Alibaba unveils its T-Head-designed Zhenwu V900 AI accelerator, claiming it is China’s most powerful chip and can be clustered in systems of up to 500,000 units to train frontier models in a challenge to Nvidia.
Google opens U.S. preorders for the $899 Googlebook AI laptop ahead of its October 4 retail launch, pairing Android and desktop Chrome with Magic Cursor, Gemini Live, and 12 months of Google AI Pro with 5TB storage.
Alibaba unveils its Zhenwu V900 AI chip, calling it China’s most powerful, as CEO Eddie Wu says the T-Head accelerator delivers triple its predecessor’s performance and can scale to 500,000-chip clusters for frontier-model training.
Apple’s M5 Ultra Mac Studio and M6 Mac mini widen its local-AI hardware lead, with the Mac mini aimed at always-on agents and the top Mac Studio offering five times the memory bandwidth of Nvidia’s DGX Spark and AMD’s Ryzen Halo.
Google launches Googlebooks, Android-based laptops starting at $899 that retain ChromeOS while centering Gemini features such as Magic Pointer, Rambler voice-to-text and Create My Widget; HP, Dell, Lenovo, Acer and ASUS models are available for U.S. preorder ahead of October 4 availability.
AMD launches the $3,999 Ryzen AI Halo local AI platform, pairing a Ryzen AI Max+ 395 chip with 128GB unified memory to run models up to 200 billion parameters and claiming 4% to 14% faster LLM token generation than Nvidia’s DGX Spark.
mizorewww launches Laya-MLX, a native Apple Silicon runtime for Convai Innovations’ typed decision models that runs without text generation, PyTorch, or cloud APIs, reporting 7.39 ms median latency for its 322M multilingual model on an M3 Max.
Enactic launches OpenArm, a fully open-source 7DOF humanoid arm for physical-AI research, with a $6,500 bimanual system and a standardized OpenArm Cell for reproducible teleoperation, imitation-learning, simulation, and contact-rich real-world experiments.
Perplexity launches Portable Computer for compatible NVIDIA RTX PCs, bringing multistep AI-agent workflows locally to Windows machines with at least 24GB of VRAM while connecting to tools including Outlook, Gmail, Slack and GitHub.
Colibrì now runs MoE models as large as Kimi K3’s 2.8 trillion parameters on machines with just 25 GB of memory by streaming routed experts from disk, though cold GLM-5.2 inference falls to 0.05–0.1 tokens per second.