforward: Scale is the Key Premise for Lower Costs

On the humanoid robot side, Deutsche Bank attributes cost reduction primarily to: increasing volumes to spread expenses + improved supplier bargaining.

  • One company claims costs have dropped from $200,000 to $100,000 per unit, with plans to reach $50,000 per unit in the next few years, provided thousands of units are sold.
  • Boston Dynamics and Hyundai Motor announced a goal for 30,000 units annual capacity by 2028; and its 2026 output is already fully allocated ahead for Hyundai’s automobile plants. The company also points out actuators account for about 60% of the BoM, and Hyundai Mobis, an internal Hyundai supplier, will make them to help scale up production.
  • Mobileye disclosed, in the context of acquiring Mentee: if annual output is 50,000 units, a simplified design (without tendon-drive system) costs about $20,000/unit to make; if annual output is 100,000 units, cost drops by half to $10,000/unit; and at 100,000 units, cost can halve again to $5,000/unit, targeting a production ramp by 2028, with manufacturing handled by Aumovio.

7. Robotaxi Momentum Is Stacking Up: 2026 Looks More Like a “Year of Commercial Acceleration”

Deutsche Bank predicts that as Tesla launches Robotaxi in 2025, multiple players will accelerate commercialization in 2026, with Waymo and Zoox’s strong presence at CES a clear signal:

  • Waymo: Over 10 million paid rides since founding; latest disclosure shows 450,000 paid rides per week as of December 2025, expanding to Houston, Miami, and markets like Tokyo and London.
  • Amazon’s Zoox: From public testing in Las Vegas to showcasing “market-ready” products for dense urban areas—a “carriage-style” Robotaxi with no traditional cockpit at all.
  • Mobileye & Volkswagen: Will launch special ID. Buzz electric vans for L4-level Robotaxi service in Los Angeles this year.
  • Additionally, partners including Nuro, Lucid, and Uber plan to launch autonomous vehicles based on the Lucid Gravity in the San Francisco Bay Area by late 2026, with further expansion to more cities.

8. Nvidia Alpamayo: “Brain + Skull” Packaged for Carmakers, But Validation Still Underway

Nvidia announced its Alpamayo, the “brain” for autonomous driving, bundled with Thor, the “skull,” aiming to lower the threshold for automakers to deploy advanced capabilities: companies like Lucid and Mercedes need not invest billions from scratch to build AI infrastructure—they can “plug in” Nvidia’s solution directly.

Deutsche Bank remains cautious: this indeed sparks debate over Tesla’s moat, but it’s too early for worry; Nvidia still needs legacy OEMs to deliver on commitments, and whether its models can cover real-world edge cases remains to be seen. Deutsche Bank points out its training data is only a fraction of what Tesla collects.

Even if Alpamayo performs ideally, Deutsche Bank still believes Tesla, with its vertical integration (vehicle, chips, AI infrastructure, networking), has a structural cost advantage; if autonomous driving/Robotaxi commoditizes, cost will be the biggest differentiation point.

9. Aptiv: End-to-End AI ADAS + Connectivity and Software Platform Means “Cross-Industry”

Aptiv’s showcase centers on the next-generation end-to-end (E2E) AI-driven ADAS platform: using newly released Gen 8 radar and PULSE sensors to deliver “human-like logic” hands-free driving (L2++) in complex urban environments.

On the software side, it launched the cloud-native LINC middleware platform co-developed with Wind River, enabling true software-defined vehicles via 5G and C-V2X; partners with Verizon to demonstrate vehicles “seeing” pedestrians/cyclists around corners by sharing real-time data. Aptiv also emphasizes expanding sensors into aerospace and collaborative robotics—Deutsche Bank says this is the “New Aptiv” narrative for seeking a valuation re-rating.

10. Visteon: 700 TOPS Domain Control, Plug-in Upgrades, Focused on “Execution”

Visteon released its SmartCore HPC domain controller at CES, with computing power up to 700 TOPS, integrating up to 14 cameras and multiple high-speed data channels into a single “central brain.” Expanded partnership with Mahindra to launch SmartCore Pro (triple screens + 360-degree view) for the upcoming XUV7X0.

To address the “installed base platform” constraint, Visteon introduced its AI-ADAS Compute Module plug-in solution powered by Nvidia DRIVE AGX Orin, allowing automakers to add AI assistants and safety features without fully overhauling their architecture; Deutsche Bank notes this product is already on China’s Zeekr vehicles.

Additionally, Visteon launched “Entry Cockpit” for screens under 7 inches, bringing smartphone casting and digital navigation to two-wheelers and entry-level models. Deutsche Bank assesses that its “surgical” vertical integration helps its cost competitiveness and further expands in previously underpenetrated marques (especially Asian OEMs).

In Deutsche Bank’s view, the message from CES 2026 is clear: Autonomous driving and humanoid robots are shifting from “can it be done” to “can it be scaled, can the cost be brought down.”

When Boston Dynamics states actuators make up about 60% of cost and allocates 2026 production ahead of time, the industry is already pricing using manufacturing language; while Waymo’s 10m+ paid rides and its pace of 450,000/week are pushing Robotaxi from concept to hard operational metrics. For investors, the next phase to track isn’t flashier demos, but supply chain tie-ins, capacity ramp-ups, and unit cost curves.

 

~~~~~~~~~~~~~~~~~~~~~~~~

The above excellent content is from Windchase Trading Desk.

For more detailed interpretation, including real-time analysis and frontline research, please join [Windchase Trading Desk ▪ Annual Membership]

Risk Warning and DisclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article fit their particular situation. Invest at your own risk.