1.CXMT's market value falls by 1 trillion in two months 2.At least 18 outages this year! DeepSeek issues downtime compensation, some users received 3,000 yuan 3.Alibaba doubles down on AI hardware brands with Tmall and AliExpress linkage 4.Apple Intelligence for China expected to launch in December with Baidu and Alibaba participating in the cooperation 5.Kling AI new-generation model Kling 4.0 officially launches, professional imaging models reach new heights 6.Huawei's Huang Yu: the biggest challenge for Tao's Law lies in EDA tools 7.Identical interiors! Overseas blogger says Chinese automakers have entered a phase of copying each other: Li Auto executive responds
8.OpenAI infrastructure head to depart
Today's Top Story
CXMT's market value falls by 1 trillion in two months
On October 15, CXMT's share price declined; at the close, the stock stood at 43.59 yuan per share, with total market value falling below 3 trillion yuan. The position of A-share "total market value champion" subsequently changed hands, with ICBC returning to first place in A-share total market value.
On July 27, CXMT's first day of listing, its intraday market value surpassed ICBC's to top A-share market value. On August 17, CXMT closed at 61.8 yuan per share, with market value reaching 4.2 trillion yuan. Over the following roughly two months, CXMT's market value first fell below 4 trillion yuan, then below 3 trillion yuan. From the August 17 close to October 14, CXMT's share price fell 25.57% and its market value dropped by 1.07 trillion yuan.
Over the past two months, CXMT has repeatedly updated its technology and product R&D progress. As previously reported, on October 10, CXMT President Cao Kanyu said CXMT expects to launch a new-generation DDR5 RDIMM product using the 4F² DRAM (dynamic random access memory) architecture by the end of the year. (Yicai)
Domestic News
At least 18 outages this year! DeepSeek issues downtime compensation, some users received 3,000 yuan
News on October 15: DeepSeek issued a notice apologizing for the API service failure that occurred on the morning of the 15th, and issued compensation to API users. As compensation, DeepSeek will fully refund the amount consumed by users' accounts within 7 days (10.8-10.14) in the form of bonus credit; the bonus credit is credited automatically, and API billing deductions are taken first from the bonus credit account.
After the notice was published, many netizens began sharing the compensation they had received. Some had used only a small amount of API in the past week, so their compensation was only a few mao or a few yuan, but some users had higher demand and heavy usage over the past week, receiving compensation of several hundred or even more than three thousand yuan. One user said DeepSeek's "compensation shows real sincerity"; another user joked that this move was a "Liang-style quota reset."
DeepSeek did not explain the cause of the service failure. It is reported that during this outage, some netizens suggested DeepSeek switch to a new batch of compute hardware, while others claimed "rumor has it the compute was diverted to work on something big." Notably, since 2026 the platform has experienced at least 18 large-scale service anomalies, averaging one to two per month, with at least 7 service fluctuations in September alone. (Yicai, Mydrivers)
Son following in father's footsteps? Film Streak responds to Tim's father becoming company chairman
News on October 15: public information shows that recently, Hangzhou Xing'ao Media Co., Ltd., a company affiliated with Film Streak, underwent a business registration change, with Pan Shuimiao, director and president of YTO Express, taking the position of chairman of the company. At the same time, the company's business scope added sales of prepackaged food, film production, and multi-channel distribution services for internet information content.
After this change, founder Pan Tianhong (Tim) remains the company's legal representative, manager, and director, and is the largest shareholder with a 51.77% stake; Pan Shuimiao holds 25.65%, making him the second-largest shareholder. According to public reports, Pan Shuimiao is Pan Tianhong's father.
Regarding this matter, "Film Streak" founder Pan Tianhong (Tim) gave a recent response: Pan Shuimiao has been an executive director since the company's founding and does not participate in operations; a board of directors was only established now out of the need for company development. This business change mainly added some business categories, and some interesting products will be launched next. (Sina Tech, Mydrivers)
Alibaba doubles down on AI hardware brands with Tmall and AliExpress linkage
News on October 15: recent reports say Alibaba has been in contact with a group of AI hardware brands, and besides providing angel-round investment, may also link its domestic and overseas e-commerce businesses to provide brands with angel-round investment and global market expansion support.
