【Data and Intelligence Weekly Report】 The new model Kimi K2.6 from Moonshot has been released and open-sourced; Deepseek official website has published the deepseek-v4 interface documentation; China’s software industry scale is expected to exceed 20 trillion yuan; Tesla China’s in-car system will connect to the Doubao large model; Open...

null

【The Weekly Intelligence Report will consolidate the most important enterprise services, cloud computing, big data trends, major policies, and industry research reports of the week.】

Opinions

Vice Chairman and CEO of AsiaInfo Security Ma Hongjun: Security requires building a data-driven, AI-native system

Vice Chairman and CEO of AsiaInfo Security Ma Hongjun: The old security construction model has reached a ceiling. What security needs is not just buying another box or adding another product, but a thorough rebuild of the underlying logic—based on coordinated defense, to create a new system driven by data and native to AI.

Tesla CEO Musk: Expect severe shortages of AI chips in the future

Tesla CEO Musk stated during the earnings call that Tesla launched the Terafab chip factory project because the company anticipates a serious shortage of AI chips in the future. He said, “In terms of industry growth, logic chips, and even more so storage chips, we expect that if we don’t manufacture our own chips, we will hit a bottleneck. That’s the reason for the birth of Terafab.”

Domestic News

First domestic park-level token (Token) base established in Baotou

“The first park-level token (Token) base project in China” was officially signed in Guyang County, Baotou City. The project aims to create the country’s first park-level token production and operation center by integrating operations, products, computing power, and energy.

New model Kimi K2.6 of Moon’s Dark Side released and open-sourced

Moon’s Dark Side released and open-sourced the Kimi K2.6 model on the evening of April 20. It is claimed to be its strongest code model to date, with improved long-range coding ability and enhanced autonomous execution of agents. Kimi K2.6 is now available on kimi.com, the latest Kimi app, Kimi API, and Kimi Code programming assistant, accessible to all users.

UCloud: First quarter net profit of 2.74 million yuan, turning losses into profits year-on-year

UCloud announced that in the first quarter of 2026, revenue reached 439 million yuan, a year-on-year increase of 16.77%; net profit attributable to the listed company’s shareholders was 2.74 million yuan, turning losses into gains. In Q1 2026, AI-related revenue accounted for over 40%; since the beginning of 2026 until the disclosure of this report, the company added nodes in multiple regions including Denver, Brazil São Paulo, Ho Chi Minh City, Vietnam, and Almaty, Kazakhstan. Additionally, the company’s Ulanqab in Inner Mongolia and Qingpu Smart Computing Center in Shanghai have been gradually put into operation with high efficiency.

Volkswagen releases “Full-Field Intelligent Agent AI” technical route, plans to introduce in-vehicle AI agents

On the eve of the Beijing Auto Show, Volkswagen Group announced the “Full-Field Intelligent Agent AI” technical route, planning to gradually introduce in-vehicle AI agents in new models to achieve an integrated intelligent experience of cabin and driving. With accelerated development of domestic R&D systems and product pace, Volkswagen’s product layout in China is shifting from single-point model updates to a systematized, multi-brand coordinated phase. On the same day, four models from Volkswagen, Jetta, and AUDI made their global debut, covering pure electric mid-size SUVs, electric sedans, and entry-level new energy vehicles. These models are built on domestically developed electronic and electrical architecture, equipped with enhanced L2-level assisted driving and intelligent cockpit systems.

Alibaba releases Qianwen AI digital human image “Qianwen Little Dimple”

Alibaba officially launched the ecosystem-level AI assistant “Qianwen Little Dimple.” As the unified digital human image of Alibaba Qianwen AI, this intelligent partner will not only be available in the Qianwen App but will also be integrated into various Alibaba ecosystem applications to provide users with a unified, convenient intelligent experience.

Baidu Alliance launches “Nebula Plan,” opens Baidu Orion AI engine

At the Baidu Alliance Conference, the “Nebula Plan” was officially launched, opening Baidu Orion AI engine to the entire industry. MCP service revenue is shared with developers, and vendors can directly access industry modules. Mainstream mobile phone manufacturers such as Xiaomi, Honor, and vivo have already joined. Baidu Orion is a comprehensive AI search service solution for partners, relying on Wenxin Assistant with over 200 million active users monthly, through Master Agent and multi-skill orchestration, capable of calling Baidu search, e-commerce, travel, health, and other quality services, completing the loop from information acquisition to task delivery. In vertical fields, Baidu launched a health app co-construction plan covering over 300k doctors and more than 10k hospitals, providing full-cycle, companion-style proactive care services for users.

Tesla China vehicle system to connect to Doubao large model

According to “Cybersecurity Shanghai,” Tesla’s vehicle voice large model service completed filing on April 20. It is learned that Tesla’s vehicle voice service will connect to Doubao large model. Notably, the previous Tesla China official website’s “Tesla Vehicle Voice Assistant Terms of Use” indicated that the Model Y L vehicle will be equipped with Doubao large model and DeepSeek model, both accessed via Volcano Engine. Doubao large model will handle voice commands such as navigation setting, media control, air conditioning temperature adjustment, and also support owner manual queries; DeepSeek provides AI interaction services.

Huawei Cloud CodeArts code intelligent agent’s daily active users increase 18 times

According to insiders, since the public beta launched in February this year, Huawei Cloud CodeArts code intelligent agent’s user base has continued to grow, with recent data showing about an 18-fold increase in daily active users. Focused on enterprise-level AI programming scenarios, the product offers core capabilities including Skills, Codebase indexing, and norm-driven development. It is expected to launch new features this week, including Skill marketplace, custom agents, and multi-task parallelism.

Tencent Cloud releases Xinference supply chain poisoning risk notice

Tencent Cloud announced that its Security Center detected that Xinference has been disclosed to have supply chain poisoning risks. Attackers could steal highly sensitive information such as cloud credentials, API keys, SSH keys, encrypted wallets, database credentials, and environment variables when users install or import affected versions, sending them to remote command and control (C2) servers. To prevent business impact, Tencent Cloud Security recommends timely security self-checks and updates or fixes within the affected scope to avoid external attacks.

Xiaomi MiMo-V2.5 series models officially open public testing

Xiaomi MiMo-V2.5 series models officially open for public testing. The series includes MiMo-V2.5, V2.5-Pro, V2.5-TTS Series, and V2.5-ASR. Additionally, Xiaomi has adjusted the Token Plan pricing: canceled the 1 Token=4 Credits billing method, and TokenPlan no longer distinguishes between 256k and 1M context window Credit multipliers; from 00:00 to 08:00 Beijing time daily, all models’ Credits consumption rate is further discounted by 20%.

Unity China releases AI OS 3D spatial intelligent cockpit

Unity China officially launched AI OS 3D spatial intelligent cockpit at the 2026 (19th) Beijing International Auto Show. AI OS 3D aims to address the display challenges of atomized content and multimodal data in the AI era, integrating multimodal outputs into a unified spatial experience layer, providing automakers with a technical solution for spatial perception and AI agent interaction transformation, and promoting the evolution of smart cockpits from 3D HMI to spatial intelligent agent operating systems. Architecturally, 3D AI OS decouples AI capabilities from spatial experience, allowing automakers to independently build AI perception and system orchestration, with the engine focusing on spatial rendering and interaction presentation. Car manufacturers can freely choose large models and AI solutions, with core AI capabilities always under their control.

DeepSeek: After the batch launch of Ascend 950 super nodes in the second half, the price of V4-Pro model will be significantly reduced

DeepSeek-V4 preview version launched and open-sourced on the 24th, including DeepSeek-V4-Pro and DeepSeek-V4-Flash. According to DeepSeek official documentation, DeepSeek V4 has validated fine-grained expert parallel (EP) solutions on both NVIDIA GPUs and Huawei Ascend NPUs. Additionally, the official API page mentions that due to high-end computing power limitations, the service throughput of V4-Pro remains limited; after the Ascend 950 super nodes are mass-produced in the second half, the Pro price will be significantly lowered.

