China Unicom expands AI compute as model and chip activity picks up
China's AI compute news centered on carrier infrastructure, model platform updates, domestic chip supply-chain moves, and reported large-model financing.
• China Unicom said its AI computing capacity has reached 45 EFLOPS across more than 1.1 million standard racks and seven 100-megawatt-class AIDC parks, with more than 17.5 billion yuan of this year's capex earmarked for computing power-related projects.
• Alibaba set its 2026 Apsara Conference for Sept. 22-24 in Hangzhou, with forums covering chips, cloud infrastructure, model capabilities, model services and agentic applications.
• Alibaba's Qwen-Image-3.0 launched on the Qianwen AI platform, with Pro and Standard versions opening API access and text-to-image pricing starting at 0.18 yuan per image.
• DeepSeek has reportedly restarted a second funding round seeking 50 billion yuan at a pre-money valuation of about 500 billion yuan, with signing planned for late August.
• ByteDance's Seed unit released SeedRealtime, a full-duplex audio-video large model that is now live in Doubao.
• Wuwen Xinqiong and MiniMax signed a strategic cooperation agreement covering model inference efficiency optimization, large-model token services, AI infrastructure and application scenarios.
• Samsung Electronics and SK Hynix are reportedly evaluating etching equipment from China's AMEC for possible use at their factories in China as they hedge against tighter U.S. export controls.
• Biwin Storage and CIX Technology agreed to establish a joint laboratory focused on storage-compute collaboration, storage chips, high-performance intelligent-agent CPU chips, coordinated product development and software-hardware compatibility.