马斯克预测星链有望成为互联网骨干，V3卫星网络容量将大幅提升

简报：

马斯克表示如果SpaceX的星链卫星网络继续发展，最终可能承载大部分互联网流量，成为在线连接的主要骨干网。
根据SpaceX文件细节显示，星链V3卫星的下行链路容量将迎来大幅提升。
星链单颗卫星的下行链路容量将从V2版本的96Gbps提升至V3版本的1024Gbps。

相关链接：

Musk Predicts Starlink Could Become the Backbone of the Internet, with V3 Satellite Network Capacity Set for a Major Boost 🚀🌐

Brief:

Musk said that if SpaceX’s Starlink satellite network continues to develop, it could eventually carry most internet traffic and become the primary backbone for online connectivity.
According to details in SpaceX documents, the downlink capacity of Starlink V3 satellites will see a significant increase.
The downlink capacity of a single Starlink satellite will rise from 96 Gbps in the V2 version to 1,024 Gbps in the V3 version. 📈

面壁智能联合清华开源首个1.58-bit大模型BitCPM-CANN，8B版本可于手机端运行

简报：

近日，面壁智能联合清华大学及OpenBMB开源社区，正式发布并开源了中国首个基于华为昇腾平台训练的1.58-bit大模型BitCPM-CANN。
该模型提供0.5B、1B、3B和8B四个尺寸，成功实现了从量化算子到训练算法的全链路原生开发。
BitCPM-CANN在推理阶段可释放约6倍显存红利，使得8B参数量级的模型能够轻松运行在当前主流旗舰手机上。
面壁智能基于MindSpeed与Megatron-LM搭建了完整的低比特训练底座，涵盖32K长序列支持、融合算子等，为后续面向昇腾的训练工作提供公共基础设施。
目前该模型的所有权重均已开源，开发者可通过HuggingFace和ModelScope平台获取并进行创新应用开发。

相关链接：

面壁智能联合清华大学发布中国首个1.58-bit大模型BitCPM-CANN

ModelBest and Tsinghua Jointly Open-Source China’s First 1.58-bit Large Model, BitCPM-CANN; the 8B Version Can Run on Mobile Phones 📱

Brief:

Recently, ModelBest, together with Tsinghua University and the OpenBMB open-source community, officially released and open-sourced BitCPM-CANN, China’s first 1.58-bit large model trained on the Huawei Ascend platform.
The model is available in four sizes—0.5B, 1B, 3B, and 8B—and successfully achieves native end-to-end development, from quantized operators to training algorithms.
During inference, BitCPM-CANN can deliver about a 6× VRAM efficiency gain, enabling 8B-parameter models to run smoothly on today’s mainstream flagship smartphones. 🚀
Based on MindSpeed and Megatron-LM, ModelBest has built a complete low-bit training foundation covering 32K long-sequence support, fused operators, and more, providing shared infrastructure for subsequent Ascend-oriented training work.
All model weights have now been fully open-sourced, and developers can access them via HuggingFace and ModelScope for innovative application development. 🌟

阿里推出全能智能体基座Qwen3.7-Max，全面接入千问多端平台

简报：

5月22日，阿里千问全新一代智能模型Qwen3.7-Max正式接入千问App（需更新至6.9.7及以上版本）、PC端和网页端。
Qwen3.7-Max定位为全能智能体基座，支持编写调试代码、办公流程自动化，并能在长达35小时、超1000次工具调用的长周期任务中稳定自主执行。
该模型具备优越的跨框架泛化能力，能够兼容并稳定运行在Claude Code、OpenClaw、Qwen Code等多个开发框架中。
目前，开发者和企业已可通过阿里云百炼平台及API调用该模型服务。

相关链接：

Alibaba Launches the All-in-One Agent Foundation Qwen3.7-Max, Fully Integrating It Across the Qwen Multi-Platform Ecosystem 🚀

Brief:

On May 22, Alibaba Qwen’s new-generation intelligent model, Qwen3.7-Max, was officially integrated into the Qwen App (requires updating to version 6.9.7 or later), as well as the PC and web versions.
Positioned as an all-in-one agent foundation, Qwen3.7-Max supports code writing and debugging, office workflow automation, and can autonomously and reliably execute long-cycle tasks lasting up to 35 hours with more than 1,000 tool calls.
The model demonstrates strong cross-framework generalization capabilities, enabling it to operate compatibly and stably across multiple development frameworks, including Claude Code, OpenClaw, and Qwen Code.
At present, developers and enterprises can already access the model service through the Alibaba Cloud Bailian platform and via API calls. 🤖