Notably, AliExpress will hold its annual and Double 11 merchant conference on October 19, and Tmall rarely appears on the publicly released agenda. Multiple sources of information point to Alibaba possibly announcing a domestic-overseas linkage plan for AI hardware brands at the conference.
Apple Intelligence for China expected to launch in December with Baidu and Alibaba participating in the cooperation
News on October 15: according to foreign media reports, Apple plans to officially launch Apple Intelligence in December this year via the iOS 27.3 system update, with the first batch being AI features released in the US two years ago. Separately, Apple Executive Chairman Cook recently revealed that Apple Intelligence has been approved in China. A test version will come later in October, with the first batch of features including real-time translation, writing tools, and image cleanup; Siri AI is not included.
Apple AI adopts a dual-partner model in China. According to foreign media reports, Apple will embed Baidu's software services in the Siri voice assistant and Visual Intelligence features. A teardown of the iOS 27 Beta 2 firmware by third-party developers verified this information. The teardown showed that a new module named "Baidu Visual Search" was added to the system, pointing to the integration of visual search capabilities.
Previously, domestic media reported that Alibaba would be integrated as AI capability into Apple Intelligence, providing Chinese users of iOS, iPadOS, macOS, and visionOS with intelligent experiences such as text and image understanding and content generation. The license obtained on September 29 covers both Apple Intelligence and Alibaba's AI models, including Tongyi Wanxiang and Qwen.
Kling AI new-generation model Kling 4.0 officially launches, professional imaging models reach new heights
On October 15, Kling AI's new-generation video generation model Kling 4.0 officially launched.
Kling 4.0 focuses on high-dynamics, strong camera-movement video creation scenarios, achieving concentrated breakthroughs in three directions: large-magnitude motion performance is more stable, improving the coherence of high-speed movement and complex camera movements; a single native generation can last up to 30 seconds, supporting long takes and continuous narrative with directorial thinking online throughout; all-around reference supports combining up to 15 reference materials, while also supporting up to 10 keyframe inputs, helping creators turn one idea into a complete work.
High-quality delivery of core shots is an important direction in the continuous iteration of the Kling model. Professional creators need not just occasional stunning generated images, but the stable completion of the key shots that determine a work's quality, and the ability to bring generated material into a complete production workflow.
Huawei's Huang Yu: the biggest challenge for Tao's Law lies in EDA tools
October 15 news, according to media reports, Huawei semiconductor EDA chief scientist Huang Yu said at the 2026 Bay Chip Expo that the biggest challenge of Tao's Law lies in EDA tools. Meanwhile, process challenges, engineering challenges, and system design challenges also exist.
Huang Yu revealed that Huawei's Tau (τ) Law consists mainly of four layers: the most fundamental device layer, which optimizes transistor structures and interconnection methods to compress device time constants at the source; the second layer is the link layer, which is their biggest innovation; the third layer is the chip layer, with software-hardware co-design forming new architectures; and the fourth layer is the system layer, namely the well-known supernodes and rack interconnection, which require better communication protocols and faster speeds.
It is understood that EDA stands for Electronic Design Automation. It refers to a software tool chain that uses computer-aided design (CAD) software to complete the entire process of functional design, synthesis, verification, placement and routing for very-large-scale integration (VLSI) chips and electronic systems. In the semiconductor and microelectronics fields, EDA is known as the 'mother of chips.' When chip designers design chips composed of hundreds of millions or billions of transistors, they must rely on EDA software. (IT之家 / ITHome)
ByteDance in talks to purchase MediaTek AI chips for building data center computing power
October 15 news: It is reported that ByteDance is in in-depth negotiations with MediaTek over the purchase of custom data center chips, which will be used in ByteDance's data centers. Informed sources revealed that MediaTek is expected to supply ByteDance with a certain quantity of application-specific integrated circuits (ASICs), but the specific scale has not been disclosed, and the two parties have not yet formally announced an agreement.
For MediaTek, if the deal is reached, ByteDance is expected to become its second major ASIC customer after Google. This would also push MediaTek to further expand its data center business, competing with vendors such as Qualcomm and Broadcom, and reduce its dependence on the smartphone chip business.
