Tesla and XPeng delivered two of the clearest signals yet this week that the electric-vehicle battle is no longer just about factories, batteries, or price cuts. On May 7-8, Tesla showed both manufacturing strength and AI ambition: its Shanghai Gigafactory reportedly delivered more than 79,000 vehicles in April, up 36% year on year, while the company simultaneously launched a major US hiring push for data annotation roles tied to Full Self-Driving (FSD) and Optimus. At the same time, XPeng said its second-generation VLA intelligent driving system has crossed a major threshold, with smart-driving mileage now accounting for more than 50% of real-world user driving. Taken together, the news highlights how Chinese EV competition is increasingly being shaped by software adoption, data infrastructure, and real-world autonomous driving usage.
Tesla's Shanghai Factory Keeps Delivering
According to China Passenger Car Association data cited by local media, Tesla's Shanghai Gigafactory delivered more than 79,000 vehicles in April, representing 36% year-on-year growth. That is a notable rebound and reinforces Shanghai's role as one of Tesla's most important export and production hubs.
The factory's influence is extending well beyond mainland China. Tesla is also expanding deliveries of the Model Y L, a six-seat SUV built for Asia-Pacific markets, with shipments now reaching:
- Japan
- South Korea
- Australia
- Thailand
- Hong Kong and Macau
Regional momentum appears strong:
- South Korea: 13,000 Shanghai-built Tesla imports in April, up 1,050% year on year
- Australia: Tesla sales up 145% in April
- Sweden: April registrations up 111%
- Denmark: April registrations up 102%
- France, the UK, and Ireland also posted double-digit to triple-digit gains
Tesla also said its global deliveries exceeded 358,000 units in Q1 2026, while its worldwide owner base has now surpassed 9 million.
Tesla Is Building a Bigger Real-World AI Data Engine
Just as important as Tesla's delivery growth is what the company is doing behind the scenes. Tesla recently posted a significant number of US job openings for data annotation specialists, with roles spread across four US cities and eight related positions in total.
The hiring drive includes both execution-level and management roles, including annotation managers and technical project managers. Some management positions reportedly offer annual salaries of $90,000-$130,000, equivalent to roughly RMB 610,000 to RMB 880,000, before additional cash and stock incentives.
A few details stand out:
- Tesla does not require prior AI experience for some roles
- The company offers on-site training
- Standard roles reportedly follow a 9-to-5 work schedule
- Benefits include healthcare, retirement support, and employee stock plans
Tesla AI engineering director Duan Pengfei reportedly reshared the recruitment post and said the company aims to build the world's largest real-world AI data engine.
That language matters. Data annotation remains a foundational layer for training advanced AI systems, especially in:
- FSD driver-assistance systems
- Computer vision stacks
- Humanoid robotics, including Optimus
Tesla's annotation teams are expected to handle image and video data involving:
- Vehicles
- Pedestrians
- Lane lines
- Road environments
- Robot-related perception tasks
Because Tesla continues to pursue a pure vision approach, annotation quality is especially critical. High-accuracy labeling directly influences how effectively the system learns to interpret the physical world. Tesla's preference for an in-house, offline workflow also reflects two priorities:
- Data security, especially where user privacy and proprietary driving data are involved
- Consistency and speed, allowing annotation tools and workflows to be refined internally
XPeng Hits a Real-World Smart-Driving Milestone
While Tesla is scaling the data pipeline, XPeng is making a case that advanced driver assistance is already moving into daily use. On May 8, XPeng said its second-generation VLA intelligent driving system has been live for one month, and that smart-driving mileage now accounts for more than 50% of users' real driving distance.
According to the company, this is the first time the global intelligent-driving industry has crossed that threshold. Whether rivals frame the milestone the same way or not, the data point is significant because it shifts the conversation away from staged demos and toward actual user behavior.
XPeng's headline metrics for the first month include:
- Smart-driving mileage share: over 50% of real-world driving
- Human takeovers per 100 km: down 25.87% month on month
- Trips completed with 100% smart driving enabled: up 27.84% month on month
- During the May Day holiday, AI-assisted driving daily usage reached 93.21%
- Cumulative intelligent-driving distance reached 84.46 million km
- Longest single smart-driving trip by one vehicle: 5,441 km
- First-week activation rate among new users: 97.43%
- Monthly active rate: 96.97%
- Total usage mileage: up 91.05% month on month
- Total usage time: up 107.86% month on month
What XPeng's VLA 2.0 Actually Changes
XPeng describes its second-generation VLA as a self-developed, production-ready physical world model for intelligent driving. The key technical claim is architectural simplification.
