Axera Semiconductor unveiled its AX8860 series high-compute AI chips, featuring a 16-core Arm Cortex-A55 CPU and proprietary NPU with 256 TOPS mixed-precision compute, up to 128GB LPDDR5X memory, 6nm process, PCIe 5.0 support, and multi-chip scaling up to 1024 TOPS. It also launched the A256S module and showcased Yuanxi inference cards and servers. On the Qwen 27B model, the chip's prefill speed is 25% faster than NVIDIA DGX Spark, and system pricing is close to half of DGX Spark's.
At a media interview after QiYuan Robotics' launch, CEO Tian Hua and chief scientist Dong Hao addressed the company's entry into home scenarios, the generalization challenge in embodied AI, positioning, and mass production. QiYuan aims to become a leader in consumer robotics, with future revenue from hardware sales, content subscriptions/IP, and maintenance services. R&D spending exceeded 160 million yuan in H1 and is expected to exceed 500 million yuan for the full year; one robot rolls off the production line every 2.5 minutes.
Spanish AI company Multiverse Computing announced on X the release of its large language model Quasar 1.1 438B, claiming it is the first AI model trained on data generated by quantum computing. Netizens questioned its European origin, noting that the model is based on China's Zhipu AI GLM-5.2, which the company's website later confirmed, sparking criticism and mockery.