ByteDance is increasing its investment in AI infrastructure and regards artificial intelligence as an important growth engine for its next stage. Since it cannot obtain the most advanced AI chips from vendors such as NVIDIA, and local suppliers such as Huawei cannot yet meet all of its demand, ByteDance is exploring alternatives such as partnering with chip design vendors. Earlier this year, ByteDance reached a similar purchase arrangement with Qualcomm, planning to buy millions of AI chips. (快科技 / MyDrivers)
Interiors look exactly the same! Overseas blogger says Chinese carmakers have entered a stage of copying each other: Li Auto executive responds
October 15 news: Recently, an overseas blogger posted on social media, sharing his views on several Chinese high-end SUVs at the Paris Motor Show. After the content was reposted to Weibo by a domestic blogger, Li Auto's 'Ying Ge' reposted it and commented: 'It's not exactly copying each other — they only copied one company.'
The overseas blogger said that Chinese automakers have entered a peculiar stage: instead of copying Western products, carmakers have begun copying each other, a phenomenon especially prominent among high-end SUV models, with about seven such models at the Paris Motor Show. He noted that these models can still be distinguished by their exteriors, but their interiors are extremely similar.
The blogger said that multiple models carry in-car software developed on Android with nearly identical interface layouts and functions, and their instrument clusters, center console designs, and hardware configurations are highly alike. The blogger also mentioned that he had test-driven many of these models, and their powertrains and driving characteristics differ very little. He also believes that it is not just high-end SUVs — products from many unrelated Chinese carmakers share this problem: the vehicles' hardware technology is impressive, but originality is lacking; however, relying on standardized component procurement is enough to support the products' market success. (快科技 / MyDrivers)Kimi's 17-year-old genius Chen Guangyu announces departure! His paper was once endorsed by Musk
October 15 news: Moonshot AI's 17-year-old genius Chen Guangyu posted on social media announcing his departure from Kimi. Looking back on his experience working at Kimi, he said frankly: 'The past year at Kimi has been the best year of my life so far.'
In March (3) of this year, the Kimi team released a technical paper titled 'Attention Residuals' (注意力残差), attempting to improve the residual connection mechanism that large models have used for many years. It sparked wide discussion in the AI community and even earned a like from Musk.
Among the three co-contributing co-first authors of this paper was the then 17-year-old Chen Guangyu, who was still in his final year of high school. Attention Residuals later became one of the key technologies of Moonshot AI's next-generation large model, Kimi K3. (智猩猩AI)World-class white hat and renowned security researcher Gu He joins Zhipu
October 15 news: World-class white hat and renowned security researcher Gu He (Yuki Chen) has recently officially joined Zhipu, working in the GLM foundation model group, with the core task of improving the cybersecurity capabilities of large models. Gu He previously topped Microsoft's MSRC 'Most Valuable Security Researchers' annual list for multiple consecutive years and served as head of 360's Vulnerability Research Institute.
Zhipu's official overseas account Z.ai has confirmed the news of his joining, and disclosed that the 23 Windows vulnerabilities discovered by Gu He and his team using GLM models have all been confirmed and acknowledged by Microsoft's Security Response Center MSRC. (科创板日报 / STAR Market Daily)
Baidu launches short drama product 'Wenjuhui,' competing head-to-head with Tencent and ByteDance
October 15 news: Baidu has launched a short drama and comic drama product 'Wenjuhui' on WeChat. The product is delivered as a mini program, with its theater section separating short drama and comic drama content. However, compared with Hongguo's mature recommendation algorithms, Wenjuhui's content library is currently small in volume and its content update pace has not been fully ramped up, making it more like a pilot product Baidu has put out in the short drama track.
From hands-on experience, Wenjuhui's interface layout is close to mature products such as ByteDance's Hongguo and Tencent's Lüpaopao; tapping a series starts playback directly, and the overall approach is free viewing supported by ads. Baidu's entry adds another major player to the showdown among big companies in the free short drama and comic drama track. (读佳)
Alibaba DAMO Academy's Rong Yu starts a business, founding 'Gezhi Yongxian' to enter the AI for Science track
LeiFeng.com (official account: LeiFeng.com)'s Jingxi channel has exclusively learned that Rong Yu, head of AI for Science at Alibaba DAMO Academy, has left to found a new company, 'Gezhi Yongxian,' with its business direction focused on AI for Science. Sources say the new company has received investment from top-tier investment institutions.
