XPENG Invited to CVPR for the Third Time, Showcasing China's Advances in Physical AI to the World

XPENG (NYSE:XPEV) showcased its Physical AI progress at CVPR 2026, marking its third participation.

Rhea-AI Summary

XPENG (NYSE:XPEV) showcased its Physical AI progress at CVPR 2026, marking its third participation. The company announced that its in-house driving foundation model VLA2.0 has entered formal mass production and reached over 50% assisted-driving mileage share in its first month.

XPENG reported a 1,010% uplift in per-GPU training efficiency, 4,360% gain in single-job efficiency, and GPU utilization rising from 40% to 90%. It is accelerating deployment of VLA2.0, robotaxis, and the IRON humanoid robot, targeting mass production by end-2026 and in-store use in Q1 2027.

Positive

• VLA2.0 driving foundation model has entered formal mass production
• VLA2.0 achieved over 50% assisted-driving mileage share in first month
• Per-GPU training efficiency improved by 1,010% over 12 months
• Single-job training efficiency increased by 4,360% over 12 months
• GPU hardware utilization improved from 40% to 90%
• IRON humanoid robot targeted for mass production by end of 2026

Negative

• None.

Argus Jun 4 session

-3.72% close to close Open Argus

Details

News Market Reaction – XPEV

In the Jun 4 session, XPEV declined 3.72%, reflecting a moderate negative market reaction.

Data tracked by StockTitan Argus on the day of publication.

Market Context

This announcement highlights XPENG’s continued build-out of its Physical AI stack, confirming VLA2.0...

Analysis

This announcement highlights XPENG’s continued build-out of its Physical AI stack, confirming VLA2.0’s entry into mass production and introducing a complementary world model with three core capacities. The company also reports large gains in training efficiency and higher GPU utilization, while targeting IRON humanoid robot mass production by the end of 2026. In context of prior AI events averaging a 3.24% move, investors may watch future adoption metrics, commercialization milestones, and integration into vehicle and robotics platforms.

Key Figures

Assisted driving share: over 50% Per-GPU efficiency uplift: 1,010% Single-job efficiency gain: 4,360% +5 more

Assisted driving share

over 50%

VLA2.0 assisted driving mileage share in first month

Per-GPU efficiency uplift

1,010%

Increase in per-GPU training efficiency over 12 months ending March

Single-job efficiency gain

4,360%

Gain in single-job training efficiency over 12 months ending March

GPU utilization start

40%

Cluster GPU hardware utilization 12 months earlier

GPU utilization current

90%

Cluster GPU hardware utilization, matching top-tier AI benchmarks

World model capacities

3

Deliberative Reasoning, Controllable Generation, Long-Horizon Forecasting

IRON mass production target

end of 2026

Timeline to bring IRON humanoid robot to mass production

Store deployment target

Q1 2027

IRON humanoid robots planned as shopping guides in offline stores

Previous AI Reports

5 past events · Latest: Apr 24

Same Type 5 events

1. Apr 24

   Auto show AI showcase

   24h Move

   +4.8%

   Showcased full-stack Physical AI ecosystem and VLA 2.0 user metrics at Auto China.

2. Apr 17

   Beijing show preview

   24h Move

   -0.8%

   Announced Physical AI ecosystem unveiling and global smart mobility vision for auto show.

3. Jan 12

   Flagship AI strategy

   24h Move

   +8.4%

   Outlined 2026 flagship P7+ launch, strong 2025 deliveries, and VLA2.0 AI rollout.

4. Sep 08

   IAA tech showcase

   24h Move

   +3.8%

   Presented AI mobility breakthroughs, European expansion, and 2026 L4 autonomy plans.

5. May 28

   AI-powered EV launch

   24h Move

   +0.0%

   Launched MONA M03 Max with AI smart driving at US$20,000 and strong delivery ramp.

24h Move is the share-price change in the day after each event; other market factors may also have contributed.

Key Terms

foundation model, embodied ai, world model, robotaxi, +2 more

6 terms

foundation model technical

"XPENG's in-house foundation model has now achieved a critical leap: VLA2.0..."

A foundation model is a large artificial intelligence system trained on vast, diverse data so it can be adapted to many tasks—like a universal engine that can be tuned to drive different products or services. Investors care because these models can lower the cost and time to build new AI-enabled offerings, create competitive advantages or concentration risks, and drive capital needs for compute, talent and regulation that affect company value.

embodied ai technical

"Workshop on Deployment of Foundation Models for Embodied AI (WDFM-EAI)..."

