Alibaba

ByteDance

CATL

Huawei

Meituan

Tencent

Xiaomi

Most Followed

KrASIA Connection

Venture Roundup

Earnings Reports

China's Auto Sector

Inside China's Startups

100 AI Creators

View All

Collections

News

Features

Insights

Ecosystem

Press Releases

DeepSeek

Li Auto

Luckin Coffee

Mixue Bingcheng

Pop Mart

Xpeng

On the Radar

Events

Latest

About

Consumer

Robotics

Fintech

Collaborate

With the US issuing new worldwide restrictions on the use of <a href="/company/huawei">Huawei's</a> artificial intelligence chips, Chinese tech giant <a href="/company/tencent">Tencent</a> said on May 14 that it has enough previously acquired high-end chips to sustain its AI training "for a few more generations" and is exploring ways to boost AI inference efficiency, including through software optimization.
"So it's a very dynamic situation, and we just … have to manage the situation," Tencent president Martin Lau said on a conference call with analysts. This means not only addressing it "in a completely compliant way," but also "trying to figure out the right solution for us to make sure that our AI strategy … can still be executed."
On May 13, the US Bureau of Industry and Security (BIS) issued new guidance on chips for AI applications, restricting anyone, anywhere in the world, from using AI chips from Huawei—an unprecedented crackdown on the company despite the recent de-escalation of the tariff war.
The BIS also said it will issue a new rule in the future to replace the Biden-era "AI diffusion rule," which would have sorted countries into three tiers and imposed limitations on the second and third tiers. Under that rule, even friendly countries such as Mexico would have faced restrictions on the advanced AI chips they could import from the US.
In April, the US government started requiring <a href="/company/nvidia">Nvidia</a> to obtain licenses for exporting its H20 chip—which was introduced after Washington tightened export controls in 2023—to China over concerns about potential military applications. But Chinese technology giants including <a href="/company/bytedance">ByteDance</a>, <a href="/company/alibaba">Alibaba</a>, and Tencent <a title="" href="https://asia.nikkei.com/Business/Tech/Semiconductors/ByteDance-Alibaba-and-Tencent-stockpile-billions-worth-of-Nvidia-chips" target="_blank" rel="noopener">had already stockpiled</a> billions of dollars' worth of H20s in advance, according to Nikkei Asia.
Lau said that while there is a short supply of graphics processing units, the company has lowered the priority of its GPU rental businesses and prioritized internal usage instead. The inventory will be used for the applications that will generate immediate returns, such as the advertising business and content recommendation products.
Training of its large language models, which requires higher-end chips, will be the next priority, Lau said.
"And now we can see, even with a smaller cluster, you can actually achieve very good training results," he said. This helps the company look at its existing inventory of high-end chips and say that "we should have enough high-end chips to continue our training of models for a few more generations going forward."
In inference, which requires relatively fewer high-end chips, the company is focusing on improving AI inference efficiency through software optimization, which could effectively double GPU capacity, Lau said.
"We can customize different sizes of … models, especially some applications do not require very large models," he said, adding that Tencent can also tailor-make and distill models for different use cases, saving on inference usage of GPUs. The company is exploring compliant chip options available in China or for importation, Lau said.
Tencent posted much better-than-expected quarterly earnings on May 14 due to strong performances of its AI-enhanced advertising and evergreen games.
In the January-March quarter, revenue of the Shenzhen-based company rose 13% on the year to RMB 180 billion (USD 25.2 billion), beating the average forecast of RMB 174.6 billion (USD 24.4 billion) from analysts surveyed by LSEG.
Net income attributable to equity shareholders for the quarter came to RMB 47.8 billion (USD 6.7 billion), up 14%.
Chairman and CEO Pony Ma said AI capabilities contributed "tangibly" to Tencent's businesses such as advertising and games in the first quarter and the company has stepped up its spending on new AI opportunities.
"We expect these strategic AI investments will create value for users and society, and generate substantial incremental returns for us over the longer term," Ma said.
In the first quarter, the company revamped its ad platform with the enhanced generative AI tools, making creating ads faster with improved image generation and video editing. The company has also used AI to boost live streaming and optimize its recommendation algorithm.
Tencent's value-added services, which includes gaming and social networks, contributed over 50% of its revenue. While its gaming division faces sector-specific headwinds like regulatory oversight and shifting consumer trends, flagship titles such as Honor of Kings and Peacekeeper Elites and the recently released Dungeon &amp; Fighter Mobile and Delta Force helped domestic games grow 24% to RMB 42.9 billion (USD 6 billion), partially against a low base. International games grew 23% to RMB 16.6 billion (USD 2.3 billion), driven by growth in revenues from Brawl Stars, Clash Royale, and PUBG Mobile.
Despite broader industry pressures, Tencent's online advertising grew 20% to RMB 31.9 billion (USD 4.5 billion). Over recent years, the company has embedded AI into its ad platform, capitalizing on its ubiquitous app WeChat to solidify its strong position in China's digital advertising market. Earlier this year, it integrated Chinese startup <a href="/company/deepseek">DeepSeek's</a> technology into its various platforms, as Tencent's self-developed AI tool Yuanbao was losing steam against major rivals.
However, its fintech revenue, which includes WeChat Pay, grew only 5% to RMB 54.9 billion (USD 7.7 billion) as China's consumption remained tepid.
Tencent's shares have rallied 24% since the start of the year, outpacing the broader Chinese equity performance in Hong Kong. Yet it remains 27% below its all-time peak.
This article first appeared on <a href="https://asia.nikkei.com/Business/Technology/Tencent-says-chip-stockpile-can-power-AI-training-for-generations-despite-US-ban">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing partnership with <a href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.