Deepseek official website releases deepseek-v4 API documentation

Deepseek’s official website now features deepseek-v4 API and model descriptions, including deepseek-v4-flash and deepseek-v4-pro.

BMW integrates with Alibaba Qianwen, launches cockpit AI intelligent agent

BMW’s customized cockpit AI intelligent agent based on Alibaba’s Qianwen large model made its debut at the Beijing Auto Show. Using the Qianwen large model as a foundation and based on Zebra Smart Spirit AI, BMW launched three China-market tailored AI agents—“Driving Expert,” “Travel Partner,” and “Encyclopedia Guru”—covering vehicle use, travel, and knowledge Q&A scenarios, truly achieving “understandable and manageable.” The BMW iX3 Long Wheelbase will be the first model equipped with this technology, followed by the new BMW 7 Series.

Qingzhou Zhihang publicly unveils “World Model + Reinforcement Learning” physics AI solution

At the 19th Beijing International Auto Show, Qingzhou Zhihang launched a physics AI model based on the unified architecture of “World Model + Reinforcement Learning.” The company will introduce the “Chuangfeng MAX” system with over 500 TOPS, adopting the VLA + World Model + Reinforcement Learning unified architecture.

Dai Jifeng’s conflict with Chen Tianqiao intensifies, true story of MiroMind’s departure exposed

Tsinghua University associate professor Dai Jifeng’s departure from MiroMind and his entrepreneurial return have reignited. On April 22, Dai Jifeng spoke publicly for the first time about his departure, claiming it was because MiroMind asked him to work overseas, which he refused. On April 23, MiroMind issued an internal statement denying this, fully exposing the conflict. The statement shows that Dai Jifeng joined in March 2025 and proactively applied to work in the US; the O-1 visa approved in November 2025 predates the “Manus incident.” On the night of January 15, 2026, Dai Jifeng proposed to resign and offered to exchange 15% equity in his new company for MiroMind’s intellectual property rights; later, his backer’s investment agency demanded to reduce the equity to 5% and requested free perpetual licensing. MiroMind refused, calling it “company employees’ shared property” that cannot be arbitrarily licensed. Currently, MiroMind claims to have evidence of infringement and defamation by Dai Jifeng’s team, has reported to authorities, and initiated legal proceedings.

Overseas News

Meta and CBRE launch “LevelUp” plan to build Meta’s data centers in the US

Meta Platforms and CBRE announced the launch of the “LevelUp” plan, a multi-year project to recruit and train thousands of technical personnel to build Meta’s data centers in the US.

Amazon and Anthropic to invest over $100 billion in AWS technology over the next decade

Amazon and Anthropic will invest over $100 billion in AWS technology over the next ten years, planning to expand AI inference capabilities in Asia and Europe. Today, Amazon announced an additional $5 billion investment in Anthropic, up to a total of $25 billion, with Anthropic’s Claude platform launching on Amazon Cloud.

OpenAI releases Images 2, ranked first by Arena.ai

Early this morning, OpenAI CEO Altman led a 20-minute online live broadcast announcing the launch of ChatGPT Images 2.0, an AI image generation tool based on GPT Image 2, with a core upgrade of “thinking ability.” After enabling “thinking,” the tool can retrieve information online, create visual analysis content based on uploaded files, and perform reasoning and planning of image structure before generation. This feature is now available to ChatGPT Plus, Pro, Business, and Enterprise subscribers.

Overseas large model evaluation agency Arena.ai posts that OpenAI’s GPT-Image-2 tops all Image Arena leaderboards and leads Google Nano-banana-2 by a record 242 points in the text-to-image domain.

Adobe expands “Agentic” ecosystem, partners with OpenAI and payment giants to accelerate enterprise AI deployment

Adobe announced expansion of its “Agentic” ecosystem. Besides launching CX Enterprise Coworker, the company confirmed deep cooperation with AWS, Anthropic, Google Cloud, IBM, Microsoft, NVIDIA, and OpenAI to extend AI agent workflows across enterprise segments. To achieve end-to-end “discovery to purchase” loops, Adobe also deepened partnerships with Adyen, PayPal, and Stripe, embedding payment functions directly into AI-driven interactions.

