Physical AI News (Feb 26, 2026 Issue)
Update Date: 2026/2/26
Executive Summary
On February 25, 2026, Wayve raised $1.2 billion from NVIDIA, Uber, and others to commercialize camera-centric, end-to-end autonomous driving that does not rely on HD maps or LiDAR. In Japan, SCSK, Net One Systems, and TechShare are mass-producing synthetic data using Isaac Sim and Cosmos to reduce the deployment time and cost of robots for non-routine tasks such as bin picking. Volkswagen's adoption of XPENG's second-generation VLA marks a shift toward horizontal specialization. Pegasus and AISIN have expanded their CVC fund to $100 million. Unitree has announced the high-performance quadruped As2.
1️⃣ Wayve raises $1.2 billion from NVIDIA and Uber: Moving toward commercialization of end-to-end autonomous driving
🔗 Source: The Tech Buzz — Wayve Raises $1.2B From Nvidia, Uber in Massive AV Round
UK-based autonomous driving startup Wayve has raised a total of $1.2 billion from NVIDIA, Uber, and three undisclosed automakers. The company employs an Embodied AI model that does not rely on HD maps or expensive LiDAR, instead using camera-based visual perception and deep learning trained on massive amounts of real-world driving data to adapt to unknown environments. Integration with Uber's ride-sharing network and the NVIDIA DRIVE platform is expected to accelerate global commercial deployment. As the largest AV funding round of 2026, this move symbolizes the industry's shift toward AI-native approaches.
2️⃣ SCSK, Net One Systems, and TechShare launch Physical AI collaboration utilizing NVIDIA technology (Japan)
🔗 Source: SCSK Press Release / PR TIMES
SCSK, Net One Systems, and TechShare have announced a collaboration to integrate NVIDIA Isaac Sim and Cosmos to generate large volumes of high-quality synthetic data, aiming for the social implementation of Physical AI that automates non-routine tasks such as bin picking. The roles are divided as follows: SCSK handles synthetic data generation, Net One Systems builds imitation learning models on advanced AI infrastructure, and TechShare conducts operational verification and accuracy feedback on actual robots. By significantly substituting the cost of collecting learning data in real-world environments with synthetic data generated in virtual space, the companies aim to reduce the deployment time and cost of robots in the domestic enterprise market. They plan to commercialize the package within fiscal year 2026, with an eye toward expanding from manufacturing and logistics to medical and home care fields.
3️⃣ XPENG mass-produces 2nd-generation VLA model: Volkswagen as launch customer
🔗 Source: Gasgoo AutoNews — He Xiaopeng Internal Memo
An internal memo from XPENG's CEO reveals that the company's second-generation VLA (Vision-Language-Action) model has reached an emergent level and will be officially deployed in Q1 2026. Global giant VW has decided to adopt it as a launch customer, establishing a horizontal specialization model where legacy OEMs outsource autonomous driving technology. XPENG is fundamentally transforming its revenue structure from an EV manufacturer to an "Embodied AI company."
4️⃣ Pegasus Tech Ventures and AISIN expand CVC fund to $100 million
🔗 Source: Business Wire — Pegasus Tech Ventures and AISIN Expand CVC Fund to $100 Million
Silicon Valley-based Pegasus Tech Ventures and Japan's AISIN have extended their jointly managed CVC fund until 2036 and increased the total to $100 million. The fund focuses on Physical AI, next-generation robotics, energy, and health tech. This Japanese-American private VC partnership, which has invested in over 40 companies since its establishment in 2018, has entered its second phase, and startup investment in areas beyond the automotive industry—such as humanoids, space, and next-generation energy—is expected to accelerate.
5️⃣ Unitree Robotics announces high-performance quadruped robot "As2"
🔗 Source: Phemex News — Unitree Robotics Launches Advanced Quadruped Robot As2
China's Unitree has announced the compact and high-spec As2. It features a maximum torque of 90 N·m, over 4 hours of operation without load, a 15kg payload, a range of over 13km, and IP54 dust and water resistance. Designed specifically for industrial and outdoor use, the specifications are intended to accelerate practical application in a wide range of field operations, such as infrastructure inspection and transport support.
Comprehensive Analysis
February 25, 2026, marks the shift of Physical AI into a competition where success depends not only on "model performance" but also on "data acquisition and deployment channels." Wayve has presented a framework that connects a ride-hailing network (Uber) with a computing infrastructure (NVIDIA) to create a cycle of learning through actual operation. Domestic collaborations are aiming for widespread adoption by lowering learning costs through synthetic data and packaging non-routine tasks. VW's external procurement shakes the OEM's self-sufficiency model, and in terms of funding, CVCs are strengthening diversified investments in robots and energy. The next focus will be on quality assurance and safety evaluation for Sim2Real, as well as the design of revenue models for continuous operation.
Future Points of Interest
The key to the Wayve x Uber x NVIDIA integration is whether they can create a cycle of "driving data acquisition -> learning -> deployment via ride-hailing," and I will be watching the design of regulatory evaluation and liability boundaries.
The deciding factor for Isaac Sim + Cosmos synthetic data will be whether it can reproduce even exceptional on-site behaviors, with the speed of the Sim2Real verification loop and commercial pricing being the main points.
With VW's adoption of XPENG's VLA, the horizontal specialization of "buying models" is accelerating, making license terms and the scope of data sharing new axes of competition.
The expansion of the Pegasus x AISIN CVC suggests a shift of funds outside of automotive applications, and I want to track the speed of commercialization to see if the investees can reach mass production and implementation.
While the Unitree As2 could boost on-site application for inspection and transport, its price range, maintenance system, and SDK maturity could become bottlenecks for widespread adoption.
