Enterprise-grade MaaS Inference Optimization and Edge Deployment Platform

📌 Background

As the price war for large models intensifies, reducing token costs through inference optimization has become critical for the survival of MaaS providers. Zhipu AI's first earnings report post-listing shows revenue exceeding 724 million RMB in 2025, a year-on-year increase of 131.9%. Its MaaS API platform ARR reached approximately 1.7 billion RMB, a 60-fold increase year-on-year, with the platform's gross margin rising from 3.3% in 2024 to 18.9% (based on company financial reporting). Inference optimization and edge deployment have thus been validated as a scalable business model.

👤 Target Customers

Enterprises developing AI applications, AI Agent developers, and manufacturers requiring intelligent hardware with edge deployment capabilities.

💰 Revenue Streams

1) API usage fees based on token throughput or call volume; 2) Annual fees for dedicated enterprise deployments and edge node hosting services; 3) Ongoing maintenance packages: Annual subscriptions for version iterations, technical support, and operational assurance for edge nodes.

🧮 Cost Structure

GPU computing power procurement and depreciation, model R&D and MLOps team expenses, network and bandwidth costs.

🛡️ Moat

Superior inference optimization technology that lowers unit token costs, and bargaining power derived from large-scale computing power pools.

🔑 Keys to Success

• Extremely low model inference costs
• Building a rich ecosystem of open-source models
• Ensuring low-latency operation on edge devices under high concurrency

⚠️ Risks

• Profit margin compression due to price wars
• Risk of supply chain disruption for computing power
• Loss of key accounts to self-built computing infrastructure

🏢 Cases

• SiliconFlow
• Zhipu MaaS
• Tencent Cloud EdgeOne

📊 SWOT Analysis