News
WAIC’s big release! Yuntian Lifei reveals the complete computing power map of chips, Wanka clusters and software stacks
2 min read
Source: zhidx.com
Zhidongxi Author | Cheng Qian Editor | Mo Ying Zhidongzhi reported on July 19 that yesterday, Yuntian Lifei released the AI inference infrastructure roadmap at the Shanghai World Artificial Intelligence Conference (WAIC 2026). This also means that Yuntian Lifei has completed its identity transformation from a single supplier of AI inference chip products to a builder of system-level AI inference infrastructure, and the domestic inference computing power competition has entered an era of system decisiveness. This roadmap includes three cloud-based high-computing inference chips, DeepVerse100P, DeepVerse100D, and DeepVerse100L, as well as a separate AI inference infrastructure based on these three chips for Wanka heterogeneous clusters, as well as the IFWA software stack that runs through the entire link, forming a complete computing power system with software and hardware collaboration. This year's WAIC clearly reflects the core trend of the AI industry. AI is no longer limited to general conversation tools. Agents that perform multiple tasks autonomously have become the standard for enterprise digitalization. Device-cloud collaboration and local device-side large-scale models have been implemented on a large scale. Computing power competition has shifted to full-stack system capability competition among chips, clusters, and software collaboration... Under the multiple changes on the application side, the industry has paid increasing attention to the two key indicators of token generation efficiency and unit token cost of AI computing power. In this context, along with the new AI inference roadmap, Yuntian Lifei has announced the long-term goal of “tens of billions of tokens for one cent”. This goal is also the general consensus in the current AI computing power industry. 1. Three inference chips + Wanka heterogeneous cluster, Yuntian Lifei unveiled a new blueprint for AI inference base. Large models have entered the stage of large-scale implementation, and the focus of industry competition has shifted from the peak computing power of a single chip to the end-to-end inference performance of the cluster. We can see from Yuntian Lifei’s latest layout plan that this change in competitive logic is reshaping the breakthrough path of domestic AI chips. The first is the underlying chip. Yuntian Lifei will launch three cloud-based reasoning chips with high computing power in response to the AI reasoning needs in the next two years: DeepVerse100P, DeepVerse100D, and DeepVerse100L.