Unique Research · Interview / Talk · November 5, 2024
At Unique Capital’s AIGC Application Summit & AIGC100 (Beijing, Oct 17, 2024), Tencent Cloud enterprise industry architecture director Li Xiang presented “Cloud Intelligence: Building AI-Native Applications One-Stop.” Edited highlights:
What is an AI-native app?
AI at the core of develop / operate / serve — not bolted onto legacy. Four traits: autonomy, adaptability, data-driven decisions, predictive power. On Gartner’s AI maturity model, AI-native sits at stage 4 (systematic implementation) of five.
Trends (IDC): China GenAI market CAGR ~55.1% 2023–2027; GenAI software CAGR ~58.2%. 2023 = foundation-model year; 2024 = AI-native app year. Marketing & office lead (Tencent Ads precision; Meeting/Docs + Hunyuan). Agents with memory/reason/plan/act become primary landing form. Multimodal (GPT-4o, Sora-class) opens new product shapes.
Architecture & challenges
AI-native is a Tencent Cloud strategic direction alongside cloud-native — same tech line across public/private/distributed cloud. Typical stack: data processing, storage, quality evaluation, business systems. Pain points: model selection (cost-effective); model engineering (fine-tune, versioning, release); GPU cost; content compliance; document prep for training; efficient vector retrieval.
Tencent Cloud toolkit (talk highlights)
End-to-end path from model choice through engineering, knowledge/RAG, agents, and industry scenes. Internal AI coding assistant: ~50% daily coverage, ~40% less coding time, ~20% R&D efficiency lift. Real-time voice agents: RTC stack >3B uplink minutes/day in APAC enables natural LLM dialogue. (Full slide stack in CN images — attach on ingest.)
— END —