South Korean FuriosaAI Unveils Third-Generation AI Inference Accelerator: 32x Improvement in Computing Power and Memory Bandwidth

Korean AI semiconductor startup FuriosaAI announced the technical details of its third-generation AI inference accelerator yesterday.

The accelerator was developed in collaboration with Broadcom, featuring two "mask limit-level" 2nm computing dies, one independent I/O die, 12 HBM4(E) memory stacks each with 48GB capacity, and using a SUE full interconnection topology within the rack, supporting PCIe Gen7.

FuriosaAI stated that the physical size of the chip is about eight times that of the second-generation chip (RGND); the computing power can reach 32 PFLOPS, 32 times that of RGND; the HBM memory capacity reaches 576GB, 12 times that of RGND; and the HBM memory bandwidth reaches 48TB/s, also 32 times that of RGND.

FuriosaAI, a South Korean AI Chip Startup, Rejects $800 Million Acquisition Bid from Meta

South Korean media reports that FuriosaAI, a startup specializing in AI application chips, recently rejected an $800 million acquisition offer from tech giant Meta. FuriosaAI stated it will continue focusing on independent R&D and production of its AI chips. The report indicates that the breakdown of acquisition talks stemmed from disagreements on business strategy and organizational structure, not the acquisition price itself. Similar to many tech companies building large language models (LLMs), Meta...

Huawei Cloud Ma Dao Fully Upgraded: The World's First HarmonyOS-Specific AI Coding Agent Launches, Providing a One-Stop Solution from Idea to Release

Huawei Cloud Ma Dao CodeArts has launched three core capabilities for HarmonyOS developers: the HarmonyOS Coding Large Model, Ma Dao HarmonyOS Intelligent Agent, and HarmonyOS Developer Practice Center. Officially called the world's first and only AI coding agent specifically designed for the HarmonyOS ecosystem, it acts as a full-cycle digital partner for developers. The core HarmonyOS Coding Large Model is enhanced through the HarmonyOS-enhanced training system, with improvements in training data, code generation, and compilation pass rate.

Tongyi Pingtouge Launches AI Chip Zhenwu V900 with Three Times the Computing Power of M890

At the 2026 Yunqi Conference, Tongyi Pingtouge launched the domestically produced AI chip Zhenwu V900, which has three times the computing power of its predecessor M890. A single cluster can be expanded to 500,000 cards, supporting advanced AI model training and inference. Alibaba stated that its chip product line is mature, with a significant increase in annual shipments expected. Wu Yongming revealed that the self-developed M890 super node is already capable of supporting large-scale model inference with up to 2 trillion parameters.

2026 First Half Global AI Glasses Shipments Surge 263%, Screenless Becomes Dominant Form

According to Counterpoint, smart glasses remain niche but are growing rapidly. Global shipments of AI glasses increased by 263% year-on-year in the first half of 2026. Most products have no AR screen and interact through speakers and microphones; screenless AI glasses, represented by Meta and Ray-Ban, dominate the market, while AR display glasses show even stronger growth.

Tencent Hunyuan Image 3.5 is here: more accurate text drawing, stronger image editing, a 2K image costs just 1.5 cents

On September 22, Tencent Hunyuan released the preview version of Image 3.5, with comprehensive upgrades in image quality, layout, and image editing capabilities, and a significant reduction in usage cost, featuring low cost and high efficiency. The official stated that the overall capability has improved by about 30% compared to the previous version, with more accurate text generation, avoiding missing strokes or characters; better understanding of visual layout needs; and enhanced image editing capabilities.