Alibaba Cloud's Tongyi Qianwen (Qwen) team released a new round of open-source results on October 4, including the Qwen3-VL-30B-A3B-Instruct and Thinking models along with their FP8 quantized versions, and launched an FP8 edition of the ultra-large-scale Qwen3-VL-235B-A22B model.
The releases underscore the continued acceleration of open-source model iteration in China, which has not been slowed by overseas safety debates. Around the same period, OpenAI updated its model line with GPT-6 Sol, Luna and Opus 5.5, followed by GPT-6.1 Sol, which approaches the capability of GPT-6 Astra at only one-fifth the cost. Anthropic, meanwhile, deepened enterprise partnerships with Salesforce and Bain to push agents into corporate workflows.
On the infrastructure side, Tencent signed an overseas leasing agreement with Oracle on October 2 covering roughly 100,000 AI chips over a five-year term, using Oracle data centers across Southeast Asia, in a deal valued at approximately 7 billion USD including about 30% prepaid. Tencent's capital expenditure for 2026 Q2 reached 52.8 billion yuan, up 176% year-on-year.
Chinese internet giants' capital spending has clearly accelerated, while overseas cloud vendors continue to see revenue growth quicken. According to second-quarter 2026 earnings, Amazon AWS posted revenue of 42.2 billion USD, up 36.7% year-on-year and accelerating 8.7 percentage points from the prior quarter. Microsoft Azure and Google Cloud grew 43% and 82% respectively, both faster than AWS and both accelerating further. Oracle's cloud infrastructure business grew 93%, with remaining performance obligations increasing substantially.
Domestic cloud vendors are showing a similar pattern of accelerating growth, supporting the view that AI capital expenditure expectations are converting into fundamental earnings verification. The focus of market discussion has shifted from whether demand will materialize to how large it will be.
Analysts noted that domestic AI agents still have room to drive a new round of annual recurring revenue growth by raising task density per user and lifting Token call volumes. The overall domestic AI sector trend is upward, with attention drawn to compute services, domestic chips and supernodes, as well as B-end AI application vendors with scenario, data and enterprise delivery capabilities.
