Has the Automotive Industry's 'OpenClaw Moment' Arrived?

No AI, No Game: The LLM Arms Race at the Auto Show

After walking through more than a dozen exhibition halls at the 2026 Beijing Auto Show, one signal is unmistakably clear — if your new car doesn't come equipped with a large language model, you can barely justify holding a press conference.

Volcano Engine announced that Doubao is now deployed in over 7 million vehicles; Tencent launched its open platform for mobility-focused intelligent agents across all scenarios; iFlytek unveiled its Spark Intelligent Cockpit; MiniMax showcased its on-device Agent framework EmbodiedClaw; even the new-generation Mercedes-Benz S-Class has a multimodal on-device vision-language model (VLM) tucked into the rear seats. Huawei's HarmonySpace 6 cockpit, BMW's jointly customized AI model with Alibaba — everywhere you look, the entire auto show is permeated with an urgency that screams 'No AI, No Game.'

This arms race to put large models in cars has arrived more fiercely than anyone anticipated.

The Awkward Reality: Can Chat, Can't Control

However, if you actually sit inside these cars and try them one by one, you'll discover a somewhat embarrassing truth: the vast majority of so-called 'AI cockpits' are essentially just smarter, more conversational voice assistants.

They've certainly gotten cleverer. They can plan sightseeing itineraries, recommend trendy restaurants, and with large model capabilities, they can chat with you about a wide range of topics, delivering plenty of emotional value. But when it comes to actual 'vehicle control' — such as automatically adjusting suspension based on driving conditions, coordinating air conditioning with seat settings, or proactively responding to driver intent — capabilities remain severely lacking.

It's like the early days of smartphones: everyone had installed touchscreens, but the app ecosystem hadn't truly been established yet. AI large models have given the cockpit the ability to 'understand language,' but they haven't yet bridged the last mile to 'understanding the vehicle.'

What Is the Automotive 'OpenClaw Moment'?

The concept of 'OpenClaw' originates from the field of embodied intelligence — when AI moves beyond merely 'talking' to truly 'doing,' a qualitative transformation occurs. For the automotive industry, this moment means: AI is no longer just a chat companion in the cockpit, but a genuine 'intelligent agent' capable of perceiving the environment, understanding intent, and proactively controlling various vehicle subsystems.

Imagine this scenario: you're driving onto a bumpy mountain road, and the AI automatically identifies the road conditions and stiffens the suspension; it detects you yawning, and the ventilated seats switch on while the music changes to an energizing beat; you say 'get ready for camping,' and the vehicle automatically switches to off-road mode, adjusts the lighting, and activates the trunk power outlet.

That is true AI vehicle control — evolving from 'conversational interaction' to 'embodied control.'

Three Critical Bottlenecks to Overcome

The automotive 'OpenClaw Moment' has been slow to arrive, held back by three core bottlenecks:

First, insufficient openness of underlying vehicle interfaces. For large models to control a vehicle, they need access to APIs for core systems such as the chassis, powertrain, and thermal management. But for safety reasons, most automakers take a conservative approach to these interfaces, leaving AI with extremely limited access permissions.

Second, the tension between on-device computing power and model capability. True AI vehicle control requires multimodal perception and real-time decision-making, which places extremely high demands on in-vehicle chip computing power. Although on-device large models are evolving rapidly, finding the right balance among power consumption, latency, and reliability remains an engineering challenge.

Third, unclear safety and liability boundaries. When AI proactively controls vehicle functions, how should liability be assigned if an error occurs? This is not just a technical issue — it's a regulatory and ethical one as well.

Who Is Most Likely to Break Through First?

Based on current strategies, several approaches are being pursued simultaneously.

Huawei's HarmonyOS cockpit, with its full-stack in-house R&D advantage, has inherent strengths in connecting the 'perception — decision — control' pipeline; MiniMax's EmbodiedClaw framework represents the technical direction of on-device agents, attempting to let large models directly drive physical-world operations at the vehicle edge; and traditional luxury brands like Mercedes-Benz, by deploying on-device VLMs in their flagship models, are sending a clear signal: the premium market is willing to pay for genuine AI vehicle control capabilities.

Notably, advances in embodied intelligence are accelerating their penetration into the automotive industry. When robots can already grasp and manipulate objects in complex environments, having AI control a highly structured vehicle is theoretically not out of reach.

Outlook: From 'AI in the Car' to 'AI Driving the Car'

The 2026 Beijing Auto Show may well be looked back upon as a pivotal milestone — large models in vehicles have completed the leap from concept to standard feature, but the true value inflection point still lies ahead.

When the industry resolves the three hurdles of interface openness, on-device computing power, and safety boundaries, the automotive 'OpenClaw Moment' will truly arrive. At that point, AI will no longer be just the chatty assistant in your car, but a 'digital driving partner' that understands you, understands the vehicle, and understands the road.

That day is not far off. But until then, automakers might need fewer large-model slogans at press conferences and more real work on underlying vehicle architecture. After all, AI that can chat is everywhere — AI that can control the car is the real competitive moat.