OpenAI launches Codex Labs

OpenAI reports that weekly active users of Codex jumped from 3 million to 4 million within two weeks. Partners include Accenture, Capgemini, CGI, Cognizant, as well as Infosys, PwC, and Tata Consulting. Codex Labs will enable OpenAI experts to directly join various organizations.

Meta invests $1 billion to build data center in Tulsa, US

It is reported that Meta has officially launched a $1 billion data center project in Tulsa, Oklahoma. The project includes over $25 million in local infrastructure improvements; at peak construction, it will create over 1,000 construction jobs and provide about 100 long-term operational positions upon completion.

Adobe announces $25 billion stock buyback amid stock price decline

Amid anxiety caused by AI disruption, Adobe’s stock price has fallen, and the company announced a $25 billion share repurchase plan over the next four years. CFO Dan Durn stated that this new buyback authorization “demonstrates our confidence in the company’s strong cash flow and our commitment to creating long-term value for investors.” The stock has declined 29% this year. This buyback plan comes as a follow-up to a $25 billion repurchase announced in March 2024.

LG AI Research Institute and NVIDIA agree to strengthen AI cooperation

On April 21, LG announced that LG AI Research Institute and NVIDIA have agreed to strengthen their technological alliance to expand the Exaone AI ecosystem. The two will combine LG’s Exaone model with NVIDIA’s Nemotron ecosystem to jointly develop AI models.

Meta’s bold move: collecting employee mouse and keyboard input data solely for AI training

According to reports citing industry insiders, Meta is installing new tracking software on its US employees’ computers to record mouse movements, clicks, and keystrokes for AI model training. This is part of a plan to build autonomous AI agents capable of performing work tasks. However, Meta’s latest move has also raised concerns among white-collar workers about potential monitoring during work.

US first to include AI in criminal investigations, regulatory urgency

Florida prosecutor James Ussmaier announced the launch of a criminal investigation into US AI company OpenAI and its chatbot ChatGPT, citing its potential role in the April 2025 Florida State University campus shooting. This is the first case in the US to include AI in criminal investigations.

IBM Q1 revenue $15.92 billion, up 9.5% YoY

IBM reported Q1 revenue of $15.92 billion, up 9.5% YoY, beating analyst expectations of $15.67 billion. Operating EPS was $1.91, compared to $1.60 last year, with analysts expecting $1.81. Free cash flow was $2.22 billion, up 13% YoY, versus expected $2.17 billion.

Google launches a series of new AI tools

Google announced a series of tools for building AI agents to help automate tasks, including a dedicated inbox for virtual robots to publish information and progress reports. The Gemini Enterprise Agent Platform will include new features like Memory Bank and Memory Profile to help agents remember past interactions. Another new feature, Agent Simulation, will help developers better test these tools before release. Google also announced Projects, a platform for employees to collaborate with colleagues and agents, integrating information from Workspace, Microsoft OneDrive, and company chat logs to enable agents to operate with proper context.

Meta internally announces 10% layoffs to improve efficiency

On April 23, local time, reports indicated Meta sent a memo to employees stating plans to cut 10% of staff to improve efficiency and offset high AI spending. The layoffs are expected to affect 8,000 people and will be implemented by May 20.

Intel releases new AI workstation platform

Intel recently launched a new AI workstation platform, including Intel® Xeon® 600 workstation processors and Intel® Arc™ Pro B70, B65 GPUs. The Xeon 600 processors are designed for demanding professional workloads, achieving breakthroughs in performance, scalability, AI, and management, with up to 86 cores, a 61% performance increase over the previous generation, and a maximum turbo frequency of 4.8GHz. Arc Pro B70, based on the second-generation Xe architecture, features 32GB memory, 32 Xe cores, and AI compute peak of 367 TOPS, suitable for graphics rendering, general parallel computing, and AI acceleration.

