Unitree Technology’s STAR Market IPO Registration Takes Effect
The Shanghai Stock Exchange’s official website shows that the IPO review status of Unitree Technology Co., Ltd. on the STAR Market has been changed to “registration effective.”
SK Hynix Kicks Off U.S. Listing Roadshow, Could Rank Among Top 3 Largest IPOs in History
On July 6, SK Hynix officially launched its roadshow for a U.S. listing. According to the filing with the U.S. Securities and Exchange Commission, SK Hynix plans to issue American depositary receipts (ADRs) representing approximately 17.79 million common shares. Based on the closing price of its Seoul-listed shares last Friday, the offering size is about $28 billion. The filing shows that SK Hynix’s ADRs are expected to start trading on July 10. At the current proposed size, this ADR offering could rank among the top three largest IPOs in history.
Tencent’s Hunyuan Hy3 Released, Already Integrated into Multiple Tencent Businesses
On July 6, Tencent’s Hunyuan Hy3 was officially released. Hy3 is a model that integrates fast and slow thinking, using a Mixture of Experts (MoE) architecture with total parameters 295B, active parameters 21B, and a context length of 256K. It has already been integrated into multiple services such as WorkBuddy/CodeBuddy, Yuanbao, Marvis, and ima, with the API now available on Tencent Cloud TokenHub. Pricing is set at 1 yuan per million input tokens, 4 yuan per million output tokens, and 0.25 yuan per million tokens for input cache hits.
Meituan’s LongCat-2.0 Open-Sourced, with Inference Code for Domestic Chips Also Released
On July 6, Meituan officially open-sourced its trillion-parameter large model LongCat-2.0. With total parameters 1.6T and an average active parameter count of about 48B, the architecture introduces LongCat sparse attention and N-gram embedding to improve long-context processing efficiency and token-level representation, while dynamic activation further enhances code understanding, generation, and execution. This open-source release includes multiple precision versions such as BF16, FP8, and INT8, covering deployment needs across different computing platforms. Additionally, inference code optimized for domestic computing power is also open-sourced, meaning you can run LongCat-2.0 stably on existing hardware even without the latest accelerators.
Frost & Sullivan’s Full-Stack AI Cloud Services Report: Alibaba Cloud Leads with 40.1%, Baidu Cloud, Volcano Engine, and SenseTime Follow
On July 6, international market research firm Frost & Sullivan released its “2025 China Full-Stack AI Cloud Services Market Report,” showing that the total market size of IaaS, PaaS, and MaaS in China reached 59.59 billion yuan in 2025. Alibaba Cloud’s total revenue was 23.9 billion yuan, taking the top spot with a 40.1% market share, while Baidu Cloud, Volcano Engine, and SenseTime followed with 16.6%, 13.4%, and 7.1% respectively.
According to the report, in 2026 China’s daily AI token call volume has surpassed 140 trillion, a more than 1,000-fold increase in two years. The unit of computing consumption has evolved from “GPU hours” to “token volume,” and the pricing system and value proposition of cloud services are being reshaped. The report also notes that AI cloud services not only provide underlying computing power but also integrate model training, inference platforms, and Model as a Service (MaaS), driving the adoption of AI technologies across various industries.