Tencent says chip stockpile can power AI training for "generations" despite US ban

East Asia

China

Artificial Intelligence

Deep Tech

AI Infrastructure

<a href="http://jq-industries.com/">JQ Industries</a>, a developer of flexible fiber sensors based in Weihai, has raised an eight-figure RMB sum in its Series A funding round. Backers include JHF, New Oriental, and Freyr Graphite. The company said the new capital will support R&amp;D in robotic skin and haptic sensing, production expansion, and market development. InvesTarget served as the financial advisor.
This marks JQ’s second funding round since its founding in early 2024. A new round is already underway to secure funding for continued R&amp;D, scale production, and build out the team.
JQ specializes in flexible fiber sensors made from plant-derived materials and textiles, aimed at use cases across <a href="/industry/robotics">robotics</a>, <a href="/industry/health-tech">healthcare</a>, and <a href="/collections/china-automobile">automotives</a>. Its portfolio includes both hardware like robotic skins and fingertip sensors, as well as software platforms for algorithmic processing, smart eldercare, and in-car comfort monitoring.
The rise of <a href="/industry/ai">artificial intelligence</a> and 5G communications has pushed demand for sensors that are adaptable, precise, and capable of functioning in complex environments. Flexible sensors are particularly relevant where conventional rigid sensors fall short, such as in motion tracking or pressure detection during physical interaction.
Still, most current flexible sensors are built on thin-film pressure technology, which is limited by issues with elasticity, breathability, and resistance to environmental interference. These constraints hinder performance and restrict mass production.
JQ’s answer is a fourth-generation resistive flexible fiber sensor. Its architecture deviates from planar films, using fiber-based layers for conductivity and sensitivity. The sensor integrates a fiber nylon core, a piezoresistive outer layer, and an electrode core—all within a flexible yarn.
This new material architecture offers exceptional flexibility and durability. The sensors can reportedly be folded to a radius of less than 0.1 millimeters and have passed automotive-grade impact tests. They retain high elasticity and rebound performance without requiring recalibration and demonstrate strong resistance to environmental interference. When integrated into textiles, the sensors remain breathable and comfortable against the skin, making them suitable for diverse use cases.
<figure id="attachment_153894" aria-describedby="caption-attachment-153894" style="width: 640px" class="wp-caption aligncenter"><img loading="lazy" decoding="async" class="size-full wp-image-153894" src="https://console.kr-asia.com/wp-content/uploads/2025/05/JQ-Industries-Demo.gif" alt="GIF shows real-time 3D pressure mapping using JQ Industries’ fiber sensor array. The system visualizes pressure distribution across a surface, offering high-resolution feedback for applications like smart bedding and robotic skin." width="640" height="360" /><figcaption id="caption-attachment-153894" class="wp-caption-text">GIF shows real-time 3D pressure mapping using JQ Industries’ fiber sensor array. The system visualizes pressure distribution across a surface, offering high-resolution feedback for applications like smart bedding and robotic skin. Graphic and header image source: JQ Industries.</figcaption></figure>
“We’re now heavily focused on humanoid robotics and scaling fiber sensors into large-area electronic skin,” founder Zhang Jingqi told 36Kr. “These robotic skins can help robots sense their environment more holistically, interact like humans, and ensure safer human-robot interaction. For example, foot-embedded sensors can detect surface hardness and balance to help robots walk more stably.”
According to Zhang, the company’s woven fiber sensors offer distinct advantages for humanoid robotics. He cited the G1 skin suite from Unitree Robotics, which includes glove, torso, and sole components, as a representative application. The suite provides high levels of flexibility, elasticity, breathability, and durability. The gloves deliver multi-point tactile feedback across the palm and support both motion capture and remote manipulation. They accommodate joint folds down to zero and can detect forces as low as 0.01 newtons, with a pressure sensitivity range up to 3,500 kilopascals. When integrated with <a href="/company/nvidia">Nvidia’s</a> Isaac Sim, the sensors enable tactile data modeling and deployment in closed-loop systems.
The company is currently negotiating with over ten humanoid robotics companies to supply upper-body and limb skin systems. Some units are already being shipped in volume.
Beyond robotics, JQ is extending its sensor technology to eldercare and smart mobility. It has signed partnerships with institutions like Tian’e Daojia, Taikang, and Yuecheng Senior Living. Its eldercare mattress monitors pressure distribution, posture, sleep quality, and breathing to detect early signs of bedsores or falls.
In the automotive sector, JQ is working with major carmakers to embed its sensors into cockpits. Use cases include comfort tuning and gamified car seat experiences. Some of these integrations are now entering mass production.
One of the biggest hurdles in sensor commercialization is manufacturing consistency. JQ has addressed this by building its own factory, retrofitting textile machinery, and optimizing dyeing and finishing. These changes have cut production costs by 80% and improved yield consistency by 200%.
The firm now runs an industrial roll-to-roll line capable of producing 200,000 sensors annually at a 95% yield. New samples can be prototyped within 24 hours.
JQ’s team of 30 includes alumni from the Chinese Academy of Sciences and top hardware firms. Zhang, who holds a degree in electronic engineering from the University of York, previously served as CTO at a <a href="/company/xiaomi">Xiaomi</a> ecosystem company.
“JQ Industries delivers a compelling solution in embodied haptics. True full-body robotic skins require scalable, cost-effective, and reliable solutions,” said Zhou Lifeng, a partner at JHF. “JQ’s fiber-textile approach enables a rare combination of flexibility, sensitivity, accuracy, range, durability, and affordability.”
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3303658591607296">article</a> was written by Zhang Zhuoqian for 36Kr.

With fresh funding, JQ Industries bets on fiber sensors to give machines a sense of touch