Nvidia CEO Huang Renxun urges all employees to use OpenAI Codex

OpenAI founder Altman shared an email from Nvidia CEO Huang Renxun, urging all 10k employees—including engineering, product, legal, marketing, finance, sales, HR—to actively use OpenAI’s Codex programming tools. He described Codex as a “teammate” and “superpower” for collaboration. Huang also revealed that Nvidia and OpenAI have collaborated on building a Codex lab within Endeavor for learning support and will hold online seminars to boost efficiency.

NEC and Anthropic announce cybersecurity cooperation

On April 23, Japan’s NEC announced a strategic partnership with Anthropic to accelerate AI application in Japanese enterprises. The cooperation will focus on “joint development of AI solutions tailored for Japan’s security industry.”

Funding and M&A

Vars Variable Robots completes nearly 2 billion yuan Series B financing, Xiaomi strategic investment leads

Vars Variable Robots founder and CEO Wang Qian confirmed that the company completed nearly 2 billion yuan in Series B funding from Xiaomi’s strategic investment between late March and early April. Recently, Meituan, Alibaba, and ByteDance led or participated in Series A, A+, and A++ rounds respectively. Vars Variable Robots has become the only domestic embodied intelligence enterprise invested in by four major internet giants.

Robot component R&D firm Quanshi Zhibo completes several hundred million yuan Series A++ financing, led by Shentou Holdings

Quanshi Zhibo, a robot component R&D company, completed several hundred million yuan in Series A++ financing, led by Shentou Holdings, with additional investments from Gofun Venture Capital, Dongfang Jiafu, Junshan Capital, Liangjiang Capital, and Junxi Capital. The funds will mainly support frontier technology exploration, product matrix expansion, and the enhancement of intelligent manufacturing and delivery capabilities, leading the next generation of robot technology and helping robots enter thousands of households.

Chinese inference GPU unicorn Sunrise secures over 1 billion yuan in new funding

Chinese full-stack self-developed AI inference GPU company Sunrise announced a new round of over 1 billion yuan in financing. Since the AI industry fully entered the “inference landing and intelligent agent popularization” era in 2026, this is one of the largest single financings in China’s GPU sector. Sunrise, which spun off as an independent company just over a year ago, has completed seven rounds of funding with a total of about 4 billion yuan, becoming China’s first unicorn valued over 300k yuan solely in inference GPUs. The funds will mainly be used for mass production and delivery of the new S3 inference GPU, full-stack software ecosystem development, and R&D iteration of subsequent chips S4/S5.

SpaceX announces right to acquire AI startup Cursor for $60 billion

SpaceX announced it has reached a cooperation agreement with AI startup Cursor, granting SpaceX the right to acquire the company later this year for $60 billion or to pay $10 billion to advance cooperation. Cursor is a rapidly growing AI startup focused on developing code-writing software.

Honeywell’s quantum computing company Quantinuum secretly files for US IPO

Honeywell’s quantum computing company Quantinuum has secretly filed for an IPO in the US. Honeywell has not disclosed financial details of this potential IPO, but Quantinuum was valued at $10 billion in a funding round last September.

Amazon-backed X-Energy raises $1.02 billion via IPO

On April 23, US-based nuclear reactor developer X-Energy, supported by Amazon, raised $1.02 billion in a scaled-up IPO, with the offering price above the previously announced range. The company sold 44.3 million shares at $23 per share. Previously, X-Energy had set the IPO price between $16 and $19 per share.

Google plans to invest up to $40 billion in Anthropic to support significant expansion of computing power

On April 24, Anthropic announced that Google committed to investing $10 billion in cash, valuing the company at $350 billion, matching its valuation at the February funding round (excluding recent additional financing). Additionally, Google will reinvest $10k when Anthropic hits performance targets and support its large-scale computing expansion.

Industry Trends

SpaceX drops “IPO blockbuster”: space AI tech unverified, may not be commercially viable

Reports indicate SpaceX warned investors in a document that its ambitions to build space AI data centers and establish human settlements on the Moon and Mars “depend on unproven technology” and may not be commercially feasible. The document, filed before the IPO, highlights business risks previously unreported. It also emphasizes its heavy reliance on the next-generation fully reusable Starship rocket, which has experienced multiple delays and test failures.

China’s software industry may surpass 20 trillion yuan

The China Software Industry Association released the “2025 China Software Industry Development Outlook Report” on the 21st, showing that China’s software industry has successfully met the “14th Five-Year Plan” development targets and has a solid foundation for new growth. Data shows that in the first two months of this year, China’s software revenue reached 256k yuan, up 11.7% YoY; total profit was 1M yuan, up 7.3%. The association’s Secretary-General Chen Baoguo stated that China’s software revenue grew from 8.16 trillion yuan in 2020 to an expected 15.48 trillion yuan in 2025, with an average annual growth rate of 13.8%, ranking among the top sectors of the national economy. During the “14th Five-Year Plan,” China’s software industry is expected to surpass 20 trillion yuan.

National Computer Virus Emergency Response Center: Beware of “Lobster” intelligent agent being poisoned

Recently, the National Computer Virus Emergency Response Center and the National Engineering Laboratory for Computer Virus Prevention and Control, relying on the National Computer Virus Collaborative Analysis Platform, discovered multiple malicious code-infected counterfeit skill packages in the “Lobster” (OpenClaw) intelligent agent system’s skill repository. Once users install these skill packages, malicious Trojan viruses will be silently downloaded and installed in the background when calling these skills, stealing privacy and causing sensitive information leaks or property losses.

Qihoo 360 releases “Mythos Incident White Paper”—deep defense system construction for government and enterprise networks

Qihoo 360 released the “Mythos Incident White Paper—Thoughts on Building Deep Defense Systems for Government and Enterprise Networks” (hereinafter “White Paper”). The White Paper points out that Mythos directly exposes a long-standing misconception among government and enterprise clients: the old approach of “finding vulnerabilities, patching, then finding again, and patching again” is no longer effective in today’s AI-driven, large-scale cyberattack environment. The reality is that vulnerabilities can never be fully patched; instead of chasing vulnerabilities, it’s better to acknowledge their existence and build robust defenses—this is the core idea of “vulnerability-tolerant protection.” To achieve this, constructing a “three-in-one” endogenous security system is the most feasible and systematic solution.

Shenzhen’s robot industry output hits a new high of over 240 billion yuan last year

On April 22, the “Shenzhen Robot Industry Development White Paper 2025” was released at the FAIR plus 2026 Robot Industry Link Conference. Data shows that in 2025, Shenzhen’s robot industry total output value reached 242.6 billion yuan, a year-on-year increase of 20.56%, setting a new record.

Industry Preview

The 2nd World Humanoid Robot Sports Games to be held in Beijing in August

The press conference for the 2nd World Humanoid Robot Sports Games was held in Beijing. Organized jointly by the Beijing Municipal Government and China Central Radio and Television Station, the event will be held from August 22 to 26 at the National Speed Skating Oval, extending the competition period to five days. The event includes two categories and 32 items, divided into competitive and scenario-based competitions. The competitive events include athletics, soccer, gymnastics, weightlifting, martial arts, street dance, sports dance, tug-of-war, and pitching; scenario-based events cover home, hotel, industrial, emergency, hospital, and commercial scenarios.

Quanmiao Silicon Valley announces “DREAME NEXT” global launch event from April 27-30

Quanmiao Technology announced that from April 27 to 30, 2026 (North American time), it will hold a “DREAME NEXT” themed global launch week in Silicon Valley. The four-day event will cover smart vehicles, smart appliances, smartphones, hardware, and personal care, showcasing Quanmiao’s latest breakthroughs in AI-driven, embodied intelligence, and bionic interaction, presenting a complete “human-vehicle-home” intelligent ecosystem. This is the first time in Silicon Valley history that a Chinese tech company has hosted an exclusive launch week.