言零_yanling

Big Companies:
Japanese company ispace has postponed the NASA-funded lunar mission to 2030.
Japanese spacecraft startup ispace announced on Friday that after two failed lunar lander missions, the company would strategically adjust and delay a NASA-funded lunar exploration mission to 2030, while also reducing its global workforce. This announcement highlights the uncertainty surrounding the company's future. Currently, the United States is restructuring its space missions in collaboration with commercial and international partners. (Sina Finance)
Cainiao Group signs strategic cooperation agreement with CP AXTRA under Thailand's Charoen Pokphand Group.
36Kr learned that today, Cainiao Group signed a strategic cooperation agreement with CP AXTRA under Thailand's Charoen Pokphand Group. Cainiao will apply its digital supply chain technology, warehouse automation technology, and overseas warehouse operation experience to fully support CP AXTRA in building an integrated online and offline instant retail logistics network in Thailand and ASEAN countries. CP AXTRA manages over 2,700 offline stores in Thailand and Asia.
Wanda Film: Plans to change its stock abbreviation to Ruyi Film.
36Kr learned that Wanda Film announced plans to change its company name to Ruyi Film Entertainment Co., Ltd., change its stock abbreviation to Ruyi Film, and accordingly revise its Articles of Association.
Alibaba's consumption platform drives growth in Fliggy car rentals, expanding 88VIP order user scale by 40%.
36Kr learned that on March 27, at the Fliggy car business ecosystem partner conference, it was disclosed that in the past fiscal year, the Fliggy car rental business maintained strong growth, particularly benefiting from resources, scenarios, and technology from Alibaba's consumption platform, with car rental bookings from Taobao increasing by nearly 60% year-on-year, and the number of ordering users growing by 40%. The cross-selling of airline tickets, hotels, and car services has reached new heights, with more than 50% of orders coming from users of the aforementioned categories for pick-up services.
Some Wahaha Hongsheng factories halt production, with some factories already on holiday.
On March 27, a person close to Wahaha stated that Wahaha notified last night (March 26) that it would halt 70% of its factory production lines, including multiple factories from both Wahaha and Hongsheng, with this notification given verbally. Currently, factory shipments have paused except for necessary orders, and some factories have also issued holiday notices, with production expected to resume around April 2. Additionally, another person close to Wahaha revealed that this is due to production scheduling issues. (Interface)
Razer: More than half of brand sales come from new products, IP co-branded new products launch exclusively on Tmall.
On March 26, Razer's global vice president Chen Xiaoping stated at the 2026 Tmall TOP TALK that over 50% of sales in our brand flagship store in 2025 came from new products, which are the core engine for business growth. Razer has always launched exclusive co-branded new products with IPs like Pokémon, Sanrio, and BLACKPINK on Tmall. Recently, Razer's annual flagship new product, the "Venom V4 Professional Edition," was launched on Tmall, providing Tmall users with exclusive rights such as customized gaming props and limited merchandise.
PingTouGe's ZhenYue 510 chip shipments exceed 500,000 units, with multiple storage companies already integrating this chip into SSDs.
36Kr learned that on March 27, at the 2026 CFMS Flash Memory Summit, PingTouGe Semiconductor announced that the cumulative shipment of its SSD controller chip ZhenYue 510 has exceeded 500,000 units, making it one of the highest shipment controller chips in the country recently. Currently, ZhenYue 510 has been scaled up in multiple core businesses of Alibaba Cloud, with storage manufacturers like Yiheng Chuangyuan, Derui Lingxin, Baiwei Storage, and Changjiang Wanrun launching storage products based on this chip.
XPeng Motors: The company’s Chinese name will change from "XPeng Motors Co., Ltd." to "XPeng Group."
36Kr learned that XPeng Motors announced that starting from April 1, 2026, its Chinese name (for identification purposes only) will change from "XPeng Motors Co., Ltd." to "XPeng Group." The company's English name remains unchanged as "XPeng Inc." Starting from 9:00 AM on April 1, 2026, the Chinese stock abbreviation of the company's Class A ordinary shares traded on the Main Board of the Hong Kong Stock Exchange will change from "XPeng Motors–W" to "XPeng Group–W."
Lloyds Bank IT failure leads to the leakage of transaction data for nearly 500,000 customers.
The UK Treasury Committee stated on Friday that earlier this month, Lloyds Banking Group experienced an IT failure that resulted in the personal information of up to 447936 customers being leaked, allowing users to view others' transaction records, including account details and national insurance numbers. (Sina Finance)
New Products:
Deep Intelligence releases a new embodied general intelligence model.
36Kr learned that on March 27, at the Zhongguancun Forum event, the first embodied intelligence company incubated by Beijing Zhongguancun Academy and Zhongguancun Artificial Intelligence Research Institute—Deep Intelligence—officially released its first embodied general intelligence base model, PhysBrain 1.0, built on human learning paradigms. PhysBrain 1.0 adopts a multimodal large model architecture, internalizing physical knowledge within its parameters, enabling the model to possess spatiotemporal consistency in understanding the physical world, achieving true generalization capability with limited data.
Ninebot: Cooperation with Pop Mart to launch co-branded electric vehicle in April.
On March 27, Ninebot announced a brand cooperation with Pop Mart's popular IP Little Sweet Bean, focusing on the young consumer demographic and emotional value expression for deep co-creation, with plans to launch the first IP co-branded electric vehicle product in April. (Securities Times)
Orion Star launches AI Agent service robots.
36Kr learned that on March 27, the "True Utility Creates Real Business" themed Orion Star Partner Win-Win Conference and Leopard Little Secret series new product launch was held in Beijing. At the conference, Orion Star officially launched two AI Agent voice interaction service robots—Leopard Little Secret Lite and Leopard Little Secret Pro, and introduced the Robot Agent store.
Other news worth noting:
The State Administration of Foreign Exchange: 2025

On March 27, Longfor Group released its 2025 annual performance report. The report shows that Longfor Group's operating revenue for 2025 is approximately 97.31 billion yuan. 
Among them, the revenue from real estate development business is 70.54 billion yuan; the revenue from operations and services is 26.77 billion yuan, accounting for 27.5% of the group's operating revenue, reaching a record high; the profit attributable to shareholders is 1.02 billion yuan; after excluding the impact of fair value changes of investment properties and other derivative financial instruments, the core loss attributable to the parent company is 1.7 billion yuan. 
Since 2022, Longfor has actively adjusted its pace and reduced its liabilities, with the proportion of development business continuing to decline. By 2025, the development business has recorded its first loss, but the operations and services business achieved core profit of nearly 8 billion yuan, offsetting the negative impact from the decline of the development business.
Longfor's Chairman and CEO Chen Xuping stated that it is expected that by 2028 at the latest, the revenue from Longfor's operations and services business will exceed that from real estate development, becoming the company's main source of income. At the same time, because the business structure of real estate companies is deeply integrated with the company's income, debt structure, cash flow sources and rhythms, and credit systems, the interrelationship between business structure and debt structure is causal. The adjusted revenue structure, which is primarily based on operational service income, will also mean that Longfor has completed the foundation construction for a new development model.

Longfor Performance Release Conference
01 
Safely passing the debt repayment peak
Debt safety is the first lifeline for companies in the real estate industry during a contraction period. During this period, Longfor has consistently adhered to the principle of "no overdue, no extension, no default" as the company's credit baseline.
According to Longfor's annual report data, by the end of 2025, its interest-bearing liabilities will be reduced to 152.8 billion yuan, with the average financing cost reduced to 3.51%, and the average loan term extended to over 12 years. Among the interest-bearing liabilities, bank financing accounts for nearly 90%, an increase of about 15 percentage points compared to the end of 2022, while the total amount and proportion of credit bonds continue to decline.
Longfor's Chief Financial Officer Zhao Yi stated that in 2025, the group will repay debts of approximately 22 billion yuan, including 13.5 billion yuan of domestic bonds and 9.2 billion Hong Kong dollars of overseas loans, with about 6 billion yuan of remaining due debt this year, including approximately 2.5 billion yuan of credit bonds, which have safely and smoothly passed the debt peak.
While reducing the total amount of debt, Longfor has also been continuously adjusting its debt structure. By the end of last year, Longfor's proportion of domestic loans rose to 83%, while the proportion of overseas credit loans dropped to 6%, and supply chain financing has basically been zeroed out. The new operating property loans added about 18 billion yuan that year, with a total exceeding 100 billion yuan, increasing its proportion to 66%.
Additionally, the balance of foreign currency liabilities among Longfor's interest-bearing debts has also continued to decrease to around 15 billion yuan, accounting for about 10% at the end of last year.
By the end of 2025, Longfor's asset-liability ratio after excluding advance payments will drop to 55%, with the net debt ratio remaining stable at around 52%, and the cash-to-convertible bond multiple after excluding pre-sale regulatory funds remaining above one time, indicating a relatively stable financial condition.
2025 is the peak year for Longfor's debt repayment, with the due amount of about 22 billion yuan, which will decline to about 6 billion yuan and 6.9 billion yuan in 2026 and 2027, respectively. It will then drop significantly again in 2029 to about 4 billion yuan.
Zhao Yi revealed that in the future, Longfor's financing management strategy will strictly adhere to three principles:

First, strictly maintain credit, ensuring that bank financing does not become overdue, does not extend, and does not default on public market debt;
Second, continuously improve the NPI of operational assets, enhance asset valuations, explore space for operating property loans, optimize financing structures, extend payment terms, while particularly controlling short-term and foreign currency debts;
Third, strengthen financing and cash management mechanisms, solidify and establish an account-level control system, and proactively reduce peaks in due debt.

In recent years, due to the deep adjustments in the industry and structural impacts on settlement projects, Longfor Group's gross profit margin has been continuously declining. However, the company's management expects that after two years of bottoming out in 2025 and 2026, with the dual decline of total debt and interest rates, the proportion of income from operations and services will continue to increase, and the gradual reduction of losses from the development business will lead to a recovery in the company's gross profit margin.
02 
The transition from old to new models is about to be completed
From Longfor Group's business revenue structure, it can be seen that the company's reliance on operations and services is becoming stronger.
Chen Xuping stated that in recent years, Longfor's real estate development business will still focus on inventory reduction, tackling hard stocks, and will not deliberately emphasize scale; the growth and stability of the company's performance will mainly rely on operational business.
From the data perspective, Longfor's revenue from development business has fallen below 100 billion yuan for the first time in 2025, while the revenue from operations and services has reached a new high.

Longfor's 2025 income structure
In 2025, Longfor's land acquisition remained relatively "conservative". Throughout the year, it newly acquired 7 projects, with a total construction area of approximately 380,000 square meters, an equity construction area of approximately 260,000 square meters, and a new value of approximately 8.2 billion yuan. Compared to the contract sales of 63.16 billion yuan, the land acquisition to sales ratio is only about 5.5%.
Longfor's Executive Director and General Manager of Real Estate Development Zhang Xuzhong stated that Longfor's strategy is to prioritize financial safety, with the priority of ensuring repayment higher than new investments. Currently, Longfor still has about 22 million square meters of land reserves, with a total unsold value exceeding 200 billion yuan, and the basic strategy is to match sales with expenditure and activate inventory, with the requirement that for every new project acquired, it must be successfully developed.
Compared to the "conservative" strategy of the development business, commercial and other channels have shown more proactive growth. In 2025, Longfor opened 13 new shopping centers, including 8 self-owned heavy asset projects and 5 light asset operation projects.
The rental income from shopping malls rose to 11.21 billion yuan in 2025, an increase of 4% year-on-year, with an overall occupancy rate of 96.8%.
By the end of the period, Longfor operated a total of 99 shopping malls, with a total built-up area of approximately 10.5 million square meters, covering 25 key cities, with total sales exceeding 82 billion yuan.

