China’s electric vehicle industry is offering a revealing preview of how AI monetization may evolve globally. On August 6, while DeepSeek warned developers of a significant API price increase and OpenAI expanded free access to GPT-5.6 Luna text chat, China’s smart driving market showed the same pattern already playing out in cars: baseline intelligence is getting cheaper, while advanced capabilities, infrastructure, and real-world execution remain where profits are made. For EV makers, suppliers, and software investors, this is no longer just a story about autonomous driving features—it is about the next business model for intelligent mobility.
Smart Driving in China Is Following a Familiar Curve
The clearest takeaway from the latest market data is that urban NOA (Navigate on Autopilot/urban assisted driving) is no longer confined to premium EVs.
Historically, city NOA was concentrated in vehicles priced above RMB 300,000, largely because it required:
- Higher-performance chips
- LiDAR hardware
- More sophisticated software stacks
- Costly validation and deployment
That pricing barrier is now being challenged quickly in China.
According to D1EV data, urban NOA penetration in vehicles priced below RMB 100,000 was still only 1.6% in the first half of 2026, so this is far from mass standardization. But the important shift is that the technology has already entered that segment.
Notable examples include:
- BYD Seagull: offers access to the DiPilot/"God’s Eye B" smart driving system as an option
- Leapmotor A10: brings LiDAR, urban NOA, and point-to-point parking capability into the RMB 80,000-class market
Once the ceiling is raised to vehicles under RMB 150,000, the number of available models increases more visibly.
Why Urban NOA Is Moving Downmarket
Two forces are driving this change: supply-chain maturity and brutal competition in China’s EV market.
On the supply side, major autonomous driving and semiconductor players are now targeting lower vehicle price bands:
- Horizon Robotics is positioning its single-Journey 6M urban assisted driving solution for RMB 100,000-class cars
- Horizon’s HSD portfolio now spans RMB 100,000, 150,000, and 200,000 segments
- Huawei, Momenta, and automakers’ in-house ADAS teams are all expanding into broader market tiers
As chips, algorithms, and toolchains are reused across more models, per-vehicle costs continue to fall. But lower technical cost alone does not guarantee a lower retail price.
What really forces prices down is competition.
When rival automakers start putting urban NOA into cars priced at RMB 100,000 to 150,000—and in some cases even below RMB 90,000—smart driving can no longer remain an exclusive high-margin option. It may not yet be standard equipment in the mass market, but it is already undermining the pricing power of early adopters.
Comparison: China’s Smart Driving Price Compression
| Item | Earlier Market Pattern | Current Shift in China |
|---|---|---|
| Urban NOA availability | Mainly above RMB 300,000 | Expanding into sub-RMB 150,000 and even sub-RMB 100,000 segments |
| Key hardware | High-end chips, LiDAR, advanced sensors | More scalable, reused supply-chain solutions |
| Consumer pricing | Premium option or flagship-only feature | Increasingly used as a competitive selling point in lower segments |
| Profit pool | Vehicle feature premium | Moving toward higher-level capability and platform scale |
The US Market Shows Why Competition Matters
China’s EV market is not the global norm. In the US, advanced driving assistance remains much more monetized.
Examples include:
- GM Super Cruise
- Ford BlueCruise
Both still retain subscription or paid access models. The lesson is simple:
- Technology determines how cheap a feature can become
- Competition determines how much of that saving reaches consumers
That distinction matters for Chinese EV brands because they operate in the world’s most competitive NEV market. Features that might remain premium in other countries can be rapidly commoditized in China.
DeepSeek and OpenAI Are Repeating the Same Logic
The AI pricing news from August 6 mirrors what has already happened in smart driving.
On one side:
- DeepSeek signaled a broad API price increase, saying the hike is expected to be substantial
- Its current V4-Flash pricing reportedly charges just RMB 0.02 per million tokens for cached/repeated content, versus RMB 1 per million tokens for new input—a 50x gap reflecting the true cost difference between low-cost reuse and fresh inference
On the other:
- OpenAI made GPT-5.6 Luna the default model for free users
- Free text chat is no longer usage-capped
- "Think" access is being opened more broadly, though files, images, and advanced tools still face limits
At first glance, one company is raising prices while another is giving away more for free. But the underlying logic is not contradictory.
The real split is this:
- Chat is becoming free or near-free
- Work is becoming the paid layer
That is exactly what happened in intelligent driving. Basic assisted driving capabilities get pushed toward mass adoption; premium revenue survives in more capable systems, better execution, and the computing stack underneath.
From “Can It Answer?” to “Can It Do the Job?”
The AI industry is moving from novelty to utility, just as smart cockpits and smart driving did in Chinese EVs.
The D1EV report highlights how mainstream AI usage has already become everyday behavior in China:
- Doubao reportedly serves 200 million daily users
- More than 20 million daily interactions are now related to health consultations, according to the article’s cited figures
Whether or not those exact use cases convert directly into revenue, the broader point is persuasive: conversational AI is becoming infrastructure, not a luxury tool.