Embodied AI is artificial intelligence built into a physical device or robot that can sense, move, and interact with the real world rather than just run in software on a server. For investors, it matters because adding a “body” turns AI into products that require manufacturing, maintenance, sensors and software updates, creating different revenue streams, capital needs, safety and regulatory risks, and clearer paths to recurring service income—like software that also sells the hardware it runs on.

world model technical

"Dr. Xianming Liu introduced the world model as another core pillar..."

A world model is an internal, computer-built map of how parts of the real world behave and interact, used by software to predict outcomes and plan actions. For investors, it matters because companies that develop accurate world models can automate decision-making, improve forecasting, and build more capable products—potentially lowering costs and increasing competitive advantage—similar to how a pilot’s mental picture of weather and terrain helps them navigate more safely and efficiently.

robotaxi technical

"XPENG is accelerating the large-scale deployment of its core Physical AI applications: VLA2.0, Robotaxi, and Humanoid Robots."

A robotaxi is a self-driving vehicle designed to transport passengers without a human driver. It operates autonomously, using sensors and computers to navigate roads and pick up riders, similar to a taxi service but without a driver. For investors, robotaxis represent a potential shift in transportation, promising lower costs and new opportunities in mobility services.

humanoid robots technical

"deployment of its core Physical AI applications: VLA2.0, Robotaxi, and Humanoid Robots."

Machines built with human-like bodies or features—such as a head, arms, and legs—designed to perform tasks, interact with people, or navigate environments similarly to a person. Investors care because these robots can change labor needs, create new markets for products and services, and affect costs and productivity in industries like manufacturing, logistics, healthcare and retail; think of them as programmable workers that can reshape how businesses operate and compete.

gpu technical

"delivered a 1,010% uplift in per-GPU training efficiency and a 4,360% gain..."

A GPU (graphics processing unit) is a specialized computer chip designed to handle many calculations at once, originally for rendering images and video but now widely used for tasks like artificial intelligence, data analysis and high-performance computing. Investors watch GPU demand and prices because strong sales often signal growth for chip makers and their customers, affect profit margins and capital spending, and can forecast wider trends in gaming, AI adoption and cloud services.

AI-generated analysis. How Rhea-AI works. Not financial advice.

See more from StockTitan in Google Search and AI answers. Adds StockTitan as a preferred source · opens Google

Add on Google Not now

DENVER, June 4, 2026 /PRNewswire/ -- XPENG (NYSE: XPEV, HKEX: 9868), a leading China-based high-tech company, kicks off its key presence at CVPR 2026 (The IEEE/CVF Conference on Computer Vision and Pattern Recognition). Dr. Xianming Liu, Head of General Intelligence Center at XPENG, spoke at the inaugural Workshop on Deployment of Foundation Models for Embodied AI (WDFM-EAI), sharing insights with global counterparts including Tesla, NVIDIA and Waymo. This marks XPENG's third attendance at the prestigious conference.

From Technical Concept to Mass Production: VLA2.0 Achieves Key Technical Breakthrough

Dr. Xianming Liu systematically deconstructed the evolution of XPENG's physical AI technology system, from concept validation and technical refinement to full-scale mass production. First unveiled at CVPR 2025, XPENG's in-house foundation model has now achieved a critical leap: VLA2.0 has entered formal mass production, marking an industry-leading closed loop from pre-research to commercial deployment. Within its first month, VLA2.0 set an industry milestone with over 50% assisted driving mileage share, establishing a new benchmark in China's assisted driving sector.

First Technical Blueprint for World Model Unveiled, Advancing Physical-World Foundation Models

Dr. Xianming Liu introduced the world model as another core pillar of XPENG's foundation model system. XPENG is developing a world model capable of Deliberative Reasoning, Controllable Generation, and Long-Horizon Forecasting. Rather than competing, the world model and VLA2.0 complement each other: VLA2.0 learns "how to act" from human driving, while the world model learns "how the world changes after an action" by predicting future states and scene evolution. Together, they form XPENG's Physical-World Foundation Model, essentially pursuing the same goal: building a sufficiently powerful foundation model for the physical world.

XPENG specifies these three core capacities as essential for qualified world models and autonomous driving. Its R&D team has published three papers: X-World, X-Foresight, X-Cache, detailing corresponding R&D approaches.