Large Companies:
Japan's ispace company has postponed the NASA-funded lunar mission to 2030
Japanese spacecraft startup ispace announced on Friday that after two failed lunar lander missions, the company is making strategic adjustments and will postpone a government-funded lunar exploration mission to 2030, while also cutting global staff. This announcement highlights the uncertain prospects for the company. Currently, the U.S. is reorganizing space missions in conjunction with commercial and international partners. (Sina Finance)
Cainiao Group signs strategic cooperation agreement with CP AXTRA under Thailand's Charoen Pokphand Group
36Kr learned that today, Cainiao Group signed a strategic cooperation agreement with CP AXTRA under Thailand's Charoen Pokphand Group. Cainiao will apply its digital supply chain technology, warehousing automation technology, and overseas warehousing network operational experience to fully support CP AXTRA in building an integrated online and offline instant retail logistics network in Thailand and ASEAN countries. CP AXTRA manages over 2,700 offline stores in Thailand and Asia.
Wanda Film: Plans to change its stock abbreviation to Ruyi Film
36Kr learned that Wanda Film announced its intention to change the company's name to Ruyi Film Entertainment Co., Ltd., change its stock abbreviation to Ruyi Film, and correspondingly revise the Articles of Association.
Alibaba's consumer platform drives growth in Fliggy car rentals, 88VIP order user base expands by 40%
36Kr learned that on March 27, according to disclosures from the Fliggy car business ecosystem partner conference, over the past fiscal year, Fliggy's car rental business has maintained strong growth, particularly benefiting from the resources, scenarios, and technology of Alibaba's consumer platform, with car rental bookings from Taobao increasing by nearly 60% year-on-year and the number of ordering users growing by 40%. The cross-selling of flight, hotel, and car services has reached new heights, with orders from users of the aforementioned categories accounting for over 50% of the total for pick-up and drop-off services.
Some Wahaha Hongsheng factories have halted production, and some factories have issued holiday notices
On March 27, a person close to Wahaha stated that Wahaha notified last night (March 26) that 70% of its factory production lines have been halted, including several factories from both Wahaha and Hongsheng. This notification was made verbally, and currently, except for essential orders, other shipments have been suspended, with some factories also issuing holiday notices, with production expected to resume around April 2. In addition, another person close to Wahaha revealed that this is due to production scheduling. (Jiemian)
Razer: More than half of brand transactions come from new products, IP co-branded new products debut exclusively on Tmall
On March 26, Razer's global vice president Chen Xiaoping stated at the 2026 Tmall TOP TALK that over 50% of transactions from our brand flagship store in 2025 came from new products, which are the core engine of business growth. Razer has consistently launched IP co-branded new products such as Pokémon, Sanrio, and BLACKPINK exclusively on Tmall. Recently, Razer's annual flagship new product "Venom V4 Professional Edition" debuted on Tmall and offered Tmall users exclusive rights such as game custom props and limited peripherals.
Tsinghua Unigroup's Zhenyue 510 chip shipment exceeds 500,000 units, with multiple storage companies integrating this chip into SSDs
36Kr learned that on March 27, at the 2026 CFMS Flash Memory Summit, Tsinghua Unigroup announced that the cumulative shipment of its SSD controller chip Zhenyue 510 has exceeded 500,000 units, making it one of the highest shipment controller chips in the country recently. Currently, Zhenyue 510 has been launched on a large scale in multiple core businesses of Alibaba Cloud, and several storage manufacturers such as Yiheng Chuangyuan, Derui Lingxin, Baiwei Storage, and Changjiang Wanrun have launched storage products based on this chip.
XPeng Motors: The company's Chinese name will change from "XPeng Motors Co., Ltd." to "XPeng Group"
36Kr learned that XPeng Motors announced that starting from April 1, 2026, the company's Chinese name (for identification purposes only) will change from "XPeng Motors Co., Ltd." to "XPeng Group." The company's English name will remain unchanged as "XPeng Inc." Starting from 9 AM on April 1, 2026, the Chinese stock abbreviation for the company's A-shares traded on the main board of the Hong Kong Stock Exchange will change from "XPeng Motors-W" to "XPeng Group-W."
Lloyds Bank IT failure leads to the leak of transaction data for nearly 500,000 customers
The UK Treasury Committee stated on Friday that earlier this month, Lloyds Banking Group experienced an IT failure that led to the leak of personal information for up to 447,936 customers, allowing users to view others' transaction records, including account details and national insurance numbers. (Sina Finance)
New Products:
Deep Intelligence releases a new embodied general intelligence large model
36Kr learned that on March 27, at the Zhongguancun Forum event, Deep Intelligence, the first embodied intelligent company incubated by Beijing Zhongguancun Academy and Zhongguancun Artificial Intelligence Research Institute, officially released the first PhysBrain 1.0 embodied general intelligence base model constructed based on human learning paradigms. PhysBrain 1.0 adopts a multimodal large model architecture, internalizing physical knowledge into parameters, allowing the model to possess temporal and spatial consistency in understanding the physical world, and achieving true generalization capability under limited data conditions.
Ninebot: Cooperation with Pop Mart reached, co-branded electric vehicle will launch in April
On March 27, Ninebot announced a brand cooperation with Pop Mart's popular IP Little Sweet Bean, focusing on young consumer groups and emotional value expression for deep co-creation, with plans to launch the first IP co-branded electric vehicle product in April. (Securities Times)
Orion Star launches AI Agent service robot
36Kr learned that on March 27, the "Real Utility Creates Real Business" themed Orion Star partner win-win conference and the launch of the Leopard Assistant series new products were held in Beijing. At the conference, Orion Star officially released two AI Agent voice interaction service robots—Leopard Assistant Lite and Leopard Assistant Pro—and launched an Agent store for robots.
Other news worth noting:
State Administration of Foreign Exchange: 2025

On March 27, Longfor Group released its annual performance report for 2025. The report shows that Longfor Group's operating income for 2025 is approximately 97.31 billion yuan. 
Among them, the income from real estate development business is 70.54 billion yuan; the income from operation and service business is 26.77 billion yuan, accounting for 27.5% of the group's operating income, reaching a historic high; the profit attributable to the company owners is 1.02 billion yuan; after excluding the impact of fair value changes in investment properties and other derivative financial instruments, the core loss attributable to the parent company is 1.7 billion yuan. 
Longfor actively adjusted its pace and reduced debt since 2022, with the proportion of development business continuing to decline. By 2025, the development business experienced its first loss, but the operation and service business realized a core profit of nearly 8 billion yuan, offsetting the negative impact of the decline in the development business.
Longfor's Chairman of the Board and CEO Chen Xuping stated that it is expected that by 2028 at the latest, the income from Longfor's operation and service business will exceed that from real estate development, becoming the company's main source of income. At the same time, because the business structure of real estate companies is deeply integrated with the company's income, debt structure, cash flow sources and rhythm, and credit system, the business structure and debt structure are mutually causal. The adjustment to an income structure mainly based on operation service income will also signify that Longfor has completed the foundation construction of a new development model.

Longfor Performance Release Conference
01 
Safely Pass the Debt Repayment Peak
Debt safety is the first lifeline for companies in the real estate industry during tightening periods. During this time, Longfor has consistently adhered to the credit bottom line of "no overdue, no extension, no default."
According to Longfor's annual report data, by the end of 2025, its interest-bearing liabilities will be reduced to 152.8 billion yuan, with an average financing cost reduced to 3.51%, and the average loan term extended to over 12 years. Among the interest-bearing liabilities, bank financing accounts for nearly 90%, an increase of about 15 percentage points compared to the end of 2022, while the total amount and proportion of credit bonds continue to decrease.
Longfor's Chief Financial Officer Zhao Yi stated that in 2025, the group will repay debts of approximately 22 billion yuan, including 13.5 billion yuan for domestic bonds and 9.2 billion Hong Kong dollars for overseas loans, with about 6 billion yuan of debt maturing remaining this year, including around 2.5 billion yuan in credit bonds, which has safely and steadily passed the debt peak.
While reducing total liabilities, Longfor is also continuously adjusting its debt structure. By the end of last year, Longfor's proportion of domestic loans rose to 83%, while the proportion of overseas credit loans fell to 6%, and supply chain financing was basically zeroed out, with new operating property loans of approximately 18 billion yuan added that year, exceeding 100 billion yuan, increasing the proportion to 66%.
Additionally, among Longfor's interest-bearing liabilities, the balance of foreign currency liabilities has continuously decreased to around 15 billion yuan, accounting for about 10% at the end of last year.
By the end of 2025, Longfor's asset-liability ratio after excluding advance payments will drop to 55%, with the net debt ratio stabilizing around 52%, and the cash conversion multiple after excluding pre-sale regulated funds maintained at above one time, indicating a relatively stable financial condition.
2025 is the peak year for Longfor's debt repayment, with a maturing amount of approximately 22 billion yuan, which will decrease to about 6 billion yuan and 6.9 billion yuan in 2026 and 2027, respectively. After that, it will drop sharply again in 2029, to about 4 billion yuan.
Zhao Yi revealed that in the future, Longfor's financing management strategy will strictly adhere to three principles:

First, strictly observe credit, with bank financing not overdue, not extended, and public market debts not defaulted;
Second, continuously improve the NPI of operating assets, enhance asset valuation, explore the space of operating property loans, optimize the financing structure, extend payment terms, while particularly controlling short-term debt and foreign currency debt;
Third, strengthen financing and cash management mechanisms, solidify and establish an account-level control system, and proactively manage peak maturing debts.

In recent years, affected by the deep adjustment of the industry and the structural impact of settlement projects, Longfor Group's gross profit margin has also been continuously declining. However, the company’s management expects that after a two-year bottoming process in 2025 and 2026, with the dual decrease of total liabilities and interest rates, the proportion of operating and service business income will continue to increase, and the gradual reduction of losses brought by the settlement rhythm of the development business will lead to a recovery in the company's gross profit margin.
02 
New and Old Model Transition Nearing Completion
From the structure of Longfor Group's business income, it can be seen that the company is increasingly dependent on its operation and service business.
Chen Xuping stated that in the coming years, Longfor's real estate development business will still focus on clearing existing inventory, tackling hard-to-sell stock, and will not deliberately emphasize scale. The growth and stability of the company's performance mainly rely on operational business.
From the data perspective, Longfor's development business income fell below 100 billion yuan for the first time in 2025, while the income from operation and service business reached a new high.

Longfor's 2025 Income Structure
In 2025, Longfor remained relatively "conservative" in land acquisition. Throughout the year, it newly acquired 7 projects, with a total construction area of about 380,000 square meters, an attributable construction area of about 260,000 square meters, and a newly added value of about 8.2 billion yuan. Compared to the contract sales of 63.16 billion yuan, the land acquisition sales ratio is only about 5.5%.
Longfor's executive director and general manager of real estate development Zhang Xuzhong stated that Longfor's strategy prioritizes financial safety, with the priority of ensuring repayment being higher than that of new investments. Currently, Longfor has approximately 22.2 million square meters of land reserves, with the overall unsold value exceeding 200 billion yuan, and the basic strategy is to spend according to sales and activate inventory, with the requirement for new projects being to successfully deliver what is acquired.
Compared to the "conservative" strategy of the development business, the commercial and other channel businesses show a more proactive performance in growth. In 2025, Longfor opened 13 new shopping centers, including 8 self-owned heavy asset projects and 5 light asset operation projects.
Shopping mall rental income rose to 11.21 billion yuan in 2025, an increase of 4% year-on-year, with an overall occupancy rate of 96.8%.
By the end of the reporting period, Longfor had a total of 99 shopping malls in operation, with a total opened construction area of about 10.5 million square meters, covering 25 key cities, and total revenue exceeding 82 billion yuan.

