Chinese EV makers' AI feature war

The most striking signal in this story isn't Chinese OEMs adopting Chinese AI — it's foreign luxury brands doing the same to stay competitive in China. Mercedes-Benz expanded its LLM cooperation with ByteDance in August 2024 to make virtual assistants 'smarter and more knowledgeable.' SAIC Audi (E7X) and SAIC Volkswagen (ID series) followed by licensing Doubao outright, and SAIC Volkswagen now appears among the nine automakers integrating Alibaba's Qwen. SCMP also reported that Tesla, Volvo and Mercedes-Benz won approval to deploy in-vehicle AI assistants in China, with Tesla reportedly preparing to integrate Doubao around April 22, 2026.

For decades the cockpit-software stack was a place where premium German brands could project quality. In China, that layer has effectively been outsourced to ByteDance and Alibaba. The implication: Chinese tech platforms now sit between global automakers and their highest-margin customers, holding the API for the experience that defines a 'modern' car.

AlixPartners' Stephen Dyer puts the math bluntly — features 'disseminate so quickly' that automakers 'have to race and keep racing.' The mechanism is structural: when ByteDance and Alibaba license Doubao and Qwen to dozens of OEMs through Volcano Engine and Alibaba Cloud, the underlying capability is the same component used by competitors.

AlixPartners notes that the top 20 best-selling EVs above 100,000 yuan already offer near-identical driver-assist and infotainment feature sets. BYD's decision to push Doubao across all five sub-brands — Yangwang, Denza, Fangchengbao, Dynasty and Ocean — confirms this; large-model cockpits are now baseline configuration, not a premium upcharge. Pair that with an 11% drop in average vehicle prices from CN¥217,000 to CN¥194,000 between 2023 and 2025 (roughly CN¥471 billion of industry revenue erased) and you get a feature war that cannot do what the price war could not: rebuild margin.

Two players now functionally control the cockpit operating system for the Chinese auto industry. ByteDance's Volcano Engine has scale — 145 models, 7M+ vehicles, 50+ brands, plus a halo from Doubao's 155M weekly active users — and is running roughly 30B-parameter on-device models with deeply customized cabin solutions (Seres alone invested 100M+ yuan in this work).

Alibaba's counter is architectural: a Qwen-Omni multimodal model on NVIDIA DRIVE AGX Thor for environmental perception, paired with cloud-side agentic AI handling multi-step workflows like booking hotels, buying attraction tickets and tracking parcels. The on-device parameter gap (ByteDance's ~30B vs Qwen's ~4B) hints at divergent bets: ByteDance leans into local intelligence and customization, while Alibaba leans into cloud agents that can transact. Either way, OEM bargaining power is shrinking — the carmakers risk becoming distribution channels for someone else's AI.

Sino Auto Insights' Tu Le argues AI 'should run in the background to support the user experience, not necessarily be a feature' — a pointed critique of the marketing-first cockpit cycle. That view has unexpected support outside the analyst class. On Reddit, a recurring counter-narrative dismisses AI cockpits as bolt-on novelty; one Porsche Macan owner's reaction — 'jam packing everything with pointless AI features is what everyone wants these days. It gets the people going' — captures the skeptical mood.

Meanwhile X.com chatter is more bullish, treating Doubao/Qwen integration as inevitable infrastructure, but even there the framing is utility, not delight. The thread connecting AlixPartners, Tu Le and the contrarian Reddit voices is the same: when a feature exists in 145 models within months, it is no longer a feature, it is plumbing — and plumbing does not justify a price premium.

If AI cockpits become plumbing, where does the next moat live? AlixPartners' Dyer expects the battleground to shift to 'outside-of-car' lifestyle experiences — the luxury-brand playbook of ecosystems, services and ownership rituals that cannot be cloned by licensing a model. MERICS analysts argue Chinese OEMs are already advantaged in the software-and-UX layer because 'there is less differentiation in motors and technical specs,' and YouTube coverage from Voyah/Dongfeng's manufacturing base showed AI-equipped vehicles rolling off the line every 118 seconds — AI absorbed into production, not just product.

The harder question is whether European incumbents, who historically anchored their identity in driving dynamics and engineering, can pivot to lifestyle-as-product fast enough. AlixPartners' projection that fewer than 10 of ~129 Chinese EV brands will be profitable by 2030 suggests the consolidation pressure will be brutal even for the home team — a reminder that winning the AI feature war is not the same as winning the auto industry.