Why AI Large Models Are Driving the Next Industrial Revolution

Background

Artificial General Intelligence (AGI) is still in the research phase, and large‑scale models are considered a crucial pathway to achieving it. These models are first trained on massive datasets and then adapted to a variety of downstream tasks through prompting, instruction fine‑tuning, and human‑feedback loops. The industry has shifted from "big model training" to "refining big models," sparking a multimodal and multi‑scenario revolution that raises the ceiling of model capabilities and significantly boosts application value.

Current Landscape

Global leadership: GPT series has accelerated its iteration cycle, moving from GPT‑1 to GPT‑3.5 in just over four years, and launching GPT‑4 within a year. The latest versions support multimodal inputs and have built an ecosystem of GPT‑based custom models (GPTs), making AI tools low‑cost, low‑threshold, and widely accessible.

Domestic competition: Chinese technology firms—including SenseTime, Du Xiaoman, DiPu, as well as cloud giants Baidu, