China EVs: SDV and Physical AI Reshape Mobility

China’s smart EV industry is moving beyond electrification and into a new phase defined by software, data, and robotics. Fresh reports from D1EV highlight three connected developments: China’s passenger-car L2 penetration has reached 64% in 2025, RTI is positioning its DDS-based Connext Drive as a core communications layer for software-defined vehicles (SDVs), and JD.com is launching an ambitious physical AI and robotics push that could influence the wider intelligent mobility supply chain. At the same time, startup Westlake Robotics is accelerating embodied AI commercialization in Hangzhou, underlining how China’s auto and robotics ecosystems are increasingly converging.

China’s EV Market Is Now a Software-Defined Vehicle Battleground

China remains the world’s largest auto market, but its real significance in 2025 is that it has become the fastest-moving test bed for software-defined vehicles. According to the D1EV source, L2-capable passenger vehicles now account for 64% of the market in China, showing just how quickly driver-assistance and intelligent cockpit features have become mainstream.

That shift matters because the competitive focus in China’s EV market is no longer just range, battery chemistry, or charging speed. Increasingly, automakers are competing on:

• Advanced driver assistance systems (ADAS)
• Urban NOA capabilities
• End-to-end AI models
• Intelligent cockpit software
• Over-the-air software iteration speed
• Cross-domain computing architecture

In other words, the center of value is moving from hardware-defined cars to software-defined cars. And once a vehicle becomes a rolling compute platform, the communications framework underneath it becomes strategically important.

Why In-Vehicle Communication Architecture Matters More Than Ever

Traditional vehicle electronics relied heavily on signal-based, point-to-point communication. That model struggles when a car must continuously move data between sensors, zonal controllers, domain controllers, high-performance computing units, cloud services, infotainment systems, and vehicle control functions.

This is where the D1EV report places focus on DDS (Data Distribution Service) and RTI’s Connext Drive platform. The argument is straightforward: if Chinese automakers want to support heterogeneous chips, operating systems, and software stacks while still delivering functional safety and cybersecurity, they need a more flexible, data-centric middleware layer.

What Connext Drive Is Designed to Do

RTI describes Connext Drive as a communications backbone spanning the full SDV stack, from chip to cloud. In practice, that means it is intended to connect:

• Sensors and actuators through zonal gateways
• High-performance computing modules for ADAS and vehicle control
• Infotainment and digital cockpit systems
• Telematics and cloud applications
• Mixed software environments across the vehicle

The key technical appeal is interoperability. According to the source, Connext Drive supports:

• Native DDS deployment
• Integration with AUTOSAR Classic
• Integration with AUTOSAR Adaptive
• Interoperability with ROS 2
• Compatibility with processor cores including Arm and TriCore
• Support for operating systems such as Android, AGL, QNX, and VxWorks
• Integration with AUTOSAR implementations from Elektrobit, ETAS, and Vector

For China’s carmakers, that matters because many are building increasingly mixed software stacks while also working with a more localized supply chain that includes domestic chips, operating systems, and Tier 1 suppliers.

RTI’s Pitch to Chinese Automakers: Flexibility Plus Safety

The Chinese EV market rewards speed, but mass production still demands automotive-grade validation. That combination is difficult: companies want internet-style iteration cycles, yet vehicles must meet rigorous functional safety standards.

This is where RTI is trying to differentiate Connext Drive.

Reported strengths highlighted in the source

• ISO 26262 ASIL D-certified development process, a major signal for safety-critical automotive deployment
• Support for containers and virtual machines, helping teams adopt microservices-style development
• Better alignment with CI/CD software workflows, which are becoming more relevant as vehicle software grows in complexity
• Licensing and deployment flexibility across the supply chain
• A single framework that can span prototyping, validation, and production

The commercial proof points cited by D1EV are also notable:

RTI’s broader deployment in autonomous systems such as air taxis, unmanned vessels, underwater robots, and defense platforms adds credibility to the claim that this is mature middleware rather than a concept-stage software layer.

For China’s EV players, the value proposition is clear: reduce integration risk, shorten safety-related development cycles, and keep room for fast feature iteration.

JD.com’s Physical AI Plan Shows Where Mobility Is Heading Next

While RTI’s story is about the software backbone inside vehicles, JD.com’s newly announced Physical AI Acceleration Plan points to the next layer of competition: real-world AI deployment at scale.

JD says it aims to build a global physical-world operating platform by linking data, models, hardware, and industrial scenarios. Although this is not an automotive announcement in the narrow sense, it is highly relevant to intelligent mobility because logistics robots, autonomous delivery vehicles, drones, and embodied AI increasingly share core technologies with next-generation EV platforms.

