Infrawaves HK IPO at $1.3B Valuation

Post LinkedIn

🐼Read original on Pandaily

#ipo #china-ai #computing-clusterinfrawavesinfrawaves

💡China's top AI compute firm IPOs at $1.3B amid 304% growth – key for infra needs

⚡ 30-Second TL;DR

What Changed

Filed for Hong Kong IPO at USD 1.3 billion valuation

Why It Matters

This IPO highlights explosive growth in China's AI infrastructure market, potentially easing compute shortages for AI training. It positions Infrawaves as a key player, attracting investment and partnerships in AI hardware.

What To Do Next

Contact Infrawaves sales to benchmark their AI computing clusters against cloud providers for cost savings.

Who should care:Founders & Product Leaders

Key Points

•Filed for Hong Kong IPO at USD 1.3 billion valuation
•China's largest independent AI computing cluster provider by 2025 revenue
•Three-year revenue CAGR of 304.5%
•Achieved first annual profit in 2024

🧠 Deep Insight

Web-grounded analysis with 6 cited sources.

🔑 Enhanced Key Takeaways

•Infrawaves operates a full-stack AI infrastructure ecosystem branded as 'Galaxy,' which includes the Mercury AI network system, Venus AI operating system, and Venus AIDOC for cluster maintenance.
•As of April 2026, the company manages infrastructure supporting over 90,000 GPUs and 34,000 PFLOPS of compute capacity, maintaining a customer repeat purchase rate exceeding 80% between 2023 and 2025.
•The company's technical differentiation centers on 'decoupling' hardware and software to improve cluster utilization, including the development of the ICCL (Infrawaves Collective Communication Library) to address limitations in NVIDIA's NCCL regarding fault tolerance and observability.

🛠️ Technical Deep Dive

•Galaxy HPAC (High-Performance AI Computing) system: A self-developed, open-source intelligent computing stack focusing on communication, scheduling, and monitoring.
•Mercury-X: High-performance integrated AI network system, including support for 102.4T near-optical packaging white-box switches.
•ICCL (Infrawaves Collective Communication Library): A production-ready library designed to replace or augment NCCL, utilizing DPDK-like kernel bypass and zero-copy mechanisms to achieve 23.4% higher P2P throughput and 28.5% lower P2P latency.
•Fault Tolerance: Implements automated, 100% rapid fault diagnosis and automated operation and maintenance to mitigate NIC port failures in 400Gbps+ environments.

🔮 Future ImplicationsAI analysis grounded in cited sources

Infrawaves will face increased margin pressure as it scales beyond its initial 30-cluster deployment.

While the company achieved 21.8% gross margins in 2025, maintaining these levels while competing with large-scale internet and telecommunications operators for 10,000-GPU cluster market share will require significant capital expenditure.

The company's reliance on 'decoupling' strategies will be tested by evolving domestic Chinese chip architectures.

Infrawaves' business model depends on its ability to orchestrate heterogeneous hardware; any shift toward proprietary, closed-loop hardware ecosystems by major Chinese AI chip vendors could limit its software-defined value proposition.

⏳ Timeline

2023-09

Infrawaves founded in Beijing; completes seed round of financing.

2024-01

Achieves first annual profit.

2025-01

Completes Series A financing round.

2025-11

Announces completion of Series B financing; reports 30 delivered clusters and 76,000 managed GPUs.

2026-04

Files for Hong Kong IPO at a USD 1.3 billion valuation.

📎 Sources (6)

Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.

🐼Read original article on Pandaily

📰

Weekly AI Recap

Read this week's curated digest of top AI events →

👉Related Updates

Same topic

Explore #ipo

Same product