Huawei speeds up AI chip roadmap
Huawei Technologies Co. Ltd. has accelerated the rollout of its next-generation AI chips and introduced what it calls the industry’s first “super node” utilizing near-packaged optics (NPO) technology, stepping up efforts to overcome data-transfer bottlenecks in training AI models. Speaking Thursday at the annual Huawei Connect conference in Shanghai, rotating chairman Wang Tao said the Chinese tech giant is aiming to deliver computing clusters that operate logically as a single machine. To meet demand, Huawei has moved up the launch timeline for its Ascend 960 series. The Ascend 960 PR chip — tailored for prefill and recommendation tasks — is now scheduled for release in the first quarter of 2027, three quarters ahead of its original schedule.
China’s AI hiring surges as industry pivots to practical applications
Job openings for AI roles in China surged nearly eightfold in the first seven months of 2026, driven by an industrywide shift from training foundational models to deploying commercial applications, according to a report released Wednesday by professional networking platform Maimai. Postings across new-economy sectors jumped 789.5% year-on-year, while average monthly salaries for newly listed positions edged up 2.7% to 63,160 yuan ($9,406). Beijing, Shanghai, and Shenzhen led in overall AI listings, but secondary hubs recorded the steepest growth, with postings in Hefei jumping 1,570.4%. The hiring surge reflects a broader pivot toward commercialization. Application development accounted for 61% of all new AI postings, compared with 24.6% for foundational technology and 14.5% for business-efficiency roles. Maimai founder and chief executive Lin Fan said demand is now concentrated in talent capable of building intelligent agent applications, while foundational model research has largely consolidated among a few well-capitalized tech giants.
Z.AI says GLM-5.3-Flash achieves 3.2-fold performance boost in two weeks
Chinese AI startup Z.AI Co. Ltd. announced on Thursday that its GLM-5.3-Flash inference system achieved a major breakthrough on a cluster of domestic AI accelerators. Deployed in just two weeks from initial run to stably handling production traffic, the system increased end-to-end throughput by 3.2 times, bringing hardware utilization and per-token costs in line with mainstream international GPU platforms. The deployment was based on a cluster of over 100,000 domestic AI accelerators. The core innovation of the project was the introduction of an Infra Agent driven by GLM-5.3, which participated in performance bottleneck analysis and optimization planning, and autonomously completed some code modifications.
Tencent’s WorkBuddy launches full-stack application generation capability
Tencent Holdings Ltd.’s WorkBuddy launched a full-stack application generation capability on Thursday. The initial phase supports the creation of full-stack web applications, allowing users to generate complete website applications equipped with cloud databases, file storage, user registration and login, and AI calling capabilities simply by describing their needs in natural language. The applications can be published online with a single click, generating a directly shareable access link.
MiniMax joins Singapore’s AI skills popularization program
Singapore’s Skills and Workforce Development Agency (SWDA) rolled out over 200 AI courses on Wednesday and offered eligible participants a six-month free subscription to advanced AI tools. Three products from Chinese AI startup MiniMax Group Inc. — MiniMax Agent, Hailuo AI and MiniMax Audio — were included in the program through the Singtel AI Pass, providing supporting tools for the local public to learn and use AI.
Doubao releases AI-native auto assistant
An AI-native auto assistant jointly developed by ByteDance Ltd.’s Doubao and its cloud service platform Volcengine was officially released on Thursday. The assistant can perceive vehicle signals and integrate with vehicle control, navigation, assisted driving and entertainment systems to coordinate overall vehicle functions.