Large Companies:
Japanese company ispace has postponed the NASA-funded lunar mission to 2030
The Japanese spacecraft startup ispace announced on Friday that after two failed lunar lander missions, the company is making strategic adjustments, pushing back a NASA-funded lunar exploration mission to 2030 and reducing its global workforce. This announcement highlights the company's uncertain prospects. Currently, the U.S. is restructuring space missions in collaboration with commercial and international partners. (Sina Finance)
Cainiao Group signs strategic cooperation agreement with CP AXTRA under Thailand's Charoen Pokphand Group
36Kr learned that today, Cainiao Group signed a strategic cooperation agreement with CP AXTRA under Thailand's Charoen Pokphand Group. Cainiao will apply its digital supply chain technology, warehousing automation technology, and overseas warehousing network operational experience to fully support CP AXTRA in creating an online and offline integrated instant retail logistics network in Thailand and ASEAN countries. CP AXTRA manages over 2,700 offline stores in Thailand and Asia.
Wanda Films: plans to change its securities abbreviation to Ruyi Films
36Kr learned that Wanda Films announced its plan to change its company name to Ruyi Film Entertainment Co., Ltd. and to change its securities abbreviation to Ruyi Films, along with corresponding amendments to the company’s articles of association.
Alibaba's consumer platform drives growth in Fliggy car rentals, 88VIP order user base expands by 40%
36Kr learned that on March 27, at the Fliggy car business ecosystem partners conference, it was disclosed that in the past fiscal year, Fliggy's car rental business maintained strong growth, particularly due to the support from Alibaba's consumer platform resources, scenarios, and technology, with car rental bookings from Taobao increasing by nearly 60% year-on-year, and the number of ordering users growing by 40%. Cross-selling among flight tickets, hotels, and car business has reached a new level, with examples such as the pickup and drop-off service, where orders from the aforementioned categories account for over 50%.
Some Wahaha and Hongsheng factories have halted production, and some factories have issued vacation notices
On March 27, a person close to Wahaha stated that Wahaha notified on the night of March 26 that it had temporarily stopped 70% of its factory production lines, including multiple factories of Wahaha and Hongsheng. This notice was communicated verbally, and currently, the shipment from the factories has been paused except for necessary orders, and some factories have also issued vacation notices, with production expected to resume around April 2. Additionally, another person close to Wahaha revealed that this is due to production scheduling issues. (Interface)
Razer: More than half of brand transactions come from new products, IP co-branded new products debut exclusively on Tmall
On March 26, Razer's Global Vice President Chen Xiaoping stated at the 2026 Tmall TOP TALK that over 50% of transactions in our brand flagship store in 2025 came from new products, with new products being the core engine of business growth. Razer has consistently launched IP co-branded new products like Pokémon, Sanrio, and BLACKPINK exclusively on Tmall. Recently, Razer's annual flagship new product, the Viper V4 Professional Edition, was launched exclusively on Tmall, providing Tmall users with game customization items and exclusive peripheral rights.
T-head's Zhenyue 510 shipment exceeds 500,000 units, multiple storage companies have SSDs equipped with this chip
36Kr learned that on March 27, at the 2026 CFMS Flash Memory Summit, T-head Semiconductor announced that the cumulative shipment of its SSD controller chip Zhenyue 510 has exceeded 500,000 units, making it one of the highest shipment controller chips in the country recently. Currently, Zhenyue 510 has been launched at scale in several core businesses of Alibaba Cloud, with multiple storage manufacturers such as Yiheng Chuangyuan, Derui Lingxin, Baiwei Storage, and Changjiang Wanrun launching storage products based on this chip.
Xpeng Motors: the company's Chinese name will be changed from "Xpeng Motors Co., Ltd." to "Xpeng Group"
36Kr learned that Xpeng Motors announced that starting April 1, 2026, the company's Chinese name (for identification purposes only) will be changed from "Xpeng Motors Co., Ltd." to "Xpeng Group." The company’s English name remains unchanged as "XPeng Inc." Starting from 9:00 AM on April 1, 2026, the Chinese stock abbreviation of the company’s Class A ordinary shares traded on the main board of the Hong Kong Stock Exchange will change from "Xpeng Motors–W" to "Xpeng Group–W."
Lloyds Bank IT failure leads to the leakage of transaction data of nearly 500,000 customers
The UK Treasury Committee stated on Friday that earlier this month, Lloyds Banking Group experienced an IT failure that led to the leakage of personal information of up to 447,936 customers, allowing users to view others' transaction records, including account details and national insurance numbers. (Sina Finance)
New Products:
Deep Intelligence releases new embodied general intelligence model
36Kr learned that on March 27, during the Zhongguancun Forum event, Deep Intelligence, the first embodied intelligent enterprise incubated by Beijing Zhongguancun Academy and Zhongguancun Artificial Intelligence Research Institute, officially launched the first PhysBrain 1.0 embodied general intelligence base model built on human learning paradigms. PhysBrain 1.0 adopts a multimodal large model architecture, internalizing physical knowledge into parameters, allowing the model to have temporal and spatial consistency in understanding the physical world, achieving true generalization capability with limited data.
Ninebot: Partnership with Pop Mart to launch co-branded electric vehicle in April
On March 27, Ninebot announced a brand cooperation with Pop Mart's popular IP Xiaotian Dou, aiming to co-create around young consumer groups and emotional value expression, planning to launch the first IP co-branded electric vehicle product in April. (Securities Times)
Orion Star launches AI Agent service robot
36Kr learned that on March 27, the "Only Useful Has Real Business" themed Orion Star Partners Win-Win Conference and the launch event for the Leopard Xiaomi series of new products was held in Beijing. At the conference, Orion Star officially released two AI Agent voice interaction service robots—Leopard Xiaomi Lite and Leopard Xiaomi Pro, and launched a robot Agent store.
Other news worth noting:
State Administration of Foreign Exchange: 2025

On March 27, Longfor Group released its 2025 annual performance report. The report shows that Longfor Group's operating revenue for 2025 is approximately 97.31 billion yuan.  
Among them, the revenue from real estate development business is 70.54 billion yuan; the revenue from operation and service business is 26.77 billion yuan, accounting for 27.5% of the group's operating revenue, reaching a historical high; the profit attributable to shareholders of the company is 1.02 billion yuan; after excluding the impact of fair value changes of investment properties and other derivative financial instruments, the core loss attributable to the parent company is 1.7 billion yuan.  
Longfor has been actively adjusting its pace and reducing debt since 2022, with the proportion of development business continuously declining. By 2025, the development business experienced its first profit loss, but the operating and service business achieved a profit close to 8 billion yuan after core equity, offsetting the negative impact of the decline in the development business.
Longfor's Chairman of the Board and CEO Chen Xuping stated that it is expected that by 2028 at the latest, the revenue from Longfor's operation and service business will exceed that of real estate development, becoming the company's main source of revenue. At the same time, because the business structure of real estate companies is deeply integrated with the company's revenue, debt structure, cash flow sources and pace, and credit system, the business structure and debt structure are mutually causal. The adjusted income structure, which is primarily based on operating service revenue, will also mean that Longfor has completed the foundation construction of the new development model.

Longfor Performance Press Conference
01 
Safely Passing the Debt Repayment Peak
Debt safety is the first lifeline for companies in the real estate industry during a tightening period. During this time, Longfor has consistently adhered to the credit bottom line of 'no overdue, no extension, no default'.
According to Longfor's annual report data, by the end of 2025, its interest-bearing debt will be reduced to 152.8 billion yuan, with an average financing cost reduced to 3.51%, and the average loan term extended to over 12 years. Among interest-bearing debts, bank financing accounts for nearly 90%, an increase of about 15 percentage points compared to the end of 2022, while the total amount and proportion of credit bonds continue to decrease.
Longfor's Chief Financial Officer Zhao Yi stated that in 2025, the group will repay debts of approximately 22 billion yuan, including 13.5 billion yuan of domestic bonds and 9.2 billion Hong Kong dollars of overseas loans, with about 6 billion yuan of remaining debts maturing this year, of which approximately 2.5 billion yuan is credit bonds, and it has safely and smoothly passed the peak of debt.
While reducing the total amount of debt, Longfor has also been continuously adjusting its debt structure. By the end of last year, the proportion of Longfor's domestic loans rose to 83%, the proportion of overseas credit loans decreased to 6%, and supply chain financing was basically zeroed out. Operating property loans increased by approximately 18 billion yuan in the year, with a total exceeding 100 billion yuan, increasing the proportion to 66%.
Additionally, among Longfor's interest-bearing debts, the balance of foreign currency debts has also continued to decrease to about 15 billion yuan, accounting for about 10% at the end of last year.
By the end of 2025, Longfor's asset-liability ratio, excluding advance payments, will decrease to 55%, and the net debt ratio will remain stable at around 52%. Excluding pre-sale regulatory funds, the cash-to-debt ratio will remain above one, indicating a relatively stable financial situation.
2025 is the peak year for Longfor's debt repayment, with a maturing amount of about 22 billion yuan, which will decline to about 6 billion yuan and 6.9 billion yuan in 2026 and 2027, respectively. It will then decrease significantly again in 2029 to about 4 billion yuan.
Zhao Yi revealed that Longfor's future financing management strategy will strictly adhere to three principles:

First, strictly adhere to credit, with bank financing not overdue, no extensions, and no defaults on public market debts;
Second, continuously improve the NPI of operating assets, enhance asset valuations, explore the space for operating property loans, optimize financing structures, extend payment terms, while particularly controlling short-term debts and foreign currency debts;
Third, strengthen the financing and cash management mechanisms, solidify and establish account-level control systems, and preemptively reduce peak maturing debts.

In the past few years, affected by the deep adjustment of the industry and the structural impact of settlement projects, Longfor Group's gross profit margin has also been continuously declining. However, the company’s management expects that after two years of bottoming in 2025 and 2026, with the dual decrease in total debt and interest rates, the proportion of revenue from operation and service business will continue to increase, and the gradual reduction in losses from development business settlements will lead to a recovery in the company’s gross profit margin.
02 
The Transition from Old to New Model is Approaching Completion
From Longfor Group's business revenue structure, it can be seen that the company is increasingly reliant on operation and service businesses.
Chen Xuping stated that in the coming years, Longfor's real estate development business will still focus on the disposal of existing inventory, tackling the hard bones of inventory, and will not deliberately emphasize scale. The growth and stability of the company's performance mainly rely on operating businesses.
From the data perspective, Longfor's development business revenue fell below 100 billion yuan for the first time in 2025, while the revenue from operation and service business reached a new high.

Longfor's 2025 Revenue Structure
In 2025, Longfor's land acquisition remained relatively 'conservative'. Throughout the year, seven new projects were acquired, with a total construction area of approximately 380,000 square meters, an equity construction area of about 260,000 square meters, and a new value of about 8.2 billion yuan. Compared to the contract sales of 63.16 billion yuan, the land acquisition sales ratio is only about 5.5%.
Longfor's Executive Director and General Manager of Real Estate Development Channel Zhang Xuzhong stated that Longfor's strategy prioritizes financial safety, placing a higher priority on guaranteed payments than on new investments. Currently, Longfor still holds approximately 22 million square meters of land reserves, with a total unsold value exceeding 200 billion yuan. The basic strategy is to determine expenditures based on sales and activate existing inventory, with the requirement for new projects being that each project must be successfully completed.
Compared to the 'conservative' strategy of the development business, the commercial and other channel businesses show a more positive performance in growth. In 2025, Longfor opened 13 shopping centers, including eight self-owned heavy asset projects and five light asset operation projects.
Rental income from malls rose to 11.21 billion yuan in 2025, a year-on-year increase of 4%, with an overall occupancy rate of 96.8%.
By the end of the period, Longfor had a total of 99 malls in operation, with a total construction area of about 10.5 million square meters, covering 25 key cities, and total operating revenue exceeding 82 billion yuan.

Text | Wang Xinyi
Editor | Wang Yuchan
One-sentence introduction
StoReel is a global AI-native entertainment content platform, with core products being the AI series platform StoReel App and the AI series tool community StoReel Canvas, dedicated to creating, distributing, and monetizing serialized series content.
The company was transformed from the overseas short drama market and began systematically exploring the application of AI in short drama production and content creation from the end of 2024 to the beginning of 2025.
Financing progress
Recently, StoReel has obtained $34 million in growth funding support, this round of financing includes a $9 million Seed round equity financing led by Play Ventures, with T-Accelerate Capital, Tirta Ventures, and VRF participating, as well as $25 million in user growth financing from PVX Partners.
According to StoReel, this funding will be used to enhance AI content production capabilities, expand the creator ecosystem, and accelerate global user growth.
Products and business
StoReel's two core products are StoReel App and StoReel Canvas.
StoReel App is an AI short drama content platform aimed at overseas markets, with over 50 AI short dramas launched, 80% of which come from the PUGC creator ecosystem. These short dramas uniformly adopt a realistic human visual style and are fully generated by AI.

△ Some self-produced drama posters, image source: provided by the company
Before fully transitioning to the AI track, StoReel had long-term plans in the overseas short drama market, accumulating mature content production and distribution capabilities. Currently, StoReel App centers on AI-native content, continuously launching multi-type series content, and constantly optimizing content production and user experience through AI technology.
Users can not only "watch dramas" on the platform but also participate in content evolution by voting on plot directions and key events, engaging in multidimensional interaction with character IPs, gradually extending from "watching" to "participatory experience."
On the product level, users can interact with AI characters through text and gradually unlock multimodal content such as voice, images, and videos, while incorporating various interactive forms like mini-games and immersive mini-theaters to enhance the connection between users and the plot and characters.
In addition, StoReel personalizes the operation of some core characters, constructing an independent character content matrix, continuously extending and interacting around plot development, enhancing user stickiness and IP value.
Meanwhile, StoReel has developed a one-stop AI film and television creation tool community, StoReel Canvas, as the core production platform for the internal AI director team and creator ecosystem.
Its tools integrate various mainstream generative models, building a full-process production workflow covering storyboard design, character and scene asset management, image and video generation, project collaboration, and publishing, achieving an integrated production capability from concept to finished product.