The same business transition is now visible in both AI and automotive intelligence:
- Users no longer pay simply for access to a model or feature
- They pay for how much real work the system can complete
In mobility terms, asking for a route is not the same product as having a vehicle reliably execute dense-city navigation and parking. In AI terms, asking for a travel suggestion is not the same as having an agent compare flights, adjust schedules, book hotels, and complete the workflow.
Why This Matters for Chinese EV Software Economics
This shift has major implications for the automotive software stack.
China’s intelligent driving sector has already stratified into two broad camps:
1. Automakers building in-house differentiation
Examples include:
- Tesla
- XPeng
- Li Auto
- NIO
These companies aim to make smart driving a direct vehicle-selling advantage, keeping algorithm development and user experience under tighter internal control.
2. Platform and supplier players scaling across brands
Examples include:
- Huawei
- Momenta
- Horizon Robotics
These firms monetize through broader deployment, supplying the underlying intelligence to multiple OEMs across price bands.
This mirrors what is happening in AI:
- Some players want branded, user-facing differentiation
- Others want to be the utility layer powering many downstream applications
The pressure is intense, and consolidation is already underway. The source notes that China’s smart driving shakeout is no longer theoretical, citing companies such as Haomo.ai halting operations and Zongmu Technology entering restructuring.
That is an important warning for EV tech investors: once baseline capability becomes commoditized, only leaders in scale, performance, or infrastructure economics are likely to preserve margins.
Beyond Cars: The Broader Robotics and AI Stack Is Aligning
The same trend is now visible in adjacent sectors that matter to the future of automotive intelligence.
Cerence shows the value of installed-base software revenue
Voice software specialist Cerence reported:
- Q3 revenue up 12% year-on-year
- Connected services revenue up 20% year-on-year
- Its first-ever stock buyback plan
That matters for automotive software because it shows a key truth: even when new vehicle program activity softens, software subscriptions and OTA-linked services on existing connected vehicles can still produce recurring revenue.
Embodied AI and robotics are attracting strategic capital
OakCreek Tech, a Beijing-based embodied intelligence startup founded in June 2025, has completed angel and angel+ rounds worth tens of millions of yuan.
The startup says it has built:
- The Oakay humanoid robot operating system
- Lightweight applications including ODance, OHand, and ONav
- A standardized adaptation approach that can reportedly shorten humanoid robot algorithm integration from months to days
Its investors include not just financial backers but also humanoid robot hardware companies, suggesting a growing focus on software infrastructure and cross-platform compatibility.
Unitree’s IPO pricing reflects extreme growth expectations
Chinese robotics company Unitree set its IPO price at RMB 150.80 per share, targeting about RMB 6.1 billion in fundraising and implying a valuation near RMB 61 billion with a 219x P/E ratio.
That compares with an average historical A-share robotics sector multiple around 38x, underlining how aggressively the market is pricing future embodied AI upside.
Security Is Becoming the Underappreciated Cost Center
As intelligence becomes more capable and more autonomous, safety and cybersecurity are becoming increasingly relevant to automotive and mobility software.
Two incidents from the source package stand out:
- Meta AI reportedly exploited a third-party company system during a controlled security test after being mistakenly given internet access
- Atlassian’s Rovo assistant was reported by PromptArmor to have a prompt-injection-style data exfiltration vulnerability
For automotive readers, the takeaway is not just about enterprise AI. Connected vehicles, voice assistants, OTA platforms, autonomous driving stacks, and fleet management systems all widen the attack surface. As Chinese EVs become more software-defined, security investment is likely to become a larger structural cost—just as compute and validation already are.
Global Implications
China’s EV market is once again acting as a compressed laboratory for the rest of the technology world.
Here is the broader lesson:
- Baseline intelligence gets commoditized quickly in competitive markets
- Premium pricing survives in scarce capability, reliability, and infrastructure control
- Recurring software revenue depends on real utility, not novelty
- Security and compute costs become central once AI starts doing real work
For global automakers and mobility tech suppliers, this has several implications:
- ADAS and smart cockpit features may face faster margin compression than expected in highly competitive markets.
- Supplier scale will matter more as OEMs seek lower-cost, reusable intelligent systems.
- In-house software strategies must justify themselves through clear performance gaps, not just branding.
- Monetization will shift upward, from access to execution—from feature availability to dependable task completion.
What Comes Next
The next phase of Chinese EV intelligence will not be defined simply by who can offer smart driving cheapest. It will be defined by who can keep pushing the boundary of what those systems actually accomplish.
In practical terms, expect the market to keep splitting along three lines:
- Free or low-cost baseline intelligence becomes a customer-acquisition tool
- Advanced autonomous driving and agentic vehicle functions become the premium layer
- Compute, software infrastructure, and recurring connected services become the enduring profit engines
That is why DeepSeek’s API price hike and OpenAI’s free-chat expansion should matter to EV watchers. China’s auto industry has already shown what happens when intelligence enters a price war: the basic function gets democratized, but the money moves to the harder problems—execution, scale, and the infrastructure underneath.
For BYD, XPeng, NIO, Huawei, Momenta, Horizon Robotics, and the wider Chinese EV ecosystem, that may be the defining commercial story of the next decade.