Internet of Things

On May 22, after weeks of muted presence following a <a href="https://kr-asia.com/three-dead-in-su7-crash-fatal-accident-renews-scrutiny-of-smart-driving-safety">high-profile traffic accident</a>, <a href="/company/xiaomi">Xiaomi</a> returned to the spotlight with the introduction of two headline products: the XringO1, a smartphone chip built on a three-nanometer process, and the YU7, its second <a href="/industry/ev">electric vehicle</a> model.
The playbook is familiar. In 2023, <a href="/company/huawei">Huawei</a> and Seres revived interest in the Aito M7 by releasing a new version with Huawei’s in-house Kirin chip. That update transformed its fortunes, as orders surged past 30,000 in the first month after relaunch. The move buoyed both Aito and the Harmony Intelligent Mobility Alliance (HIMA).
Xiaomi appears to be charting a similar course. By spotlighting a significant chip breakthrough, it is injecting fresh energy into its EV efforts, using the YU7 to reshape public perception and restore market momentum.
<h3>Design, range, and safety take center stage</h3>
Design has consistently been a key pillar of Xiaomi’s EV strategy, and the YU7’s introduction was no exception. The vehicle closely follows the visual language of the earlier SU7, with consistent proportions, color palette, and detailing. The major difference? Aerodynamics.
The YU7 incorporates 40 aerodynamic enhancements, including ten through-body air channels and 19 vents to optimize airflow. These features yield a drag coefficient of just 0.245, contributing directly to improved driving range.
The YU7 comes in three variants. The Standard and Pro models both feature a 96.3 kilowatt-hour lithium iron phosphate battery. The Standard offers a range of 835 kilometers, while the all-wheel-drive Pro version delivers 770 km. The Max model uses a 101.7 kWh ternary lithium battery, with a combined range of 760 km.
These figures place the YU7 among the top-performing EVs in its class. The Max variant features an 800-volt architecture and supports 5.2C fast charging, capable of adding 620 km of range in just 15 minutes, which may be an appealing feature for performance-minded buyers.
<h3>Safety architecture takes a leap forward</h3>
Passive safety was a major focus of the launch. Xiaomi enhanced the YU7’s structural integrity with reinforced crumple zones and a 1,500-megapascal anti-scratch crossbeam to protect the battery from road debris. The battery itself is encased in the same anti-ballistic material used in the premium SU7 Ultra.
The company also developed new materials in collaboration with partners, including 2,200 MPa hot-formed steel. This steel is used for side-impact beams on all four doors and integrated into the A and B pillars using a roll-cage-inspired design. The intent is clear: maximize cabin protection in the event of a collision.
The YU7 passed more than 50 crash tests, meeting all requirements under C-NCAP and C-IASI standards, signaling that Xiaomi is serious about safety.
Assisted driving technology also received a boost. All YU7 models come standard with <a href="/company/nvidia">Nvidia’s</a> Thor platform, built on a 4-nm process and capable of 700 TOPS (tera operations per second). The vehicle’s sensor suite includes long-range LiDAR (light detection and ranging), 4D millimeter-wave radar, 11 high-definition cameras, and 12 ultrasonic sensors.
To ensure optical clarity, Xiaomi applied ALD (atomic layer deposition) coatings to all seven external cameras to reduce glare and enhance visual input, further strengthening safety during assisted driving.
With deliveries set to begin in July, just three months after the SU7 accident, questions have emerged about whether the YU7's safety features were introduced under pressure. But industry sources told 36Kr that components like the new steel structures and crash beams have been in development for over a year. “These weren’t reactive stopgaps after the March incident,” one insider said.
Moreover, unlike the SU7, which featured tiered assisted driving capabilities, the YU7 adopts a uniform hardware configuration across all variants, which may have been a factor that contributed to the delayed launch.
whereRelatedArticle
<h3>Redemption through reinvention?</h3>
The SU7 accident in late March cast a long shadow over Xiaomi’s credentials as a new automaker. Can a three-year-old EV brand truly deliver on safety?
Xiaomi’s answer has been decisive. By early May, it had quietly rebranded the SU7’s <a href="https://kr-asia.com/after-a-fatal-su7-crash-china-reins-in-smart-driving-hype">“smart driving”</a> system on its website to the more conservative “assisted driving.” On stage in May, it doubled down, highlighting everything from materials science to hardware design to its own driving academy, all geared toward one message: safety is now core to the brand.
These measures appear to be working, and the YU7 may prove a turning point in restoring trust and generating demand.
Founder Lei Jun compared the YU7 lineup to <a href="/company/tesla">Tesla’s</a> Model Y variants, noting that Xiaomi’s vehicles surpass Tesla’s in range and offer features like air suspension, which the Model Y lacks.
That speaks to the ambition of Xiaomi: to take on the world’s bestselling electric SUV. If pricing mirrors the SU7’s competitive strategy, the YU7 could challenge Tesla not just on specs, but on affordability too.
As Chinese EV buyers place greater emphasis on practical factors like safety and reliability, the YU7—with its strong specifications and reinforced safety architecture—represents Xiaomi’s clearest shot yet at another breakout success.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3303648590093056">article</a> was written by Xu Caiyu for 36Kr.