△ StoReel Canvas, image source: provided by the company
In terms of specific capabilities, StoReel Canvas can meet film-level production needs, including ultra-high-definition and super-resolution processing, character consistency control, lip-sync generation, lighting and visual style optimization, and significantly reduce repetitive production costs through templated processes and asset reuse mechanisms, notably improving content production efficiency and stability.
Overall, StoReel Canvas can not only serve as a generation tool but also as an AI production system that supports scalable, standardized content production, providing underlying capability support for high-frequency, high-quality content supply.
The series creation at StoReel is not a one-off but an iterative creative model. On the series content side, the platform has incorporated plot voting gameplay, allowing users to participate in voting to influence the direction of the plot and the content development of the script.
If certain episodes' plots are found unsuitable, creators can make adjustments to the script at any time and continue with content production.
From a time perspective, it may take only a few weeks from the completion of user voting to the completion of series creation.
StoReel's goal is to achieve a monthly production of 100 AI short dramas by the end of 2026, which includes both internally produced content and content provided by the PUGC creator ecosystem.

△ Script voting feature on StoReel App, image source: provided by the company
Compared to live-action short dramas, AI short dramas also perform well in data metrics. From the unit economic model of a single piece of content, the ROI performance of AI series continues to improve, and high-quality content can achieve stable positive returns, while also having a stronger efficiency advantage in customer acquisition costs.
StoReel's main revenue model is subscription-based. In the North American market, the subscription price for weekly membership is $19.9 to $29.9, and $269 for annual membership, with slight price adjustments based on different regions and devices. Some series will also adopt periodic free viewing or ad-unlocking models.
Additionally, StoReel is exploring monetization possibilities in gamification and interactive gameplay, such as virtual gifts, character-related in-app purchases, and brand collaborations.
Currently, StoReel primarily focuses on the English content market, mainly targeting users in North America, the UK, Canada, and other English-speaking countries, while also covering markets in Japan, South Korea, Southeast Asia, and Latin America.
It is worth mentioning that the community ecosystem version of StoReel Canvas will be launched in mid-April and enter internal testing. Based on providing a complete AI creation tool, the platform will further open task matching and collaboration capabilities, allowing creators to select scripts provided by StoReel for production, which will then be uniformly distributed and monetized by the platform, forming a closed loop from creation to distribution.
Meanwhile, the platform will gradually go live.

 
The biggest bottleneck is not in the algorithm, but in the hardware's response capability.

 
Written by |  Barley
Edited by | Weiwei
Cover image source |  Provided by interviewee
Jingzhi Technology seems to be the "king of anti-consensus" in the humanoid robot industry.
This company, founded in May 2024, has made a series of choices that are completely opposite to the industry mainstream in less than two years —
While others are pursuing smarter robots, they focus on making robots faster;
Small humanoids favored by capital and shipped quickly, but they insist on making full-sized robots;
Knowing that choosing the VLA (Vision-Language-Action model) route would yield better financing and point more directly to the endgame, they opted for remote control;
While others are eager to send robots into factories and homes, they want to place robots on the racetrack first — the real kind of sports track.
The results of these anti-consensus choices currently seem quite explosive.
In November 2025, Jingzhi's quadruped robot Black Panther II was invited to race against Paris Olympic 100m champion Noah Lyles, and on the 50m track, Lyles was almost chased past the finish line by Black Panther. "If it were 10 more meters, Black Panther could win (Lyles)," said Jin Yongbin, co-founder and CTO of Jingzhi Technology. The video of this sprint battle between man and robotic dog has been viewed over hundreds of millions of times online.

 
Just a month later, the speed of Black Panther II increased from about 11m/s during the competition to 13.4m/s, breaking Boston Dynamics' 12-year-old world record for quadruped robot running speed — faster than the top speed of new national standard electric vehicles after secretly removing speed limits (around 40km/h).
In February 2026, Jingzhi released its first full-sized humanoid robot, Bolt, standing 175cm tall and weighing 75kg, with a peak treadmill speed of 10m/s, making it the fastest full-sized humanoid robot in the world.
This clearly attracted industry attention, and on March 17 at the Yabuli Forum, Yushu founder Wang Xingxing mentioned that by mid-year, someone would run the 100m humanoid robot in under 10 seconds, with the core argument being Bolt's performance.
According to the two main factions of humanoid robots today, this type of fast-performing robot generally doesn't look good enough — most joints are bulky, motors are exposed, and they don't look much like humans; while another type that looks human still lags behind in performance compared to humans.
However, Jingzhi's Bolt is one of the most humanoid among functional robots. The overall lines are smooth, with no extra components exposed, and the feet are carefully designed, featuring a clearly bendable sole, rather than a flat plate that needs shoes to reduce friction. According to Jin Yongbin's explanation, this additional degree of freedom can alleviate impact while running at high speeds, while also achieving greater strength.

 
Bolt
While most of the industry focuses on "the brain" of robots, Jingzhi is focusing on "the body."
According to Jin Yongbin's judgment, with the progress of AI algorithms, the control ability of robots has approached the limits of the hardware itself; the real bottleneck is not that the brain is not smart enough, but that the robot's body is not functional enough.
Born in 1995, Jin Yongbin has been studying mechanics at Zhejiang University since his undergraduate days, and began researching high-speed legged robots after his PhD in 2017.
“When I started researching robots 10 years ago, they asked why they should fund you? With so much funding, the performance of a robot is still worse than that of a hunting dog.” This is a fundamental confusion that has been stuck in Jin Yongbin's heart, “The muscle power density of humans is about 300W/kg, while motors can reach 3000W/kg or even higher, which is an order of magnitude better than biological performance, yet the overall performance of robots still lags behind biology.”
Therefore, Jin Yongbin's solution is to push hardware capabilities to their limits with speed, correcting mechanical shortcomings in the most efficient exposure. So that the physical abilities of the robot match the intelligence it will have to carry in the future.
As Wang Xingxing said, “This breakthrough in physical limits is not only a display of speed but also a sign of the maturity of robot hardware performance and control algorithms.”
Starting in 2023, Jin Yongbin found that robots are gradually being integrated into various industries, and more and more open-source algorithms and hardware are rapidly catching up with their technical advantages in the laboratory. Therefore, in May of the following year, he co-founded Jingzhi Technology with his mentor, Wang Hongtao, executive director of Zhejiang University's Interdisciplinary Mechanics Center, focusing on AI-powered high-mobility legged robot motion research.

 
Jin Yongbin with quadruped robot Black Panther
A thoroughly academic founding team, according to last October's article "Sorry, No More Investing in Professors' Startups" in the venture capital circle, this type of founder is no longer favored by capital. Their biggest problem is their lack of interest in commercialization — obsessed with technology, neglecting the market, and never completing the journey from laboratory to shelf.
However, the eggs Jingzhi has laid along the way may be more than many people think. The drive technology honed in the pursuit of extreme speed has already been applied to the industrial quadruped robot Apollo — capable of carrying 70kg and running at 8m/s, it has set a Guinness record and is already on sale. There is also a wheeled and legged dual-mode robotic dog with a screen that can function as a smart speaker, which is now ready for mass production and was officially released at the recently concluded AWE.
We are uncertain whether Jingzhi's path will ultimately succeed, but on an increasingly homogenized track, a completely different answer is at least worth serious consideration. Below is a conversation between "Future Human Laboratory" and Jin Yongbin in mid-February, organized and published —

 
Fast, not just speed
Future: Why must it run fast?
Jin Yongbin: Fast is far more than just speed; fast is a comprehensive ability of strength, density, intensity, smoothness, and so on.
When we buy a car, we usually care about the 0 to 100 km/h acceleration time; the shorter the time, the better the car's structure, and the engine.

Text | Wang Xinyi
Editor | Wang Yuchan
One-sentence introduction
StoReel is a global AI-native entertainment content platform, with core products being the AI episodic platform StoReel App and the AI episodic tool community StoReel Canvas, dedicated to creating, distributing, and monetizing serialized episodic content.
The company was transformed from the overseas market short drama track and has systematically explored the application of AI in short drama production and content creation since the end of 2024 to early 2025.
Financing progress
Recently, StoReel has secured $34 million in growth funding, with this round of financing including a $9 million Seed round equity financing led by Play Ventures, joined by T-Accelerate Capital, Tirta Ventures, and VRF, as well as $25 million in user growth financing from PVX Partners.
According to StoReel, these funds will be used to enhance AI content production capabilities, expand the creator ecosystem, and accelerate global user growth.
Products and business
StoReel's two core products are StoReel App and StoReel Canvas.
StoReel App is an AI short drama content platform aimed at the overseas market, with over 50 AI short drama contents online, of which 80% come from the PUGC creator ecosystem. These short dramas uniformly adopt a realistic human visual style, fully generated by AI.

△ Some self-produced series posters, source: provided by the company
Before fully shifting to the AI track, StoReel has long been laid out in the overseas short drama market, accumulating mature content production and distribution capabilities. Currently, StoReel App centers on AI-native content, continuously launching various types of episodic content, and continuously optimizing content production and user experience through AI technology.
Users can not only “watch dramas” on the platform but also participate in content evolution by voting on plot directions and key plot points, interacting with character IP in multiple dimensions, gradually extending from “watching” to “participatory experience.”
At the product level, users can interact with AI characters through text and gradually unlock multimodal content such as voice, images, and videos, while incorporating various interactive forms like mini-games and immersive small theaters to enhance connections between users, plots, and characters.
Additionally, StoReel operates some core characters with personalization, building an independent character content matrix that continuously extends content and interactive operations around plot developments, enhancing user stickiness and IP value.
At the same time, StoReel has developed a one-stop AI film and television creation tool community, StoReel Canvas, as the core production platform for the internal AI director team and creator ecosystem.
Its tools integrate various mainstream generative models, constructing a full-process production workflow covering storyboard design, character and scene asset management, image generation, video generation, project collaboration, and release, achieving integrated production capabilities from creativity to finished products.

△ StoReel Canvas, source: provided by the company
In terms of specific capabilities, StoReel Canvas can meet various film-level production needs including ultra-clear and super-resolution processing, character consistency control, lip-sync generation, lighting and visual style tuning, etc., and significantly reduce repetitive production costs through templated processes and asset reuse mechanisms, greatly improving content production efficiency and stability.
Overall, StoReel Canvas can not only serve as a generative tool but also as an AI production system supporting large-scale and standardized content production, providing underlying capabilities for high-frequency, high-quality content supply.
StoReel's episodic creation is not a one-time process but an iterative creation model. On the episodic content side, the platform has introduced a plot voting feature, allowing users to participate in voting to influence the direction of plot development and the content development of the script.
If certain episodes' plots are found to be unsuitable, creators can adjust the script at any time and continue content production.
In terms of time cycle, it may only take a few weeks from the completion of user voting to the completion of episodic creation.
StoReel aims to achieve a monthly output of 100 AI short dramas by the end of 2026, which includes content internally produced by the platform as well as content provided by the PUGC creator ecosystem.

△ Script voting feature on StoReel App, source: provided by the company
Compared to live-action short dramas, AI short dramas also perform well in data metrics. From the unit economic model of a single piece of content, the ROI performance of AI episodic content continues to improve, with quality content able to achieve stable positive returns, while also having a stronger efficiency advantage in user acquisition costs.
StoReel's main revenue model is subscription-based. In the North American market, the membership subscription price ranges from $19.9 to $29.9 for weekly memberships and $269 for annual memberships, with slight price adjustments for different regions and models. Some series may also adopt a model of periodically free viewing or ad unlocking.
In addition, StoReel is also exploring monetization possibilities in gamification and interactive gameplay, such as virtual gifts, character-related in-app purchase revenues, and brand collaborations.
Currently, StoReel mainly targets the English content market, primarily catering to users in English-speaking countries such as North America, the UK, and Canada, while also covering markets in Japan, South Korea, Southeast Asia, and Latin America.
Notably, the community ecological version of StoReel Canvas will be launched in mid-April and enter beta testing. On the basis of providing complete AI creation tools, the platform will further open task matching and creation collaboration capabilities, allowing creators to select scripts provided by StoReel for production, which will then be uniformly distributed and monetized by the platform, forming a closed loop from creation to distribution.
At the same time, the platform will gradually launch

 
The biggest bottleneck is not in the algorithm, but in the hardware's response capability.

