---
title: Infra Completes 2.3 Billion CNY Series C Funding
url: https://www.elseif.net/infra-completes-23-billion-cny-series-c-funding
published: 2026-10-15T16:04:50+00:00
language: en
section: Harnesses
source: https://www.qbitai.com/2026/10/502767.html
organizations: Infra, NVIDIA, AMD, Huawei, Tencent, Baidu
publisher: elseif
---

# Infra Completes 2.3 Billion CNY Series C Funding

Infra has completed a Series C funding round of 2.3 billion CNY. This brings the company's total funding to 4.3 billion CNY, setting new records for both single-round and cumulative financing within the domestic AI native infrastructure sector.

Investors in the company include state-owned institutions such as Hangzhou Gaoxin Jintou, Social Security Zhongguancun Fund, Zhuhai Technology Industry Group, Xiamen Guoxing Investment, Xuhui Capital, Futeng Capital, China Insurance Investment, and China Development Bank Innovation. Industrial investors include Dabanc Holding, Lenovo, Jiukun Venture Capital, Tongge Venture Capital, Tencent, and Baidu. Financial institutions involved are Sequoia China, Hony Capital, Huiyuan Capital, Qiming Venture Partners, Fortune Capital, Shunwei Capital, and Lisi Capital.

The company operates as a layer between models and chips, focusing on connecting the two to optimize efficiency and costs across various model algorithms and chip hardware. It does not train its own models or develop applications. Instead, it aggregates heterogeneous and fragmented computing resources to produce low-cost, production-grade tokens.

Infra's technical system is based on diverse heterogeneity, software-hardware synergy, and autonomous AI. Its platform has adapted 16 types of chips, including those from NVIDIA, AMD, Huawei Ascend, Biren, Haiguang, Moore Threads, Moore Threads, MetaX, and Enflame, serving 37,000P of computing power. The company claims it can adapt a new chip within two working days, reducing adaptation costs for domestic chips by over 70 percent.

The company has achieved mixed training of heterogeneous chips at a scale of one thousand cards with a computing power utilization rate of 97.6 percent. It completed mixed training for a 70 billion parameter large model using four domestic chips, Huawei Ascend, Tianshu Zhixin, MetaX, and Moore Threads, alongside AMD and NVIDIA. Additionally, it collaborated with China Telecom to complete a cross-domain mixed training of a 30 billion parameter model over 4,000 kilometers between Hami, Xinjiang, and Shenzhen, Guangdong, which improved performance by 165 percent.

Infra has also implemented a cross-cluster heterogeneous PD separation architecture called PDD, which reduced first token latency by 51.5 percent and single token cost by 37.5 percent. The company uses an autonomous AI mechanism that dynamically switches between large models for planning and small models for real-time responses. It has deployed a swarm of agents, including systems for platform management, cluster operations, and chip operator generation.

The company's strategy consists of three layers: an Agentic Infra autonomous infrastructure platform for computing resource distribution, an Agentic MaaS large model service platform called the Token Factory, and an AI productivity store for industry solutions. The Token Factory has partnered with large model companies such as Zhipu and Moonshot AI. Its industry solutions cover fields including entertainment, gaming, healthcare, legal services, and energy.

Reports indicate that Infra has confidentially submitted an IPO application to the Hong Kong Stock Exchange in recent months.