Public records show that Rong Yu is a senior technical expert at Alibaba DAMO Academy. He received his PhD in Systems Engineering and Engineering Management from the Chinese University of Hong Kong in 2016 and did postdoctoral research there. In June 2017, Rong Yu joined Tencent AI Lab as an expert researcher, and in June 2024 he joined Alibaba DAMO Academy, leading research in the AI for Science direction.
Rong Yu's move to commercialize this technical direction independently as a startup is seen as a key signal of AI for Science moving from the laboratory to industry. Among the multiple AI for Science-related directions, the virtual cell is currently a hot topic of market discussion. (鲸犀 / Jingxi)
OpenCode announces DeepSeek V4.1 Flash offered for free, but only supports 200K context
October 15 news: OpenCode announced that DeepSeek V4.1 Flash is now available for free. Real-world testing shows the free version supports a 200K context window, a reduction compared with the 1M context version.
This free offering is a further step based on the earlier 'Operation Cheepseek' plan. In September this year, OpenCode temporarily increased the usage quota for DeepSeek V4.1 Flash in the OpenCode Go plan from 15 USD to 60 USD, then adjusted that quota to 80 USD on October 12. With this free offering, calls to the model within the OpenCode platform no longer count against the quota.
DeepSeek V4.1 Flash was officially released on September 10, 2026. It is a natively multimodal MoE model with a backbone parameter scale of 552B, plus 196B Engram conditional memory parameters. (IT之家 / ITHome)
OPPO partners with Manus; Find X10 series to debut with it
October 15 news: OPPO has partnered with Manus, and the AI Mind Pilot feature, powered by Manus, will debut on OPPO Find X10 series phones running ColorOS 17. Media learned that this partnership is limited to overseas versions. (科创板日报 / STAR Market Daily)
Wahaha subsidiary suspected of 'theft'; over 60 tons of plastic 'disappear'
On October 15, it was reported that a Wahaha subsidiary, 'Baishan Hongsheng Beverage Co., Ltd.,' was recently found to have suffered raw material 'theft,' with over 60 tons of plastic pellets stolen.
According to informed sources, the case is still under investigation, and multiple employees have confirmed that a case has been filed. "It started when an internal employee reported a theft; the company's internal audit found the accounts did not match. Some employees also reported that many plastic pellets were hauled away directly without ever being entered into the warehouse. The specific (loss) quantity is subject to the final ruling of the group company or the police investigation." Wahaha has not yet responded to the above situation.
It is understood that plastic pellets are the raw material used by beverage factories to produce PET bottles. In industrial packaging, they are typically stored in warehouses in large industrial bulk bags of 1 ton or 1.05 tons each. Public information shows that Baishan Hongsheng Beverage Co., Ltd. was originally named "Baishan Wahaha Beverage Co., Ltd." and was officially renamed Baishan Hongsheng Beverage Co., Ltd. in December 2025. (Sina Tech)
International News
Musk Purchases ASML EUV Lithography Machines Ahead of Schedule
Musk's chip project Terafab is under rapid construction to prepare for an early start to chip design and manufacturing. According to informed sources, the project may receive its first ASML EUV lithography equipment around the middle of next year. Previously, research institutions had predicted that the main structure of the Terafab R&D fab would not be completed until the end of 2027, with equipment installation only beginning later that year. As a result, the rumored EUV lithography equipment may sit idle for a period of time.
Musk has previously hinted that Terafab plans to apply a new extreme ultraviolet light generation method, using a free-electron laser (FEL) in place of ASML's laser-produced plasma light source. The news that Terafab is purchasing ASML lithography machines ahead of schedule also means that the project's first-generation products will still rely on more conventional equipment. But this is not surprising, because Terafab plans to adopt Intel's 14A manufacturing technology. Intel's 14A process will use not only existing low-numerical-aperture EUV scanners but also next-generation high-numerical-aperture EUV equipment.