 
Written by |  Ba Rui
Edited by | Wei Wei Zi
Cover image source |  Provided by the interviewee
Jingzhi Technology seems to be the "anti-consensus king" in the humanoid robot industry.
This company, established in May 2024, has made a series of choices that are completely contrary to the industry mainstream in less than two years—
While others are pursuing smarter robots, they are focused on making robots run faster;
Mini humanoids are favored by capital and sell quickly, but they insist on making full-sized ones;
Knowing that choosing the VLA (Vision-Language-Action model) route would enable better financing and lead to a clearer endgame, they chose remote operation;
While others are rushing to send robots into factories and homes, they want to first send robots onto the racetrack—truly on that kind of sports track.
And the results of these anti-consensus choices seem quite explosive at the moment.
In November 2025, Jingzhi's quadruped robot Black Panther II was invited to race with Noah Lyles, the 100-meter champion of the Paris Olympics, and on the 50-meter track, Lyles was almost chased across the finish line by Black Panther. "Another 10 meters, and Black Panther could have beaten (Lyles)," said Jin Yongbin, co-founder and CTO of Jingzhi Technology. This video of the sprint battle between humans and robotic dogs has already garnered over 100 million views online.

 
Just a month later, the speed of Black Panther II increased from around 11m/s during the competition to 13.4m/s, breaking the world record for quadruped robot running speed that had been sealed for 12 years by Boston Dynamics—faster than the top speed of new national standard electric vehicles after secretly lifting the speed limit (around 40km/h).
In February 2026, Jingzhi released the first full-sized humanoid robot Bolt, standing 175cm tall and weighing 75kg, with a peak speed on the treadmill of 10m/s, making it the fastest full-sized humanoid robot in the world today.
This clearly attracted attention in the industry; on March 17 at the Yabuli Forum, Yu Shu founder Wang Xingxing mentioned that by mid-year, a humanoid robot would run the 100 meters in under 10 seconds, with the core argument being Bolt's performance.
According to the two main factions in humanoid robotics today, these fast and high-performance robots generally do not look good enough—most have oversized joints, exposed motors, and look very unhuman; while another type that resembles humans still lags behind in performance compared to humans.
However, Jingzhi's Bolt is considered one of the most human-like among functionally good robots. The overall lines are smooth, with no extra components exposed, and the feet are also carefully designed, featuring a distinctly bendable sole, rather than a flat board that requires shoes to reduce friction. According to Jin Yongbin's explanation, this additional degree of freedom can alleviate impact while running at high speed and gain greater strength.

 
Bolt
While most of the industry is focusing on the "brain" of robots, Jingzhi is focusing on the "body".
According to Jin Yongbin's assessment, with the advancement of AI algorithms, the control capabilities of robots are approaching the limits of the hardware itself; the real bottleneck is not that the brain isn't smart enough, but that the robot's body isn't practical enough.
Born in 1995, Jin Yongbin has been studying mechanics at Zhejiang University since his undergraduate years, and after obtaining his PhD in 2017, he began researching high-speed legged robots.
"When I started researching robots 10 years ago, they asked, why should we fund you? So much funding for a robot that performs worse than a hunting dog." This is a fundamental confusion that has stuck with Jin Yongbin, "The muscle power density of humans is about 300W/kg, while motors can reach 3000W/kg or even higher, which is an order of magnitude better than biological performance, yet the overall performance of robots is still inferior to biology."
Therefore, Jin Yongbin's solution is to push hardware capabilities to the limit with speed, correcting mechanical shortcomings in the most efficient way possible. This allows the physical capabilities of robots to match the intelligence they are expected to carry in the future.
As Wang Xingxing stated, "This breakthrough in physical limits is not only a demonstration of speed but also a sign of the maturity of robot hardware performance and control algorithms."
Starting in 2023, Jin Yongbin found that robots are gradually landing in various industries, and that more and more open-source algorithms and hardware are rapidly catching up to their technological advantages in the laboratory. In May of the following year, he co-founded Jingzhi Technology with his teacher, Wang Hongtao, executive director of the Zhejiang University Interdisciplinary Mechanics Center, focusing on AI-powered high-mobility legged robot motion research.

 
Jin Yongbin with the quadruped robot Black Panther
A thoroughly academic founding team, according to last October's viral article "Sorry, I won't invest in professor startups" in the venture capital circle, this type of founder is no longer favored by capital. Their biggest problem is their lack of interest in commercialization—obsessed with technology, neglecting the market, and never completing the road from laboratory to shelf.
However, Jingzhi's eggs along the way may be more than many people think. The driving technology honed in the pursuit of extreme speed has already been transferred to the industrial quadruped robot Apollo—capable of carrying 70kg, running at 8m/s, has set a Guinness record, and is already on sale. There is also a wheeled bipedal robot dog with a screen that can function as a smart speaker, which is now ready for mass production and was officially launched at the recently concluded AWE.
We are uncertain whether Jingzhi's route will ultimately succeed, but on a racetrack where homogenization is becoming increasingly severe, a completely different answer is at least worth listening to once. Below is the dialogue between "Future Human Laboratory" and Jin Yongbin in mid-February, organized and published—

 
Fast is not just speed
Future: Why is speed so important?
Jin Yongbin: Fast is far more than just speed; it encompasses a series of capabilities including strength, density, intensity, and fluidity.

Text | Wang Xinyi
Editor | Wang Yuchan
One-sentence introduction
StoReel is a global AI-native entertainment content platform, with core products being the AI drama platform StoReel App and the AI drama tool community StoReel Canvas, dedicated to creating, distributing, and monetizing serialized drama content.
The company has transformed from the overseas market short drama track and has systematically explored the application of AI in short drama production and content creation from the end of 2024 to the beginning of 2025.
Financing progress
Recently, StoReel has secured $34 million in growth funding, this round of financing includes $9 million in Seed round equity financing led by Play Ventures, with participation from T-Accelerate Capital, Tirta Ventures, and VRF, as well as $25 million in user growth financing from PVX Partners.
According to StoReel, this funding will be used to enhance AI content production capabilities, expand the creator ecosystem, and accelerate global user growth.
Products and business
StoReel's two core products are StoReel App and StoReel Canvas.
StoReel App is an AI short drama content platform aimed at the overseas market, having launched over 50 AI short drama contents, of which 80% come from PUGC creator ecosystems. These short dramas uniformly adopt a photorealistic visual style, generated entirely by AI.

△Some self-produced drama posters, source: provided by the company
Before fully transitioning to the AI track, StoReel had long been laid out in the overseas short drama market, accumulating mature content production and distribution capabilities. Currently, StoReel App focuses on AI-native content, continuously launching various types of drama content, while continuously optimizing content production and user experience through AI technology.
Users can not only "watch dramas" on the platform but also participate in content evolution by voting on plot directions and key scenes, engaging in multidimensional interactions with character IPs, gradually extending from "viewing" to "participatory experience."
On the product level, users can interact with AI characters through text and gradually unlock multimodal content such as voice, images, and videos, while incorporating various interactive forms like mini-games and immersive mini-theaters to enhance the connection between users and the plot and characters.
In addition, StoReel operates some core characters in a personalized manner, building an independent character content matrix, conducting continuous content extension and interactive operations around plot development, enhancing user stickiness and IP value.
Meanwhile, StoReel has developed a one-stop AI film and television creation tool community StoReel Canvas, serving as the core production platform for the internal AI director team and creator ecosystem.
Its tools integrate various mainstream generation models, constructing a full-process production workflow covering storyboard design, character and scene asset management, image generation and video generation, project collaboration, and publishing, achieving integrated production capability from concept to finished product.

△StoReel Canvas, source: provided by the company
In terms of specific capabilities, StoReel Canvas can meet film-level production needs, including ultra-high definition and ultra-resolution processing, character consistency control, lip-sync generation, lighting and image style tuning, and significantly reducing repetitive production costs through templated processes and asset reuse mechanisms, greatly improving content production efficiency and stability.
Overall, StoReel Canvas can not only serve as a generation tool but also as an AI production system supporting large-scale and standardized content production, providing underlying capability support for high-frequency, high-quality content supply.
StoReel's drama creation is not a one-time event but an iterative creation model. On the drama content end, the platform has introduced plot voting gameplay, allowing users to participate in voting to influence the direction of the plot and the development of the script content.
If certain episodes are found to be unsuitable, creators can adjust the script at any time and continue content production.
In terms of time cycle, it may take only a few weeks from the completion of user voting to the completion of drama creation.
StoReel's goal is to achieve a monthly output of 100 AI short dramas by the end of 2026, which includes both internally produced content and content provided by the PUGC creator ecosystem.

△The script voting function on StoReel App, source: provided by the company
Compared to real-life short dramas, AI short dramas also perform well in terms of data. From the unit economic model of a single piece of content, the ROI performance of AI dramas continues to improve, and high-quality content can achieve stable positive returns, while having a stronger efficiency advantage in user acquisition costs.
StoReel's main revenue model is subscription-based. In the North American market, the membership subscription price ranges from $19.9 to $29.9 for weekly subscriptions, $269 for annual subscriptions, with slight price adjustments for different regions and devices. Some dramas will also adopt a model of regular free viewing or ad unlocking.
In addition, StoReel is also exploring monetization possibilities in gamification and interactive gameplay, such as virtual gifts, character-related in-app purchase revenues, and brand collaborations.
Currently, StoReel primarily focuses on the English content market, targeting users in English-speaking countries such as North America, the UK, and Canada, while also covering markets in Japan, South Korea, Southeast Asia, and Latin America.
It is worth mentioning that the community ecosystem version of StoReel Canvas will be launched in mid-April and will start internal testing. Based on providing complete AI creation tools, the platform will further open up task matching and collaborative creation capabilities, allowing creators to choose scripts provided by StoReel for production. After completion, the platform will unify distribution and monetization, forming a closed loop from creation to distribution.
Meanwhile, the platform will gradually roll out.

 
The biggest bottleneck is not in the algorithm, but in the hardware's responsiveness.

 
Written by |  Bai Rui
Edited by | Wei Wei Zi
Cover image source |  Provided by interviewee
Jingzhi Technology seems to be the "king of counter-consensus" in the humanoid robot industry.
Founded in May 2024, this company has made a series of choices that are completely contrary to the industry mainstream in less than two years—
While others are pursuing smarter robots, they focus on making robots run faster;
Smaller humanoid robots are favored by capital and sell quickly, but they insist on making full-sized ones;
Knowing that choosing the VLA (Vision-Language-Action model) route would allow for better financing and point more directly to the endgame, they chose remote operation;
While others are rushing to send robots into factories and homes, they want to send robots onto the racetrack first—truly, the kind of sports track.
The result of these counter-consensus choices seems quite explosive so far.
In November 2025, Jingzhi's quadruped robot Black Panther II was invited to race against Noah Lyles, the 100-meter champion of the Paris Olympics, on a 50-meter track, where Lyles was almost chased across the finish line by the Black Panther. "If it were 10 more meters, the Black Panther could win (against Lyles)," said Jin Yongbin, co-founder and CTO of Jingzhi Technology. This video of the short race between humans and the robotic dog has had over 100 million views online.

 
Just a month later, the speed of Black Panther II increased from around 11 m/s during the competition to 13.4 m/s, breaking the world record for quadruped robot running speed set by Boston Dynamics, which had been dormant for 12 years—faster than the top speed (around 40 km/h) of the new national standard electric vehicles after a speed limit was quietly lifted.
In February 2026, Jingzhi released its first full-sized humanoid robot, Bolt, which stands 175 cm tall, weighs 75 kg, and can reach a peak speed of 10 m/s on a treadmill, making it the fastest full-sized humanoid robot in the world.
This obviously caught the industry's attention; at the Yabuli Forum on March 17, Yu Shu founder Wang Xingxing mentioned that by mid-year, there would be humanoid robots capable of running 100 meters in under 10 seconds, with the core argument being based on Bolt's performance.
According to the two main schools of thought in humanoid robots today, this type of robot that runs fast and performs well generally does not look very appealing—most joints are enlarged, motors are exposed, and they look very little like humans; while another type that resembles humans still has performance gaps compared to humans.
However, Jingzhi's Bolt is one of the better-functioning robots that closely resembles a human. Its overall lines are smooth, with no extra components exposed, and the feet have been carefully designed with a clearly bendable sole, rather than a flat piece that needs shoes to reduce friction. According to Jin Yongbin's explanation, this added degree of freedom can alleviate impact while running at high speeds and gain greater strength.