After SU7 setback, Xiaomi reboots EV push with safer, smarter YU7

Transportation

Electric Vehicles

Mobility

Chinese <a href="/industry/robotics">robotics</a> startup <a href="http://www.dp-robot.com/Default.html">Slabor</a> has raised an eight-figure RMB sum in a seed funding round led solely by Plum Ventures. The company said the funds will accelerate product iteration, expand its hardware lines, and scale distribution channels.
Headquartered in Guangming, Shenzhen, Slabor develops robots to automate screw tightening, which is a common but intricate step in industrial manufacturing. Its offerings include four- and six-axis tightening robots, intelligent fastening tools, and integrated robotic workstations. The technology is already in use at companies like <a href="/company/byd">BYD</a>, Ecovacs, and Sunwoda.
Manual tightening processes remain inefficient and inconsistent globally, particularly as labor costs rise and traceability becomes more important. Traditional XYZ module-based systems often fall short in high-precision settings, where the ability to extract production data, accommodate design changes, and support customization is essential.
Founder Bai Fengfeng told 36Kr that products like new energy vehicles and smart home appliances now have shorter iteration cycles and higher demands for assembly reliability. This creates greater expectations for tightening consistency, robotic flexibility, and data integration. To meet these demands, Slabor offers a dedicated tightening robot that combines a robot-based mechanical system with PC-based control and smart tools.
Foreign players like Epson and Atlas Copco have long dominated the high-end tightening space with costly, complex solutions that may offer only limited support. Meanwhile, domestic competitors like Inovance and Estun typically focus on general-purpose industrial robotics. Slabor targets a less saturated niche: process-specific robotic automation.
According to Bai, legacy systems often require assembling disparate components—robots, electric screwdrivers, vision systems—resulting in high costs and poor integration. Slabor’s approach bundles proprietary robotics and tightening technology into one system. The company claims this reduces machine costs to around 60% of comparable systems and cuts delivery lead times significantly.
On the hardware front, Slabor’s SCARA (selective compliance assembly robot arm) robots are equipped with proprietary smart tightening tools. These tools reportedly deliver top-tier repeatable positioning accuracy and support a wide range of screw sizes. Slabor says its system improves tightening efficiency by 15–30% per screw compared to conventional setups. Real-time monitoring of torque, angle, and pressure ensures control precision within 1%, with performance metrics said to rival international leaders.
The software system is modular and integrates robot control with tightening processes. It features guided workflows, role-specific interfaces, and intuitive design. Setup is simplified through one-click teaching that overlays CAD files with vision data.
On the commercial front, following deployments with major automotive clients like BYD and Great Wall Motor, as well as smart home appliance makers such as Ecovacs and Roborock, Slabor is ramping up its channel marketing. The company is leveraging standardized product models alongside a robust sales network to broaden its market reach. It has signed nearly 100 distribution partners and is further developing its end-to-end smart assembly solutions.
Looking ahead, Bai said the company will focus on three core pillars: performance, cost-efficiency, and ease of use. Structurally, Slabor aims to boost component reusability and customization through a modular product architecture. On the software side, it plans to refine configuration tools and interface design to support more efficient workflow management. To enhance operational performance in demanding environments, the company is introducing industry-specific training models.
Slabor’s leadership team brings deep expertise in motion control and automation. Bai previously held sales leadership positions at two US-listed companies. The core R&amp;D team includes graduates from Harbin Institute of Technology and Beijing Institute of Technology, each with over a decade of experience in the field.
“We’ve always believed in technology’s potential to reshape the future, and artificial intelligence-powered tightening robots are a critical part of that transformation,” said Wu Shichun, founding partner at Plum Ventures. “Slabor stands out for its technical edge, strong product portfolio, and market insight. We see significant value in its approach.”
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3302324200856070">article</a> was written by Lin Qingqing for 36Kr.

A smarter way to tighten screws? Slabor lands early funding to prove it

… “That means we can move gradually,” Wu said. “We already have a broad suite of financial APIs, and now we can combine them with AI agents to build real financial automation.”
In the SaaS era, businesses bought tools. In the AI era, those tools become part of an agent’s toolkit, delivering services autonomously.
That’s also where precision matters. Finance teams spend huge amounts of time reconciling accounts and tracking fund flows. AI can automate much of this, but only if accuracy is guaranteed. That’s the rationale Airwallex uses for deploying feedback loops between humans and machines, identifying high-impact use cases and adjusting where needed.
AI is also central to the company’s scaling strategy. When assessing new tools, Airwallex considers three criteria: technical compatibility, disruption potential, and user response.
Ultimately, Airwallex’s goal is to pioneer “AI agentic finance,” a model where AI not only supports, but in some cases replaces roles like CFOs. An AI agent could manage vendor relationships, approve expenditures, and handle invoice payments, taking over traditional accounts payable functions.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3302382387911175">article</a> was written by Ren Qian for 36Kr.