资本狂涌AI赛道：一季度全球吸金超千亿，国产大模型加速规模化与商业变现

简报：

全球AI创投热度攀升，今年一季度共发生近600起融资事件，总额突破1100亿元，同比激增185.4%。
5月本土创投动能强劲，月之暗面、阶跃星辰等国产大模型企业接连获超300亿元注资；维他动力、鹿明机器人等具身智能初创企业也密集斩获数亿元融资。
创业公司融资主要流向三大方向：达到数十亿元规模的底层研发投入、算力基础设施建设（GPU采购与云服务租赁占融资额30%-50%）以及全球顶尖人才争夺。
截至2026年，中国大模型技术迭代周期已普遍缩短至3个月内，推理成本大幅下降，加速了AI产业向“规模化应用”演进和商业化变现。
DeepSeek估值已飙升至450亿美元，拟融资规模高达700亿元，宁德时代、京东、网易等巨头正接洽或考虑参投；同时，DeepSeek宣布其V4-Pro API将于6月1日起永久降价至原价的四分之一。
Zoom披露其2023年对AI初创企业Anthropic的5100万美元早期投资，现已增值至近13亿美元，获得了超十亿美元的财务回报。

相关链接：

Capital Floods into the AI Race: Global Funding Surpasses RMB 110 Billion in Q1, as Chinese Foundation Models Accelerate Scaling and Commercial Monetization 🚀

Briefing:

Global AI venture investment continues to heat up. In the first quarter of this year alone, nearly 600 financing deals were completed, with total funding exceeding RMB 110 billion, marking a year-on-year surge of 185.4%.
In May, domestic venture momentum remained strong. Chinese foundation model companies such as Moonshot AI and StepFun successively secured more than RMB 30 billion in fresh funding; embodied intelligence startups including Vita Dynamics and Luming Robotics also frequently raised hundreds of millions of yuan. 🤖
Startup financing is mainly flowing into three key areas: underlying R&D investment reaching tens of billions of yuan, computing infrastructure construction (with GPU procurement and cloud service leasing accounting for 30%–50% of total funding), and competition for top global talent.
By 2026, the technology iteration cycle for China’s large models has generally been shortened to within three months, while inference costs have dropped significantly, accelerating the AI industry’s shift toward “large-scale application” and commercial monetization.
DeepSeek’s valuation has soared to US$45 billion, with a planned fundraising scale of as much as RMB 70 billion. Giants such as CATL, JD.com, and NetEase are in talks or considering participation. At the same time, DeepSeek announced that its V4-Pro API will be permanently reduced to one-quarter of its original price starting June 1. 💡
Zoom disclosed that its early 2023 investment of US$51 million in AI startup Anthropic has now appreciated to nearly USdJyH6kDkfC5RlM16.answerText.3 billion, generating a financial return of more than USdJyH6kDkfC5RlM16.answerText billion.

目录

马斯克预测星链有望成为互联网骨干，V3卫星网络容量将大幅提升

Musk Predicts Starlink Could Become the Backbone of the Internet, with V3 Satellite Network Capacity Set for a Major Boost 🚀🌐

面壁智能联合清华开源首个1.58-bit大模型BitCPM-CANN，8B版本可于手机端运行

ModelBest and Tsinghua Jointly Open-Source China’s First 1.58-bit Large Model, BitCPM-CANN; the 8B Version Can Run on Mobile Phones 📱

阿里推出全能智能体基座Qwen3.7-Max，全面接入千问多端平台

Alibaba Launches the All-in-One Agent Foundation Qwen3.7-Max, Fully Integrating It Across the Qwen Multi-Platform Ecosystem 🚀

资本狂涌AI赛道：一季度全球吸金超千亿，国产大模型加速规模化与商业变现

Capital Floods into the AI Race: Global Funding Surpasses RMB 110 Billion in Q1, as Chinese Foundation Models Accelerate Scaling and Commercial Monetization 🚀