 
Bolt
While most of the industry is focusing on the "brains" of robots, Jingzhi is focusing on the "bodies."
According to Jin Yongbin's judgment, as AI algorithms improve, the control capability of robots has approached the limits of the hardware itself; the real bottleneck is not that the brains are not smart enough, but that the bodies of the robots are not user-friendly enough.
Born in 1995, Jin Yongbin has been studying mechanics at Zhejiang University since his undergraduate years, and after pursuing his doctorate in 2017, he began researching high-speed legged robots.
"When I started researching robots 10 years ago, they asked why they should fund you? With so much funding, a robot's performance is still not as good as a hunting dog." This is one of the most primal confusions in Jin Yongbin's heart, "The power density of human muscles is about 300W/kg, while motors can reach 3000W/kg or even higher, which is an order of magnitude better than biological performance, but the overall performance of robots is still not as good as that of biological entities."
Therefore, Jin Yongbin's solution is to push the hardware capabilities to the limit with speed, correcting mechanical shortcomings in the most efficient way possible. This is to ensure that the physical abilities of the robots match the intelligence they are meant to carry in the future.
As Wang Xingxing said, "This breakthrough of physical limits is not only a display of speed but also a sign of the maturity of robotic hardware performance and control algorithms."
Starting in 2023, Jin Yongbin discovered that robots are gradually landing in various industries, and more and more open-source algorithms and hardware are rapidly catching up to their technological advantages in the laboratory. In May of the following year, he co-founded Jingzhi Technology with his teacher, Wang Hongtao, executive director of the Zhejiang University Interdisciplinary Mechanics Center, focusing on AI-empowered high-mobility legged robot motion research.

 
Jin Yongbin with quadruped robot Black Panther
A completely academic founding team, according to the venture capital perspective from the article "Sorry, No More Funding for Professors' Startups" that went viral last October, this type of founder is no longer favored by capital. Their biggest problem is their lack of interest in commercialization—obsessed with technology, neglecting the market, and forever unable to complete the journey from laboratory to shelf.
However, the eggs Jingzhi has laid along the way may be more than many people think. The driving technology honed in the pursuit of extreme speed has already been transferred to the industrial quadruped robot Apollo—capable of carrying 70 kg, running at 8 m/s, and has set a Guinness record, already on sale. There’s also a wheeled dual-form robot dog with a screen that can double as a smart speaker and is now in mass production, officially launched at the recently concluded AWE.
We are uncertain whether Jingzhi's route will ultimately be successful, but on a track where homogenization is becoming increasingly severe, a completely different answer is at least worth listening to seriously. Below is a conversation between the "Future Human Laboratory" and Jin Yongbin, organized and published in mid-February—

 
Fast, not just speed
Future: Why must it run fast?
Jin Yongbin: Fast is far more than just speed; it is a combination of strength, density, intensity, smoothness, and a series of other capabilities.
When we buy a car, we usually care about the acceleration time to 100 kilometers; the shorter the time, the better the structure of the car.

Text | Wang Xinyi
Editor | Wang Yuchan
One-sentence introduction
StoReel is a global AI-native entertainment content platform, whose core products are the AI series platform StoReel App and the AI series tool community StoReel Canvas, dedicated to building a creation, distribution, and monetization system for serialized series content.
The company was transformed from the overseas market short drama track and began systematically exploring the application of AI in short drama production and content creation from the end of 2024 to early 2025.
Financing progress
Recently, StoReel has secured $34 million in growth funding, which includes an $9 million Seed round equity financing led by Play Ventures, with participation from T-Accelerate Capital, Tirta Ventures, and VRF, along with $25 million in user growth financing from PVX Partners.
According to StoReel, this funding will be used to enhance AI content production capabilities, expand the creator ecosystem, and accelerate global user growth.
Products and business
StoReel's two core products are StoReel App and StoReel Canvas.
StoReel App is an AI short drama content platform aimed at overseas markets, which has launched over 50 AI short drama contents, 80% of which come from the PUGC creator ecosystem. These short dramas uniformly adopt a realistic human visual style, with the entire process generated by AI.

△ Some self-produced series posters, image source: provided by the company
Before fully transitioning to the AI track, StoReel had long been laying out in the overseas short drama market, accumulating mature content production and distribution capabilities. Currently, StoReel App focuses on AI-native content, continuously launching various types of series content, and constantly optimizing content production and user experience through AI technology.
Users can not only "watch dramas" on the platform but also participate in content evolution by voting on plot directions and key events, interacting with character IP in multiple dimensions, gradually extending from "watching" to "participatory experience."
On the product level, users can interact with AI characters through text and gradually unlock multimodal content such as voice, images, and videos, while enhancing the connection between users and the plot and characters through various interactive forms like mini-games and immersive theaters.
In addition, StoReel operates some core characters with personality, building an independent character content matrix, continuously extending and interacting around plot development to enhance user stickiness and IP value.
At the same time, StoReel has developed a one-stop AI film and television creation tool community, StoReel Canvas, as the core production platform for the internal AI director team and creator ecosystem.
Its tools integrate various mainstream generation models, constructing a full-process production workflow covering storyboard design, character and scene asset management, image and video generation, project collaboration, and publishing, achieving integrated production capabilities from creativity to finished product.

△ StoReel Canvas, image source: provided by the company
In terms of specific capabilities, StoReel Canvas can meet film-level production needs including ultra-high-definition processing, character consistency control, lip-sync generation, lighting, and visual style adjustments, while significantly reducing repetitive production costs through templated processes and asset reuse mechanisms, greatly improving content production efficiency and stability.
Overall, StoReel Canvas can not only serve as a generation tool but also as an AI production system supporting scalable and standardized content production, providing underlying capabilities for high-frequency, high-quality content supply.
The series creation of StoReel is not a one-time event but an iterative creation model. On the series content side, the platform has introduced a plot voting mechanism, allowing users to participate in voting to influence the direction of the plot and the development of the script content.
If certain episodes' plots are not suitable, creators can adjust the script at any time and continue content production.
In terms of time cycle, it may only take a few weeks from the completion of user voting to the completion of series creation.
StoReel's goal is to achieve a monthly output of 100 AI short dramas by the end of 2026, including both internally produced content and content provided by the PUGC creator ecosystem.

△ The script voting function on StoReel App, image source: provided by the company
Compared to live-action short dramas, AI short dramas perform well in terms of data. From the unit economic model of a single piece of content, the ROI of AI series has been continuously improving, with high-quality content achieving stable positive returns and demonstrating stronger efficiency advantages in user acquisition costs.
StoReel's main revenue model is subscription-based. In the North American market, membership subscription prices range from $19.9 to $29.9 for weekly subscriptions and $269 for annual subscriptions, with slight price adjustments based on different regions and device types. Some series also adopt periodic free viewing or ad-unlocking models.
Additionally, StoReel is exploring monetization possibilities in gamification and interactive gameplay, such as virtual gifts, character-related in-app purchases, and brand collaborations.
Currently, StoReel primarily focuses on the English content market, targeting users in North America, the UK, Canada, and other English-speaking countries, while also covering markets in Japan, Southeast Asia, and Latin America.
It is worth mentioning that the community ecosystem version of StoReel Canvas will go live in mid-April and begin internal testing. Based on providing a complete AI creation tool, the platform will further open up task matching and creation collaboration capabilities, allowing creators to choose scripts provided by StoReel for production, and after completion, the platform will uniformly distribute and monetize, forming a closed loop from creation to distribution.
At the same time, the platform will gradually launch more features.

 
The biggest bottleneck is not in the algorithm, but in the response capability of the hardware.

 
Written by |  Ba Rui
Edited by | Wei Wei Zi
Cover image source |  Provided by the interviewee
Jing Shi Technology seems to be the "King of Anti-Consensus" in the humanoid robot industry.
This company, founded in May 2024, has made a series of choices that are completely contrary to the mainstream of the industry in less than two years—
While others are pursuing making robots smarter, they focus on making robots run faster;
Small humanoid robots are favored by capital and are shipped quickly, but they insist on making full-size robots;
Knowing that choosing the VLA (Vision-Language-Action model) route would facilitate better financing and point more clearly to the endgame, they chose remote operation;
While others are rushing to send robots into factories and homes, they want to send robots onto the racetrack first— the kind of sports track in the true sense.
The results of these anti-consensus choices, so far, seem quite explosive.
In November 2025, Jing Shi's quadruped robot Black Panther II was invited to race against Noah Lyles, the 100-meter champion of the Paris Olympics, and on the 50-meter track, Lyles was almost chased across the finish line by Black Panther. "If it were another 10 meters, Black Panther could have beaten (Lyles)," said Jin Yongbin, co-founder and CTO of Jing Shi Technology. This short video of the sprint battle between man and robot dog has over a hundred million views online.

 
Just a month later, the speed of Black Panther II increased from around 11m/s during the competition to 13.4m/s, breaking Boston Dynamics' 12-year-old world record for quadruped robot running speed—faster than the top speed (around 40km/h) of the new national standard electric vehicle after secretly removing the speed limit.
In February 2026, Jing Shi released its first full-size humanoid robot, Bolt, which stands 175cm tall, weighs 75kg, and can reach a peak speed of 10m/s on a treadmill, making it the fastest full-size humanoid robot in the world.
This clearly attracted attention in the industry; on March 17 at the Yabuli Forum, Yu Shu founder Wang Xingxing mentioned that by mid-year, a humanoid robot would run the 100 meters in under 10 seconds, with the core argument being Bolt's performance.
According to the two main factions of humanoid robots today, this type of fast-performing robot generally doesn't look very attractive—most joints are enlarged, motors are exposed, and they don't resemble humans very much; while another type that looks human still has performance gaps compared to humans.
However, Jing Shi's Bolt is one of the more functional robots, with a body structure that resembles a human the most. The overall lines are smooth, with no extra components exposed, and the feet have been carefully designed, featuring a distinctly bendable sole instead of a flat piece that needs shoes to reduce friction. According to Jin Yongbin's explanation, this additional degree of freedom allows the robot to mitigate impact while running at high speeds and achieve greater strength.

 
Bolt
While most of the industry is focusing on the "brains" of robots, Jing Shi is focusing on the "body."
According to Jin Yongbin's judgment, with the advancement of AI algorithms, the control capabilities of robots have approached the limits of the hardware itself; the real bottleneck is not that the brain is not smart enough, but that the robot's body is not sufficiently usable.
Born in 1995, Jin Yongbin has been studying mechanics at Zhejiang University since his undergraduate years and began researching high-speed legged robots after starting his doctoral studies in 2017.
"When we started researching robots 10 years ago, they asked why they should fund you? So much funding for a robot that performs worse than a hunting dog." This is one of the most primitive confusions in Jin Yongbin's heart. "The muscle power density of humans is about 300W/kg, while motors can reach 3000W/kg or even higher, which is an order of magnitude better than biological performance, yet the overall performance of robots is still inferior to that of biology."
Therefore, Jin Yongbin's solution is to push hardware capabilities to the limit with speed, using the most efficient exposure methods to correct mechanical shortcomings. This allows the physical capabilities of the robot to match the intelligence it will need to carry in the future.
As Wang Xingxing said, "This breakthrough in physical limits is not only a display of speed but also a sign of the maturity of robot hardware performance and control algorithms."
Starting in 2023, Jin Yongbin noticed that robots are gradually being implemented in various industries, and more and more open-source algorithms and hardware are rapidly catching up with their technological advantages in the lab. In May of the following year, he co-founded Jing Shi Technology with his teacher, Wang Hongtao, the executive director of the Zhejiang University Cross-Disciplinary Mechanics Center, focusing on AI-powered high-mobility legged robot movement research and development.