Airwallex raises USD 300 million, eyes IPO readiness and AI-led finance

Payments

Southeast Asia

… picture? And what role does Eagoal play in it?</h3>
CB: Battery recycling plays a key role in the broader push toward a circular economy. With growing global demand for sustainable development and resource efficiency, effectively recycling used batteries can relieve resource shortages and reduce environmental impact. It also accelerates the shift to low-carbon growth.
Eagoal is focused on aligning with green and circular economic trends. We’ve gone beyond just manufacturing and using batteries. We’re deeply involved in closing the loop at the end of the battery life cycle. This end-to-end approach allows us to integrate upstream and downstream resources, boost overall supply chain efficiency, and secure a strong position in the new energy post-market.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3281853106856837">article</a> was written by Wang Shun for 36Kr.

As EV batteries retire, recyclers like Eagoal race to scale and standardize

Batteries

Energy

… that handles shipments to America. "We recommend shipping now because we don't know what will happen after three months. [Trump] might even change his mind within two months," he said.
China's slowed export growth in April suggests some US importers canceled or scaled down shipments. Some products bound for the US sat in Chinese warehouses, waiting for tariffs to fall. Analysts predict a boost in Chinese exports as shippers front-load deliveries in the 90-day window.
An employee at a Yiwu-based manufacturer of Santa Claus dolls and other Christmas decorations said he had expected tariffs to come down eventually, but was relieved that a deal was reached sooner than anticipated. "30% is not an issue. Many customers can accept it," he said, referring to the lowered US tariffs on Chinese imports.
Peter Sand, chief analyst at Xeneta, expects this year's peak ocean shipping season, which usually falls in the third quarter, to start earlier. Shipping rates may consequently rebound, he added.
"April and May are our busiest months, and we are already making orders from other countries," said the employee from the Yiwu manufacturer. "So even if you order today, you will not be able to receive the goods quickly."
This article first appeared on <a href="https://asia.nikkei.com/Economy/Trade-war/US-China-trade-war-truce-5-things-to-know">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing partnership with <a href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.

US-China trade war truce: 5 things to know

Xia Zhongpu, the executive who led development of <a href="/company/li-auto">Li Auto’s</a> end-to-end (E2E) assisted driving model, has left the company, signaling a broader strategic shift at the Chinese automaker.
Xia, who joined Li Auto in 2023 and reported directly to Lang Xianpeng, vice president of assisted driving R&amp;D, was a prominent figure within the company’s advanced driving unit. People familiar with the matter said he had already stepped away from Li Auto’s vision-language-action (VLA) model development team and had not attended internal meetings for several weeks. His next move remains unclear.
Li Auto did not respond to 36Kr’s request for comment.
Xia’s departure marks a turning point in the company’s evolving approach to assisted driving technology. He was instrumental in Li Auto’s early E2E architecture, which gained traction following promising results and led to a <a href="https://kr-asia.com/li-auto-leans-into-end-to-end-autonomy-with-smart-driving-overhaul">restructuring</a> of the assisted driving division in November 2024. The company established three dedicated teams at the time: E2E modeling, world modeling, and production R&amp;D. Xia was named head of the E2E unit.