Moreover, a lithography system using an unconventional and insufficiently tested light source would require major modifications to the manufacturing process, extensive equipment certification, and possibly changes to photomasks and photoresists, which would complicate the transfer of Intel's 14A technology and delay Terafab's production ramp-up. (Kechuang Board Daily)
OpenAI's Head of Infrastructure to Depart
On October 15, according to foreign media reports, OpenAI's head of infrastructure, Sachin Katti, announced on Wednesday that he will leave the artificial intelligence company after working there for about a year. In a post on LinkedIn, Katti said he "have decided to leave to pursue a new venture," and that after his last day on November 15, he will continue to support OpenAI as an advisor.
A spokesperson said that Uday Ruddarraju, who has been OpenAI's CTO of Compute, will succeed Katti in leading the company's overall compute division. According to his LinkedIn profile, Uday Ruddarraju joined OpenAI in July of last year. Previously, he had served as head of infrastructure engineering at Elon Musk's xAI, a company later acquired by SpaceX.
Katti's departure is the latest in a series of executive departures at OpenAI this year. In July of this year, OpenAI's head of products and business, Fidji Simo, said she would move to an advisory role to focus on recovering from a "serious flare-up of a chronic illness." The following month, OpenAI veteran executive Brad Lightcap and chief revenue officer Denise Dresser also left the company. (Sanyan Tech)
Mainland China Market Accounts for 23%! ASML CFO: Opposes a DUV Sales Ban on China, Which Would Accelerate China's Self-Developed Lithography Machines
On October 15, according to foreign media reports, ASML Chief Financial Officer Roger Dassen said at a press conference call following the release of third-quarter results that the company opposes a proposal in the U.S. Congress to restrict its sales of deep ultraviolet lithography equipment to China. His characterization was blunt: it is a bad idea. His reasoning is that if the MATCH Act passes after the U.S. midterm elections, it would accelerate China's research and development of related technologies.
However, he also stated publicly that he is optimistic that common sense will ultimately prevail among decision-makers, and that such unreasonable proposals most likely will not ultimately take effect.
According to ASML's latest publicly released Q3 financial report, the mainland China market accounted for exactly 23% of ASML's total net sales in the third quarter of this year. Roger Dassen subsequently predicted that for the full year of 2026, the overall business share of the mainland China market will remain stable at around 20%. (MyDrivers)Musk: Starlink Cannot Be Used for Offensive Combat; Use Cases Limited to Defense and Ordinary Civilian Services
On October 15, SpaceX CEO Elon Musk recently spoke out publicly, making clear that the company's Starlink system will not be used to carry out offensive combat missions, and that its use cases are limited to defense purposes and ordinary civilian services.
On social media, Musk laid out in detail the boundaries of responsibility between the two related systems: the Starshield system, specially developed for military customization, is entirely owned by the U.S. government and directly controlled by the U.S. side throughout. All restrictions on this military system's usage priority and scope of operation are established by the U.S. military, with ultimate decision-making authority resting with the President of the United States.
He specifically emphasized that the ordinary Starlink service offered to the public for commercial use is a commercial service system, with ownership and day-to-day operational control held by SpaceX. Starlink's entire positioning is limited to civilian scenarios, or serving defensive military missions under compliant conditions, and it will absolutely not be connected to the execution chain of any offensive military operations, because such high-risk missions are completely unsuitable for a private commercial company. (MyDrivers)
Jensen Huang: NVIDIA Must Cooperate with Chinese Companies on AI Safety
On October 15, Jensen Huang said in an interview with foreign media that it is "absolutely necessary" for NVIDIA to cooperate with Chinese AI companies to jointly monitor the risks posed by this frontier technology. NVIDIA CEO Jensen Huang believes AI is potentially dangerous. He believes his company absolutely must join hands with Chinese AI companies to monitor these dangers.
Huang explained: "We should absolutely work together to advance AI development in a safe manner." He believes cooperation with Chinese AI companies should be achieved through "dialogue," and revealed that he has been in communication with researchers in China.
AI safety has become an important topic of discussion between China and the United States, and as artificial intelligence develops rapidly, both countries are weighing how to address the risks. U.S. Treasury Secretary Bessent has proposed to the Chinese side establishing a bilateral AI safety risk notification mechanism and communication channels to address potential crises such as AI agent loss of control, cybersecurity, and biological threats. (IT Home)