 
Jin Yongbin with quadruped robot Black Panther
A completely academic founding team; according to the views from the venture capital circle in last October's viral article "Sorry, No More Funding for Professors' Startups," such founders are no longer favored by capital. Their biggest problem is that they do not love commercialization—obsessed with technology, neglecting the market, and can never finish the journey from the lab to the shelves.
However, the eggs laid by Jing Shi along the way may be more than many people think. The driving technology honed in the pursuit of extreme speed has already been applied to the industrial quadruped robot Apollo—capable of carrying 70kg and running at 8m/s, it has set a Guinness record and is already on sale. There is also a model that can function as a smart speaker with a screen and is now ready for mass production, officially released at the recently concluded AWE.
We are uncertain whether Jing Shi's route can ultimately succeed, but on an increasingly homogeneous track, a completely different answer is at least worth listening to seriously. Below is the dialogue between "Future Human Laboratory" and Jin Yongbin from mid-February, organized and published—

 
Fast, not just speed
Future: Why must it run fast?
Jin Yongbin: Fast is far more than just speed; it encompasses a series of abilities including strength, density, intensity, fluidity, and so on.
When we buy cars, we usually care about the acceleration time over a hundred kilometers; the shorter the time, the better the car's structure, and the engine

36Kr has officially settled on the ClawHub platform, launching three official Skills designed specifically for OpenClaw users. With no complicated operations, you can upgrade your OpenClaw into a new economic information hub with just one click, easily seizing information opportunities.
The power of OpenClaw comes from its infinitely expandable Skill ecosystem—these "skill packages" act like exclusive plugins for AI assistants, enabling originally powerful agents to have more precise and efficient specialized capabilities. As the official Skill store for OpenClaw, ClawHub gathers a vast array of high-quality skills, serving as the core base for users to acquire safe and reliable expansion capabilities. The three official Skills brought by 36Kr rely on 36Kr's professional content ecosystem and data resources, accurately matching the core information needs of new economic participants, allowing every OpenClaw user to conveniently access high-value content.
No need to switch platforms or manually search; install the 36Kr official Skills to transform OpenClaw into your exclusive new economic information assistant. Each of the three skills focuses on different aspects, covering hot topic tracking, startup exposure, and AI evaluation across three core scenarios, fully meeting your information needs!
Click here: ClawHub-36Kr Skill address
Skill 1: 24-hour hot list data, closely following industry trends
The new economic field is changing rapidly, with hot events and industry trends updated at any time. Missing a key piece of information could mean missing an opportunity. The official 36Kr 24-hour hot list gathers the hottest industry articles, event interpretations, and data reports within the platform, covering core areas like technology, venture capital, and the internet, truly reflecting industry dynamics and market trends. This Skill allows you to access complete hot list data through OpenClaw in real-time, without frequently refreshing the 36Kr platform, keeping you in sync with the latest industry hot spots, capital movements, and trend predictions, easily achieving "know yourself and know your enemy" and accurately grasping industry opportunities.
Click here: Get the 36Kr official 24-hour hot list
Skill 2: Self-service reporting articles directly link startups and capital
For entrepreneurs, project exposure is key to acquiring investments and expanding resources; for investors, uncovering quality projects is a core need. The 36Kr self-service reporting section serves as an important platform for startups to publish financing reports and seek exposure, carrying a wealth of content published by entrepreneurs and investors. It is the core bridge connecting startup projects with investment institutions. With this Skill, OpenClaw users can directly access content from the self-service reporting section, quickly browsing various startup project details, financing needs, and industry insights. Entrepreneurs can conveniently understand peer dynamics and connect with potential resources, while investors can efficiently uncover quality project leads, making information matching more efficient and direct.
Click here: Get articles from the 36Kr self-service reporting section
Skill 3: Daily AI evaluation notes, seizing the AI tool advantage
With the rapid iteration of AI technology, various AI tools are emerging, making it challenging to quickly filter out practical and efficient tools. The 36Kr AI evaluation website, known as the "Yelp of AI tools," integrates navigation, evaluation, and community interaction, gathering a wealth of real user feedback and professional evaluations. This Skill allows you to automatically receive the latest notes from 36Kr AI evaluations daily through OpenClaw, covering practical experiences, functional comparisons, and usage tips of various AI tools. No need to browse manually; you can grasp the latest trends in the AI tool industry and precisely choose the AI tools that suit you, enhancing work and learning efficiency.
Click here: Get the latest daily evaluation notes from 36Kr AI evaluations
Deepening the new economy, linking value creators. 36Kr is committed to enhancing business efficiency in China, providing valuable information and services to new economic participants. The launch of the ClawHub official Skills is another important measure for 36Kr to expand the boundaries of information services and empower users, aiming to make high-value information accessible through the intelligent ecosystem of OpenClaw.
Open ClawHub now, install the 36Kr official Skills, upgrade your OpenClaw, and obtain industry hot lists, self-service reports, and AI evaluations anytime and anywhere. Gain insights into new economic trends more efficiently, seize the opportunities of the era, and with 36Kr, see the future first and grasp the future first!

---
📊 Market Data: BTC $69,252.18 (-3.43%)
#Crypto #Bitcoin #Binance

Author | Huang Nan
Editor | Yuan Silai
On March 25, Sutech JunChuang (02498.HK) announced its fourth quarter and full-year performance report for 2025.
The financial report shows that in 2025, Sutech JunChuang achieved revenue of approximately 1.941 billion yuan, with total sales of laser radars around 912,000 units, and a gross profit of about 514 million yuan, a year-on-year increase of 81.3%. The performance in the fourth quarter was particularly outstanding, achieving an operating profit of 130 million yuan and a net profit of 104 million yuan, marking the first time a quarterly profit has been achieved. The comprehensive gross profit margin for the fourth quarter reached 28.5%, a year-on-year increase of 6.4 percentage points; the gross profit margin in the robotics sector reached 41.5%.
It can be seen that the company's dual-engine strategy of "automotive + robotics" has been substantially formed. Among them, the revenue from robotics and other products in the fourth quarter was approximately 347 million yuan, while the revenue from ADAS products was about 361 million yuan. The robotics business is no longer just a long-term story but has become a current growth engine alongside the automotive business.
Specifically, in terms of product sales. Official data shows that in 2025, Sutech JunChuang's robotics business experienced explosive growth, with fourth-quarter sales reaching 221,200 units, a year-on-year surge of 2565.1%; total annual sales reached 303,000 units, securing the top position in total sales of laser radars in the global robotics field.

Sutech JunChuang 2160 line laser radar (Image source/Company)
Behind the growth lies the company's precise positioning of the trend of AI technology moving from "above the screen" to the "physical world". Based on differentiated performance in terms of performance, cost, and mass production capability, their digital laser radars have secured a large number of orders in various segmented scenarios.
Among them, in the lawn mowing robot market, Sutech JunChuang has captured an absolute share of leading customers, with expected shipments in 2026 reaching 450,000 to 600,000 units; in the unmanned delivery sector, the company has a coverage rate of over 90% of leading customers, with companies like New Stone Tools, JD.com, and Meituan adopting their solutions. With the scaled mass production of RoboVan, the number of units mounted per vehicle is expected to increase to 2-4 units; in the field of embodied intelligence, nearly 50 global core customers have been covered, and the AC series “robotic eye” and hand-eye coordination solutions are undergoing scenario validation with leading customers to strategically prepare for future market explosions.
According to Sutech JunChuang CEO Qiu Chunchao, the company has reached cooperation with nearly 50 leading humanoid and quadrupedal robot companies, including Zhiyuan, Yushu, Zhongqing, and Galaxy General.
While the robotics business is experiencing rapid growth, Sutech JunChuang has not slowed down its layout in the ADAS sector. In 2025, its ADAS business demonstrated strong resilience and risk resistance, cumulatively achieving targeted agreements with 35 automotive companies for 167 models; the customer structure of the company in 2026 will become more diverse and healthy.

Full digital laser radar product matrix (Image source/Company)
A noteworthy increment comes from joint venture brands and overseas business. According to data released by Zosi Automotive Research on the market share of laser radar suppliers for joint venture automotive brands in 2025, Sutech JunChuang ranks first with over 70% share. The company expects that by 2026, sales from joint venture brands will reach 200,000 to 250,000 units.
On the technical level, Qiu Chunchao also provided a clear judgment on the development trends of laser radars. “Digital laser radars can easily break through 1000 lines while maintaining a high cost-performance ratio, fundamentally overturning the traditional perception that laser radar performance has limited upper limits and high costs.” Qiu Chunchao pointed out that the decline in industry ASP over the past two to three years mainly comes from technological iteration and architectural switching, with the current architecture being basically stable, and no drastic price reductions expected in the short term.
However, in terms of gross profit margin, the ADAS business still faces pressure in the short term. Qiu Chunchao stated, “On one hand, the automotive industry itself is highly competitive; on the other hand, the company is also actively promoting laser radars to be compatible with more vehicle models, moving towards higher penetration rates, and even future standard configurations.” However, as new products stabilize and sales grow, and product and chip advantages are released, ADAS gross margins will gradually improve and stabilize.

Robotaxi sector orders (Image source/Company)
In the Robotaxi sector, Sutech JunChuang predicts that 2026 to 2027 will be a key window period for a surge in orders. Currently, leading companies are advancing the iteration of new-generation models, with each vehicle typically equipped with 6-10 units. The company has established cooperation with over 90% of industry-leading companies, including Baidu's LuoBo Kuaipao, Didi Autonomous Driving, Wen Yuan Zhi Xing, Xiaoma Zhixing, Uber, etc.
The achievement of Sutech JunChuang's first quarterly profit marks the entry of its "automotive + robotics" dual-engine strategy into a phase of financial validation. This performance also reflects a change in industrial logic: the valuation of laser radar companies is shifting from a single market's shipment volume to multi-scenario coverage capabilities and generational advantages in technological architecture. And this shift in evaluation system may have just begun.

---
📊 Market Data: BTC $69,252.18 (-3.43%)
#Crypto #Bitcoin #Binance

Image source: 36kr
Weekly selection of 30+ quality positions covering fields such as AI large models, embodied intelligence, quantum computing, 3D printing, semiconductors, etc., helping you find your next career explosion point.
 This issue's partner companies: State-owned background AI, listed technology leaders, global unicorns, cutting-edge hard tech newcomers | Series B and above account for over 60%

 
Highlights of this issue

Image source: 36kr

 
Selected positions
| Zhongcheng Jiao Technology · Angel round · Transportation large model
One-sentence selling point:
Shanghai's first state-owned background transportation vertical large model company, with hundreds of billions in data assets, dual engine technology system
Location: Beijing/Shanghai
Job positions:
Backend Development Engineer (FastAPI/vector database/AI process orchestration | 3+ years | Salary negotiable)
Product Manager (AI large model product 0-1 implementation | 5+ years of experience | Salary negotiable)
Visual Algorithm Engineer (Traffic scene object detection/tracking | Bachelor’s degree or above | Salary negotiable)
Large Language Model Algorithm Engineer (LLM fine-tuning/RAG/Agent | Master’s degree or above | Salary negotiable)
CTO (AI large model transportation implementation | Doctorate preferred | Salary negotiable)
Job description:
Stable platform with state-owned background, deep data barriers in the transportation industry
Deep integration of large models + transportation scenes, participating in national-level industry intelligence projects
| Qunhe Technology · Pre-IPO · Spatial Intelligence
One-sentence selling point:
The parent company of the world's largest spatial design platform, Cool Home, led by the founder & chief scientist
Location: Hangzhou
Job positions:
Research Algorithm Engineer (Neural Rendering direction | >30K)
Algorithm Engineer (Video/Image model direction | 25K-30K)
AI Product Manager (Industrial AI/Digital Twin/3D Simulation | Master’s degree | 20K-25K)
Research Algorithm Engineer (3D graphics reconstruction and generation | Priority for Master’s/PhD)
Research Algorithm Engineer (Large model and AIGC direction | Doctorate)
Job description:
Enter core strategic projects, sufficient computing power and innovative fault tolerance funding
| Bairong Cloud Creation · Listed · Enterprise Intelligent Agent
One-sentence selling point:
Benchmark enterprise-level Agent platform, reconstructing productivity with AI silicon-based employees, serving over 8000 institutions
Location: Beijing
Job positions:
Large Language Model Application Algorithm Engineer (Post-training/multimodal | Master’s degree or above | Salary negotiable)
Agent Product Manager (Multi-agent/prompt engineering | Bachelor’s degree or above | Salary negotiable)
Large Model Inference Optimization Algorithm Engineer (Inference optimization | Bachelor’s degree or above | Salary negotiable)
Job description:
Stable listed platform, large-scale implementation verification, strong business certainty
Self-developed large model + Agent OS, participating in enterprise-level AI benchmark projects
Financial/Internet/Retail multi-industry implementation, broad technical vision
| SHEIN · Listed · Global Fashion Retail
One-sentence selling point:
The world's leading fashion retail platform, a highland of large-scale data and algorithm applications
Location: Nanjing
Job positions:
Senior Algorithm Engineer/Expert (Marketing risk control | Bachelor’s degree or above | Salary negotiable)
Job description:
Global massive users and business data, broad space for algorithm implementation
Global platform, high technical vision and career ceiling
| Agassi Intelligent Technology · Series A · Edge AI
One-sentence selling point:
Edge AI solution expert, core service providers for humanoid robots/industrial quality inspection, etc.
Location: Chengdu
Job positions:
Algorithm Engineer (CV/model deployment/embedded optimization | 15K-25K)
Job description:
Unique integrated operating system, supporting multi-system single-chip operation
Serving multiple humanoid robot and industrial AI clients, rich implementation scenarios
| Daily Interaction · Listed · Data Intelligence
One-sentence selling point:
The domestic listed company with the highest "data inclusion rate," empowering thousands of industries with data intelligence
Location: Hangzhou
Job positions:
AI Algorithm Engineer (Large model application/RAG/Agent | 1-3 years)
AI Application Testing Engineer (Large model/Agent system testing | 2-3 years)
AI Application Development (AI Agent platform architecture | 3-5 years)
AI Algorithm Expert (Voiceprint recognition & speech translation, image small models | Master’s degree or above)
Multimodal Agent Algorithm Engineer (Multimodal input/model training | Master’s degree or above)
Job description:
Relying on the company’s core AI product (Individual Intelligence - Intelligent Workstation), empowering the full process scenes of government affairs
Deeply cultivating AI large models/multimodal/Agent technology implementation, exploring technology implementation in real business scenarios
| Lipusi Semiconductor · Series B+ · SiC Power Semiconductor
One-sentence selling point:
High-end SiC module manufacturer, focusing on high-power charging, recruiting hardcore talents with high salaries
Location: Wuxi
Job positions:
Power Module Design/Product Definition Expert (8+ years | 600K-1M/year)
DSP Software Development Expert (5+ years | 400K-600K/year)
Hardware Development Expert (5+ years | 400K-600K/year)
FAE Expert (5+ years | 400K-600K/year)
PCB Layout Expert (5 years of relevant experience | 400K-600K/year)
Job description:
Mature stage of Series B+, overseas channels are complete, clear commercialization
| Tuo Bamboo Technology · Series B · Global Desktop 3D Printing Unicorn
One-sentence selling point:
Five