During his tenure, Xia rose quickly within Li Auto’s hierarchy. However, internal sources suggest his exit may stem from differing views on the company’s technological direction. “Xia believed the E2E approach could still be optimized,” said one person familiar with the matter. “But internally, the company had already gone all-in on VLA.”
The transition was underscored on May 7, when CEO Li Xiang described VLA during a company-hosted “AI Talk” session as “a large model that drives like a human.” He also revealed that Li Auto had tripled its planned GPU training allocation to support VLA development.
This pivot has coincided with a reshuffling of leadership. Lang has reportedly been promoted, and Jia Peng, previously in charge of the company’s world model framework, now leads the VLA project as head of autonomous driving technology R&amp;D.
Li Auto’s assisted driving roadmap has undergone multiple changes since 2023. The company moved away from high-definition maps and rule-based logic, adopted E2E frameworks, and now appears committed to VLA as its core development path.
Originally popularized by Tesla, the E2E model allowed Li Auto to streamline core modules—perception, prediction, planning, and control—into a unified system driven by machine learning. The company formally began E2E development in November 2023 and rolled out a combined E2E and vision-language model (VLM) to Max users by October 2024, reportedly two months ahead of schedule.
That rollout was seen as a major win, helping Li Auto shed its image as a laggard in assisted driving and emerge as a top-tier contender. Xia, as the head of the production-ready E2E system, benefited from the momentum and earned internal promotions.
Yet, consensus on E2E’s long-term viability never fully materialized. At the same May 7 event, CEO Li Xiang noted a critical limitation: “End-to-end doesn’t fully understand the physical world. It’s more of an imitation process,” he said, pointing out that the system performed well in general scenarios but struggled in complex, unfamiliar ones.
Management saw those constraints as fundamental. While Li Auto had already added VLM features to its E2E stack, the team concluded that vision-language alone could not close the gap in higher-order decision-making.
VLA, originally developed by Google DeepMind for robotics applications, is now viewed within Li Auto as the future of embodied intelligence. Unlike VLM systems that parse 2D visual inputs, VLA incorporates an action component, allowing models to not only interpret their environment but also perform physical tasks like driving. These systems support 2D and 3D vision, language comprehension, and chain-of-thought reasoning, essentially enabling vehicles to see, think, and act more like humans.
Industry peers are also trending in this direction. Nio has released a world model focused on reading road signs and textual input, while Xpeng has introduced a foundation model with advanced reasoning capabilities that has since been optimized for in-car use.
Still, the VLA route is not without skeptics. Some experts interviewed by 36Kr noted that the approach remains largely untested in real-world conditions. Even Li acknowledged the risks: “In truth, we’re venturing into uncharted territory,” he said.
More than just a personnel change, Xia’s exit may perhaps reflect the end of an ambitious phase in Li Auto’s assisted driving journey, and the company’s deeper bet on a new frontier.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3300770853882121">article</a> was written by Li Anqi for 36Kr.