Instead of relying on the traditional vision-language-action pipeline with intermediate translation steps, XPeng says the new system directly converts visual inputs into driving actions in an end-to-end process. The company says this reduces decision latency to under 80 milliseconds.
Other claimed advantages include:
- Reduced reliance on high-definition maps
- Better generalization across varied road conditions
- Broader coverage for:
- Highways
- Urban roads
- Older residential neighborhoods
- Complex mixed-traffic scenarios
This is strategically important because Chinese EV makers are increasingly competing on how quickly they can turn advanced driver-assistance systems into a habit rather than a novelty.
Tesla vs XPeng: This Week's AI Signals Compared
| Metric | Tesla | XPeng |
|---|---|---|
| Main news focus | Manufacturing growth + AI data hiring | Real-world smart-driving usage milestone |
| Key date | May 7-8 | May 8 |
| Latest delivery/usage figure | Shanghai delivered 79,000+ vehicles in April | Smart-driving mileage share exceeded 50% |
| Growth metric | +36% YoY Shanghai deliveries | Takeovers per 100 km down 25.87% |
| AI strategy signal | Hiring data annotation teams for FSD and Optimus | Scaling end-to-end VLA intelligent driving |
| Geographic focus | China production, global exports, US AI hiring | China user fleet, real-world deployment |
| Notable technical angle | Pure vision, in-house labeling, real-world AI data engine | End-to-end action generation, <80 ms latency |
Why This Matters for the Chinese EV Market
The biggest takeaway is that the EV race is becoming an AI operations race.
For years, the Chinese EV market has been defined by:
- Fast product cycles
- Battery cost competition
- Aggressive pricing
- Localized software features
Those factors still matter, but the next phase is being shaped by a more difficult challenge: converting fleet scale into usable training data, then converting that data into better autonomous driving performance.
Tesla and XPeng are approaching this from different directions:
- Tesla is reinforcing the infrastructure layer: data collection, annotation, toolchains, and training support for FSD and robotics
- XPeng is emphasizing the application layer: real-world user adoption, lower intervention rates, and faster end-to-end intelligent driving
This distinction is important. A company can have strong algorithm demos, but if users do not trust or regularly activate the system, its commercial value remains limited. Conversely, high usage without robust data operations can slow long-term progress.
Global Implications
These developments also matter outside China.
Tesla's Shanghai factory is not just a China plant; it is a global export base feeding Asia-Pacific and influencing Tesla's position in Europe. If Shanghai output keeps rising while demand in overseas markets improves, the plant's strategic value will increase further.
XPeng's milestone, meanwhile, adds to the growing evidence that Chinese EV brands are becoming leaders in real-world advanced driver assistance deployment, not just cost-competitive hardware. That has implications for:
- Global intelligent-driving benchmarks
- Supplier competition in compute, sensors, and software tools
- Regulatory debates over what constitutes safe and usable driver assistance
- Potential export competitiveness of Chinese smart EV platforms
It also highlights an industry-wide labor shift. As AI tools automate basic labeling tasks, demand is rising for more specialized workers who understand traffic scenarios, robotics, and edge-case review. In other words, EV intelligence is creating not only better vehicles, but also a more specialized automotive AI workforce.
The Bigger Trend: EVs Are Becoming Data Companies on Wheels
This week's Tesla and XPeng news points to the same conclusion: the most competitive EV makers are increasingly behaving like vertically integrated AI companies.
The core assets now include:
- Vehicle production scale
- Software deployment speed
- User activation rates
- Real-world driving data
- Annotation quality
- Model iteration efficiency
That is why Tesla's hiring push and XPeng's usage milestone should be read together. One shows the invisible labor and infrastructure needed to improve machine intelligence. The other shows what happens when that intelligence reaches a level users are willing to rely on every day.
What Comes Next
The next questions are straightforward but consequential:
- Can Tesla translate its expanding data engine into visibly better FSD performance and stronger Optimus capabilities?
- Can XPeng maintain a smart-driving mileage share above 50% while continuing to reduce interventions?
- Will other Chinese EV brands such as NIO, Zeekr, and BYD respond with comparable real-world usage metrics rather than feature lists?
Expect the market to focus less on headline autonomy claims and more on measurable indicators such as:
- Intervention frequency
- Monthly active usage
- Per-vehicle smart-driving mileage
- Latency and processing efficiency
- Cross-scenario reliability
In China's EV market, the companies that win the next chapter may not simply be the ones building more cars. They will be the ones that collect better data, train better models, and persuade drivers to let the software take on more of the journey.