On March 25, Gu Ming released the 2025 annual performance report, showing stable growth in both revenue and profit.
The financial report indicates that in 2025, Gu Ming achieved total revenue of approximately ¥12.9 billion, a year-on-year increase of 46.9%; adjusted profit (not measured by international financial reporting standards) was approximately ¥2.575 billion, a year-on-year increase of 66.9%.
By the end of 2025, the number of Gu Ming stores reached 13,554. In 2025, 4,292 new stores were opened, with a net increase of 3,640 stores. The store network covers over 200 different levels of cities nationwide, with a proportion of 82% in second-tier and below, and the proportion of town stores grew to 44%, further strengthening the layout in lower-tier markets, earning the title of “King of Lower-tier Markets.”
After ensuring store scale, Gu Ming's store operational efficiency has also greatly improved. During the reporting period, the total GMV of Gu Ming stores reached ¥32.732 billion, a year-on-year increase of 46.1%; the average daily GMV per store reached approximately ¥7,800.
In addition, regarding supply chain capacity, by the end of 2025, Gu Ming operated 24 warehouses, with a total construction area of approximately 258,000 square meters, of which cold storage capacity exceeds 70,000 cubic meters, supporting various temperature range storage needs. About 75% of Gu Ming stores are located within 150 kilometers of the warehouse, allowing for cold chain delivery services to about 98% of stores every two days as needed.
From the results, based on these self-owned cold chain storage and logistics infrastructure, Gu Ming has established refined management standards from supply procurement, to raw material processing, to warehousing and transportation, and then to stores. At the same time, leveraging regional densification strategies, Gu Ming can continuously deliver the freshest raw materials to tens of thousands of stores at relatively economical costs.
In 2025, the delivery cost from Gu Ming's warehouses to stores is less than 1% of the total GMV. While improving supply chain efficiency, Gu Ming is also increasing investment in product research and development. In the past year, Gu Ming launched a total of 106 new products, among which significant progress was made in the coffee category. By the end of last year, over 12,000 Gu Ming stores were equipped with coffee machines, and 27 new coffee drinks were launched throughout the year.
Good products and services are reflected at the consumer end; the financial report shows that Gu Ming sold over 1.9 billion cups of drinks in 2025. The number of registered members on the mini-program reached approximately 206 million, with about 52 million active members per quarter.
Entering 2026, due to Gu Ming's strong market growth performance, several institutions, including Goldman Sachs and Bank of America Securities, have given a “buy” rating. Bank of America Securities believes that Gu Ming is likely to become a winner in market share, with growth surpassing its peers.

---
📊 Market Data: BTC $69,964.13 (-1.82%)
#Crypto #Bitcoin #Binance

Source: 36kr
Selected 30+ high-quality positions every week, covering fields such as AI large models, embodied intelligence, quantum computing, 3D printing, semiconductors, etc., helping you find the next career breakthrough point.
 This issue's partner companies: State-owned AI with background, listed technology leaders, global unicorns, cutting-edge hard technology newcomers | Series B and above accounting for over 60%

 
Highlights of this issue

Source: 36kr

 
Selected Positions
| Zhongcheng Jiaotong Technology · Angel Round · Traffic Large Model
Selling point in one sentence:
The first state-owned background traffic vertical large model company in Shanghai, with a data asset worth hundreds of billions and a dual-engine technology system
Location: Beijing/Shanghai
Recruitment Positions:
Backend Development Engineer (FastAPI/vector database/AI process orchestration | 3+ years | salary negotiable)
Product Manager (AI large model product 0-1 implementation | 5+ years experience | salary negotiable)
Visual Algorithm Engineer (Traffic scene object detection/tracking | Bachelor's degree or above | salary negotiable)
Large Language Model Algorithm Engineer (LLM fine-tuning/RAG/Agent | Master's degree or above | salary negotiable)
CTO (AI large model traffic implementation | Doctorate preferred | salary negotiable)
Position Introduction:
Stable platform with state-owned background, deep data barriers in the traffic industry
Deep integration of large models + traffic scenes, participating in national-level industry intelligence projects
| Qunhe Technology · Pre-IPO · Spatial Intelligence
Selling point in one sentence:
The parent company of the world's largest spatial design platform, Cool Home, directly led by the founder & chief scientist
Location: Hangzhou
Recruitment Positions:
Research Algorithm Engineer (Neural Rendering direction | >30K)
Algorithm Engineer (Video/Image model direction | 25K-30K)
AI Product Manager (Industrial AI/Digital Twin/3D Simulation | Master's | 20K-25K)
Research Algorithm Engineer (3D reconstruction and generation | Master's/Doctorate preferred)
Research Algorithm Engineer (Large model and AIGC direction | Doctorate)
Position Introduction:
Entering core strategic projects, ample computing power and innovative fault-tolerant funding
| Bairong Cloud Creation · Listed · Enterprise-level Intelligent Agents
Selling point in one sentence:
Benchmark enterprise-level Agent platform, reconstructing productivity with AI silicon-based employees, serving 8000+ institutions
Location: Beijing
Recruitment Positions:
Large Language Model Application Algorithm Engineer (Post-training/multimodal | Master's and above | salary negotiable)
Agent Product Manager (Multi-agent/prompt engineering | Bachelor's and above | salary negotiable)
Large Model Inference Optimization Algorithm Engineer (Inference optimization | Bachelor's and above | salary negotiable)
Position Introduction:
Stable listed platform, large-scale landing verification, strong business certainty
Self-developed large models + Agent OS, participating in enterprise-level AI benchmark projects
Financial/Internet/Retail multi-industry landing, broad technical vision
| SHEIN · Listed · Global Fashion Retail
Selling point in one sentence:
The world's leading fashion retail platform, a high ground for large-scale data and algorithm applications
Location: Nanjing
Recruitment Positions:
Senior Algorithm Engineer/Expert (Marketing Risk Control | Bachelor's and above | salary negotiable)
Position Introduction:
Massive global users and business data, extensive algorithm implementation space
Global platform, high technical vision and career ceiling
| Agassi Intelligent Technology · Series A · Edge AI
Selling point in one sentence:
Expert in edge AI solutions, core service provider for humanoid robots/industrial quality inspection, etc.
Location: Chengdu
Recruitment Positions:
Algorithm Engineer (CV/model deployment/embedded optimization | 15K-25K)
Position Introduction:
Unique integrated operating system, supporting multi-system single-chip operation
Serving multiple humanoid robots and industrial AI clients, rich application scenarios
| Daily Interaction · Listed · Data Intelligence
Selling point in one sentence:
The domestic listed company with the highest "data inclusion rate", empowering thousands of industries with data intelligence
Location: Hangzhou
Recruitment Positions:
AI Algorithm Engineer (Large model application/RAG/Agent | 1-3 years)
AI Application Testing Engineer (Large model/Agent system testing | 2-3 years)
AI Application Development (AI Agent platform architecture | 3-5 years)
AI Algorithm Expert (Voiceprint recognition & speech translation, image small models | Master's and above)
Multimodal Agent Algorithm Engineer (Multimodal input/model training | Master's and above)
Position Introduction:
Relying on the company's core AI product (Individual Knowledge - Intelligent Workstation), empowering government affairs full-process scenarios
Deeply cultivating AI large models/multimodal/Agent technology landing, exploring technology implementation in real business scenarios
| Lipus Semiconductor · Series B+ · SiC Power Semiconductor
Selling point in one sentence:
High-end SiC module manufacturer, focusing on high-power charging, recruiting hardcore talents with high salaries
Location: Wuxi
Recruitment Positions:
Power Module Design/Product Definition Expert (8+ years | 60W-100W/year)
DSP Software Development Expert (5+ years | 40W-60W/year)
Hardware Development Expert (5+ years | 40W-60W/year)
FAE Expert (5+ years | 40W-60W/year)
PCB Layout Expert (5 years relevant experience | 40W-60W/year)
Position Introduction:
Mature stage of Series B+, overseas channels improved, clear commercialization
| Tuo Bamboo Technology · Series B · Global Desktop 3D Printing Unicorn
Selling point in one sentence:
Five

On March 25th, GuMing released its 2025 annual performance report, showing stable growth in both revenue and profit.

The report shows that in 2025, GuMing achieved total revenue of approximately RMB 12.9 billion, a year-on-year increase of 46.9%; adjusted profit (non-IFRS) was approximately RMB 2.575 billion, a year-on-year increase of 66.9%.

As of the end of 2025, GuMing had 13,554 stores. 4,292 new stores were opened in 2025, resulting in a net increase of 3,640 stores. The store network covers more than 200 cities of different tiers nationwide, with 82% located in second-tier cities and below, and the proportion of stores in townships increasing to 44%, further strengthening its presence in lower-tier markets and making it the undisputed "King of Lower-Tier Market Penetration."

With its store scale maintained, GuMing also significantly improved its store operating efficiency. During the reporting period, GuMing's total GMV reached RMB 32.732 billion, a year-on-year increase of 46.1%; the average daily GMV per store reached approximately RMB 7,800.

Furthermore, regarding supply chain capabilities, as of the end of 2025, GuMing operated 24 warehouses with a total building area of ​​approximately 258,000 square meters, including cold storage capacity exceeding 70,000 cubic meters, supporting storage needs across multiple temperature ranges. Approximately 75% of GuMing stores are located within a 150-kilometer radius of these warehouses, enabling on-demand cold chain delivery services every two days to about 98% of the stores.

As a result, based on these self-owned cold chain warehousing and logistics infrastructures, GuMing has established refined management standards from supply procurement and raw material processing to warehousing, transportation, and finally, store operations. Simultaneously, through a geographically concentrated strategy, GuMing can continuously deliver the freshest raw materials to tens of thousands of stores at a relatively economical cost.

In 2025, GuMing's warehouse-to-store delivery cost was less than 1% of its total GMV. While improving supply chain efficiency, GuMing also increased investment in product development. In the past year, GuMing launched 106 new products, achieving significant progress in the coffee category. By the end of last year, over 12,000 GuMing stores were equipped with coffee machines, and 27 new coffee beverages were launched throughout the year.

This positive product and service performance is reflected at the consumer end. Financial reports show that GuMing sold over 1.9 billion cups of beverages in 2025. The number of registered members on its mini-program reached approximately 206 million, with about 52 million quarterly active members.

Entering 2026, due to GuMing's strong market growth performance, several institutions, including Goldman Sachs and Bank of America Securities, have given it a "buy" rating.Bank of America Securities believes that GuMing is poised to become a market share winner, outperforming its peers in growth.

--- 📊 Market Data: BTC $69,911.79 (-1.71%)
#Crypto#Bitcoin #Binance