JD’s headline targets are striking

• Build what it calls the world’s largest embodied-intelligence data collection center
• Collect more than 10 million hours of real-world human-scene video data within two years
• Deploy 80+ RoboBase robot industry bases across China over five years
• Help 100 robot OEMs reduce costs across full product categories within three years
• Help 100 components suppliers double performance
• Invest RMB 10 billion-level resources by 2028 to help 100 brands surpass RMB 1 billion in standalone sales
• Purchase 3 million robots, 1 million autonomous vehicles, and 100,000 drones within five years for logistics applications
• Build after-sales service coverage in 100+ countries
• Create 100,000+ robot service engineer jobs

These numbers are ambitious, but the strategic logic is sound. China’s competitive edge increasingly comes from combining hardware manufacturing scale with deployment scale. In EVs, that has meant rapid vehicle launches and software updates. In robotics and embodied AI, it could mean faster data collection, faster model training, and faster iteration in real-world environments.

Westlake Robotics Adds Momentum to China’s Embodied AI Push

Another D1EV report shows that embodied AI is not just a big-company story. Westlake Robotics, based in Hangzhou and originating from Westlake University’s AI and robotics research, has completed an A1 funding round after finishing its A round in August.

The company says the new funding will support three priorities:

• Development of its general embodied foundation model WR1
• Expansion into more commercial application scenarios
• Building a stronger talent base in embodied intelligence

Westlake Robotics follows a full-stack path built around what it describes as:

• A general brain
• A humanoid whole-body cerebellum
• A self-developed humanoid body

The company says it has already developed two core technologies:

• The GAE general motion model
• A dual pre-training architecture for large and small robot “brains”

Its humanoid robot, Westlake o1, is designed to move beyond pre-programmed, single-task robotics. According to the report, the system can use motion-capture equipment to replicate full-body human movement in real time without writing action scripts, while dynamically correcting posture and executing long-sequence, multi-step whole-body tasks.

That is especially interesting for the automotive and mobility sectors because the same embodied AI stack can support factory automation, inspection, warehousing, logistics, and data collection for autonomous systems.

Commercial traction is also emerging

Westlake Robotics says it has:

• Completed a full loop from technical validation to pilot deployment and project implementation
• Signed deep partnerships with Alibaba, Intel, and JD.com
• Secured nearly RMB 100 million in orders
• Expanded into scenarios including:
  • Scientific research and education
  • Data collection
  • Power inspection
  • Commercial services

It has also signed a strategic agreement with the Longyou local government to build a county-wide humanoid robot real-world training base, aimed at closing the loop between data collection, model training, and live-machine scenario validation.

How SDVs and Embodied AI Are Starting to Converge

These three stories may look separate at first glance, but together they reveal a broader industrial pattern in China.

Shared trends across EVs and robotics

The overlap is growing because modern EVs, autonomous delivery vehicles, warehouse robots, drones, and humanoids all increasingly depend on:

• High-bandwidth middleware
• Standardized interfaces
• AI model deployment pipelines
• Real-world data feedback loops
• Cloud-to-edge orchestration

China is uniquely positioned here because it has the market size to commercialize these systems faster than most regions, along with a domestic supply chain willing to move quickly.

Why This Matters Globally

For global automakers and suppliers, the takeaway is not simply that China is selling more EVs. It is that China is becoming the proving ground for a broader intelligent-machines industry.

A few implications stand out:

• China’s SDV stack is maturing fast. Middleware, safety-certified software frameworks, domestic chips, and mixed-platform integration are moving from theory to production.
• Robotics and automotive supply chains are beginning to overlap. JD’s plan suggests future scale may come from shared AI infrastructure rather than from vehicles alone.
• The competitive benchmark is shifting. It is no longer enough to build an EV with strong hardware specs; the real differentiator is how quickly a company can collect data, train models, integrate software, and deploy updates safely.
• Standards and interoperability are strategic. Middleware layers like DDS could become more important as OEMs try to avoid lock-in while still scaling complex cross-domain systems.

For Western carmakers watching China, this is a reminder that competition is increasingly about software architecture and industrial execution, not just battery cost or export pricing.

The Road Ahead

The next phase of China’s EV industry will likely be defined by the convergence of software-defined vehicles, AI-native development, and embodied intelligence. RTI’s Connext Drive represents one answer to the architectural challenge inside the car, while JD.com and Westlake Robotics show how physical AI infrastructure outside the car is scaling just as quickly.

That combination could give Chinese mobility players an even bigger advantage: not just the ability to build EVs cheaply and quickly, but the ability to connect vehicles, robots, cloud services, and real-world AI systems into one continuously learning ecosystem.

If that happens, the winners in the Chinese EV market may not simply be the brands with the best battery pack or the flashiest cockpit. They may be the companies that best integrate communication middleware, safety-certified software, AI training loops, and deployment scale across the entire intelligent mobility stack.