DeepSeek rollout widens as China’s AI supply chain advances
China’s AI compute news centered on new foundation-model deployments, memory progress and data-center materials investment.
• DeepSeek said the official DeepSeek-V4-Flash API has entered public beta, and xFusion made DeepSeek-V4-Flash-0731 available on its AI Lab platform for online PC trials.
• Benchmark updates cited by IT Home said DeepSeek-V4-Flash achieved a roughly 98% cache hit rate, while Elon Musk followed DeepSeek’s X account after the release drew broad attention.
• Reports said some large U.S. companies are turning to Chinese large models to reduce costs, with Coinbase shifting to Chinese AI models and Airbnb adopting Alibaba models.
• Tesla China’s in-car system officially connected to ByteDance’s Doubao large model, while reports said ByteDance is dismantling Feishu as Doubao connects to Douyin and seeks more usage scenarios.
• Yaohong Electronics signed an agreement to build an AI data center high-speed motherboard core materials R&D and production base in Wuxi, with total investment exceeding 5 billion yuan.
• CXMT’s LPDDR6 is reportedly close to completing R&D validation, with its first product designed for a 12,800 Mbps data rate, 16 Gb die capacity and 1,295-ball PoP packaging.
• KTC showed a mobile smart display integrated with StepFun’s Step-series large model solution at ChinaJoy 2026, alongside another device described as a Doubao large model UMAX version.