According to supply chain sources, ChangXin Memory recently declined to quote Apple for DRAM supply, prioritizing the Chinese domestic market. In August, Apple's price-cut request for LPDDR5X and other mobile DRAM had been rejected. ChangXin's capacity is fully booked by long-term orders from domestic terminal makers.
OpenAI announced it will upgrade all grants under Codex for OSS to the Pro 200 tier, increase the number of grants in this round from 5,000 to 10,000, and encourage previous recipients to reapply.
Nvidia announced that multiple European cloud service providers and infrastructure companies are adopting its DSX AI factory platform to reduce data center cooling water and power consumption while accelerating compute deployment. The platform's closed-loop liquid cooling and dry coolers can cut cooling water use to near zero, saving 54 million liters per day at 2 GW scale, and DSX MaxLPS software can reduce power use by up to 30% for the same AI token output.
fal.ai today released FLUX 3 Max, a video model developed with Black Forest Labs, claiming 10-second clips can be generated in 12.45 seconds—about 22× faster than Seedance 2.5—along with a deep launch discount. Artificial Analysis' leaderboard shows Elo 1138, tied with Seedance 2.5, though overlapping confidence intervals leave quality superiority unclear.
At the AIIC 2026 main forum, Zhou Jun, chairman of Alexander Intelligent Technology, released GomROS-Brain, a brain-inspired architecture for the company's domestic robot operating system GomROS, detailing features including five brain-region mappings, sub-millisecond response via the Zhiyao high-speed bus, a 6MB minimal installation package, and deployment across more than 200 enterprises.
Black Forest Labs released FLUX.3 Image Klein, with day-one support in Diffusers. The model supports text-to-image and multi-reference image-to-image, and adds layout control by specifying bounding boxes in prompts. It ships two open checkpoints: a distilled 16-step no-CFG version and a base 50-step CFG version. The distilled checkpoint generates a 1024×1024 image in about two seconds on an H100, with NVFP4 and FP8 quantized weights and LoRA training scripts provided.
On October 13, CiXin Technology launched the AGX Station, a desktop AI supercomputing center positioned against NVIDIA's DGX Spark. Powered by its self-developed 6nm agentic SoC CiXin P1, it natively supports 7B-35B model inference, with expandable computing power up to 350 TOPS and support for 122B and 150B parameter models.
American Express has launched the Intent Intelligence Framework, designed to capture and verify consumers' purchase intent before AI agents complete transactions, ensuring the intent is carried through the entire transaction lifecycle. The framework is part of the ACE Developer Toolkit introduced in April, with the Create Intent API as its initial component.
Musk responded on X to the chip stock selloff triggered by ChangXin Memory's technology breakthrough, saying that once China figures out how to manufacture a product, its scaling capacity often exceeds that of the rest of the world combined, citing lithium batteries and solar as examples. He noted China is rapidly tackling universal standard specs like LP5 and LP6, while pointing out that HBM architecture, SRAM layers, and 3D-stacked RAM heat dissipation remain harder challenges.
On October 12, 2026, Hefei Fourth Law Robotics, a company backed by Xiangxin Technology, announced that its self-developed embodied AI model Axiom-0 ranked first on the public leaderboard of the RoboCasa365 Benchmark. Founded in June 2026, the company received a seed investment of over ten million yuan from Xiangxin Technology in August, and the two parties will cooperate around industrial scenarios such as automotive manufacturing and intelligent assembly.