Driven by the national strategy of autonomous and controllable AI computing power, the domestic GPU sector has welcomed a major player again. AI inference-specific GPU chip company Sunrise announced that within one year of its establishment, it has completed nearly 3 billion RMB in financing, setting a new record for the largest early-stage funding scale among domestic AI chip startups. This round of financing has gathered rarely all three forces: industry capital, top VC/PE firms, and national-level funds, demonstrating the market's high recognition of its technical path and potential for domestic substitution.

The investors include:

- Industry capital: Huaxu Fund under the SanYi Group (focusing on high-end manufacturing and intelligent equipment), FantasAI (HKEX: 6682, a large model and AI platform enterprise), Hangzhou Data Group (a local data element infrastructure operator);

- Market-oriented VC/PE: IDG Capital, GaoRong Venture Capital (HKEX: 885413), WuJi Capital, etc.;

- State-owned capital funds: ChengTong Mixed-Ownership Reform Fund, etc.

Beyond this collaboration lies a clear strategic consensus: AI inference chips have become a key bottleneck for large model deployment, and the window for domestic substitution is rapidly closing. Sunrise focuses on high-performance, low-power inference scenarios, with its GPU architecture optimized for Transformer-type models. It aims to match international mainstream products in energy efficiency and deployment costs, targeting data centers, edge servers, and industry AI terminal markets.

According to the information, this round of financing will be mainly used for three directions:

1. Next-generation inference GPU chip R&D, improving the efficiency of low-precision computing such as FP8/INT4;

2. Building large-scale production capacity, establishing a complete supply chain from design to packaging and testing;

3. Building a software ecosystem, including drivers, compilers, operator libraries, and deep compatibility with mainstream frameworks (such as PyTorch, TensorRT, etc.).

Notably, Sunrise chose "inference-first" rather than pursuing both training and inference, reflecting a practical strategy — in the current scenario where training chips are still monopolized by big players, the inference market, due to fragmented application scenarios and strong willingness for domestic substitution, has become a more easily breakthrough "golden track" for domestic GPUs.

As large models enter the stage of "cost reduction and efficiency improvement," enterprises' demand for high-cost-effective, self-controlled inference computing power has surged. Sunrise's rapid rise not only fills the critical gap in the inference field of domestic AI chips but also marks that China's AI infrastructure is moving from "usable" to a new phase of "user-friendly and reliable." In this competition for computing power sovereignty, Sunrise has already secured 3 billion in "ammunition" and is ready to make a full sprint.