Continuing Validation of Scaling Law: XPENG Accelerates Large-Scale Rollout of Physical AI

In the 12 months ending March this year, XPENG's cluster delivered a 1,010% uplift in per-GPU training efficiency and a 4,360% gain in single-job training efficiency, while GPU hardware utilization climbed from 40% to 90%, matching benchmarks set by top-tier global AI firms.

As VLA2.0's capabilities continue to expand, XPENG is accelerating the large-scale deployment of its core Physical AI applications: VLA2.0, Robotaxi, and Humanoid Robots. The IRON humanoid robot is progressing toward mass production by the end of 2026, with plans to enter XPENG's offline stores as a shopping guide in Q1 2027.

Appendix: XPENG World Model Related Academic Papers

X-World Paper: https://arxiv.org/pdf/2603.19979

X-World Official Site: https://x-world-1.github.io/

X-Cache Paper: https://arxiv.org/abs/2604.20289

X-Cache Official Site: https://x-cache-1.github.io/en/

X-Foresight Paper: https://arxiv.org/abs/2605.24892

X-Foresight Official Site: https://x-foresight-1.github.io/en/

About XPENG
Founded in 2014, XPENG is a leading Chinese AI-driven mobility company that designs, develops, manufactures, and markets Smart EVs, catering to a growing base of tech-savvy consumers. With the rapid advancement of AI, XPENG aspires to become a global leader in AI mobility, with a mission to drive the Smart EV revolution through cutting-edge technology, shaping the future of mobility.

To enhance the customer experience, XPENG develops its full-stack advanced driver-assistance system (ADAS) technology and intelligent in-car operating system in-house, along with core vehicle systems such as the powertrain and electrical/electronic architecture (EEA). Headquartered in Guangzhou, China, XPENG also operates key offices in Beijing, Shanghai, Silicon Valley, and Amsterdam. Its Smart EVs are primarily manufactured at its facilities in Zhaoqing and Guangzhou, Guangdong province.

XPENG is listed on the New York Stock Exchange (NYSE: XPEV) and Hong Kong Exchange (HKEX: 9868).
For more information, please visit https://www.XPENG.com/.

Contacts:
For Media Enquiries:
XPENG PR Department
Email: pr@xiaopeng.com

 

View original content:https://www.prnewswire.com/news-releases/xpeng-invited-to-cvpr-for-the-third-time-showcasing-chinas-advances-in-physical-ai-to-the-world-302791744.html

SOURCE XPENG

FAQ

AI-generated questions and answers. How Rhea-AI works. Not financial advice.

What did XPENG (NYSE:XPEV) announce about its VLA2.0 model at CVPR 2026?

XPENG announced that its in-house foundation model VLA2.0 has entered formal mass production. According to XPENG, VLA2.0 also reached over 50% assisted-driving mileage share within its first month, forming a closed loop from pre-research to commercial deployment.

How has XPENG improved AI training efficiency for autonomous driving by March 2026?

XPENG reported major gains in AI training efficiency for its Physical AI workloads. According to XPENG, per-GPU training efficiency rose 1,010%, single-job efficiency increased 4,360%, and GPU hardware utilization climbed from 40% to 90% in the 12 months ending March.

What is XPENG's Physical-World Foundation Model and world model strategy?

XPENG is building a Physical-World Foundation Model combining VLA2.0 and a complementary world model. According to XPENG, VLA2.0 learns how to act from human driving, while the world model focuses on deliberative reasoning, controllable generation, and long-horizon forecasting of how the environment changes.

When will XPENG's IRON humanoid robot reach mass production and in-store deployment?

XPENG plans for its IRON humanoid robot to progress toward mass production by the end of 2026. According to XPENG, IRON is expected to appear in its offline stores as a shopping guide starting in Q1 2027, supporting Physical AI commercialization.

How is XPENG using Physical AI to support robotaxi and humanoid robot deployment?

XPENG is accelerating deployment of its Physical AI applications, including VLA2.0-based robotaxis and humanoid robots. According to XPENG, these systems leverage its foundation models and world model research, aiming to bring advanced assisted driving, robotaxi services, and in-store robotic assistants to large-scale use.

What academic work has XPENG released to support its world model for autonomous driving?

XPENG has released three academic papers related to its world model: X-World, X-Foresight, and X-Cache. According to XPENG, these papers describe key R&D approaches for deliberative reasoning, long-horizon forecasting, and efficient caching in physical-world foundation models.