ZhenFund angel project "Zhizi Xinyuan" closes new round of funding worth tens of millions of RMB, doubling down on AI for Computing

ZhiZi Xinyuan recently announced the completion of a tens-of-millions-of-yuan angel+ funding round. This round was led by Dingfeng Chuangke, Inno Chuangke Fund, and Shoucheng Capital, with existing investor Tongchuang Weiyue making an additional oversubscribed investment. As ZhiZi Xinyuan's first institutional investor, ZhenFund has firmly supported the team since the company's founding in 2025, witnessing its evolution from a technical concept to industrial deployment, and continues to believe in the team's long-term value in computational infrastructure.

This funding round will support continued R&D of the automated compute acceleration platform KernelCAT, the development of domestic computing ecosystem adaptation capabilities, and expansion of the core technical and engineering teams. ZhiZi Xinyuan will continue advancing its AI for Computing direction — AI optimization for computing systems — pushing computing systems from "usable" toward a new stage of "efficiently optimizable."

The Critical Window for AI for Computing

From ZhiZi Xinyuan's perspective, three key trends make AI-driven compute optimization inevitable:

First, rapid hardware and software evolution. From conversational AI to the Agent era, computational workloads have undergone massive changes, with complexity continuously escalating.

Second, extreme scarcity of senior talent capable of HPC, compiler, and operator optimization.

Third, the complexity of computational tasks has approached or even exceeded the limits of manual handling.

In the AI era, computing demand is growing rapidly, yet hardware process technology is approaching physical limits. Relying solely on hardware iteration can no longer satisfy demand. The gap in compute supply is becoming increasingly pronounced, and only through global optimization of the computing process via AI can this impasse be broken.

This is precisely the starting point for ZhiZi Xinyuan's choice of the AI for Computing technical path — "buying chips does not equal owning usable compute power." ZhiZi Xinyuan must reconstruct the efficiency boundaries of computing through software and AI.

From scientific research to industrial production, all high-value industries are being redefined by computational capability. The faster, cheaper, and more efficient the computing, the faster the pace of industrial innovation. Continuously improving computational efficiency is the key to unlocking next-generation scientific, industrial, and intelligent productivity.

Current model vendors generally prioritize performance within the CUDA ecosystem, which further highlights the urgency for domestic computing stacks to evolve from "usable" to "good to use." Only by breaking down hardware-software barriers through AI optimization can domestic computing power truly meet industrial demands.

KernelCAT: Automated Compute Acceleration Platform

Based on this assessment, ZhiZi Xinyuan has built its core product — KernelCAT, an automated compute acceleration platform.

Rather than relying on throwing people or rules at the problem, it employs China's first technical paradigm combining "large models + operations research optimization + automated algorithm discovery" to build compute acceleration agents capable of automatic design, execution, verification, and iteration.

KernelCAT has formed a complete closed loop from problem understanding, strategy generation, operator optimization to hardware verification — transforming the compute acceleration process, which was previously highly dependent on expert experience, into an automatically executable, continuously evolving system-level optimization capability. This covers operator development, model migration, framework tuning, and inference acceleration at multiple levels.

In multiple mainstream benchmark tests, KernelCAT has demonstrated generalization capabilities beyond single-point optimization, achieving leading results across multiple mainstream benchmarks.

It can not only reuse underlying optimization logic across different hardware and tasks, but also autonomously explore new solutions based on mathematical reasoning. For example, in one test task, the system discovered that a certain operator failed to meet precision requirements in large-value-range and special-value scenarios. Without human prompting, it autonomously adopted polynomial approximation to reimplement the solution, and through self-iteration ultimately achieved the required precision.

This automation capability has already begun entering real industrial environments. Operators generated and optimized by KernelCAT have been merged into Ascend's official CANN operator library. On multiple domestic chips, it has automatically completed migration and adaptation of mainstream large models, significantly shortening the cycle compared to manual approaches while achieving precision alignment and performance improvement.

As underlying computational infrastructure, KernelCAT currently serves key domains including AI computing, scientific computing, and industrial simulation, continuously helping customers transform "paper compute power" into genuinely usable computational efficiency.

As one of the earliest institutions to support ZhiZi Xinyuan, ZhenFund continues to be optimistic about the AI for Computing direction. As computing becomes the underlying variable for all high-value industries, continuously improving computational efficiency means seizing the initiative for next-generation productivity. We look forward to ZhiZi Xinyuan continuing to drive paradigm-level upgrades in underlying computational efficiency through AI for Computing.