Li Auto’s E2E architect departs as company shifts focus to VLA

Autonomous Vehicles

… advancement and real-world validation.
Working with academic institutions allows us to test and refine our ideas in practical settings. We’re now expanding our model beyond Peking University to include partnerships with Peng Cheng Laboratory, Hong Kong University of Science and Technology, Beijing Jiaotong University, and Harbin Institute of Technology. Each collaboration is structured to suit the strengths and goals of both parties.
<h3>36Kr: Which industries will Shenqi Digital prioritize going forward?</h3>
SY: Our technology is cross-sector by design. Every industry needs better data governance and intelligence.
That said, we’ve already built momentum in areas like supply chain, finance, water management, transportation, and hospitality. For instance, we currently serve 80% of China’s high-end hotel market.
We’ll continue to deepen our reach in these sectors while expanding into new ones. Through our wider ecosystem, we’re connecting with major players in consumer goods, automotive, electronics, and beauty.
We’re also actively exploring opportunities in industrial manufacturing and energy, thanks to resources and connections provided by our partners.
<h3>36Kr: Is Shenqi Digital planning to expand overseas?</h3>
SY: Yes. Our outbound approach is what we call “accompaniment-style” globalization. Many of our domestic clients are expanding overseas and want us to replicate their digital systems in new markets.
As we support them, we’re also building local service networks abroad and exploring “reverse expansion,” helping international companies establish operations in China.
We already have a strong presence in Hong Kong and Macau, with longstanding relationships in academia and finance. Our Hong Kong-based delivery teams understand the region’s language, legal frameworks, and standards, putting us in a strong position to expand into Southeast Asia and beyond.
<h3>36Kr: What are Shenqi Digital’s priorities over the next three to five years?</h3>
SY: First, advancing our product and technology pipeline. With Yanyun Infinity as the foundation, we’re developing AI applications tailored to vertical industries and assembling hybrid teams of engineers and industry experts to lead innovation.
Second, expanding our service and sales network. We’re upgrading local service hubs to meet rising demand, including deployments in fourth- and fifth-tier cities.
Third, ecosystem development. Our guiding principle is built on altruism, openness, and win-win collaboration. We’re strengthening our academic partnerships and broadening our partner ecosystem to co-develop AI solutions that create new value in traditional industries.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3202733866410505">article</a> was written by Song Wanxin for 36Kr.

With IT data drying up, Shenqi Digital backs “AI for process” to sharpen enterprise decisions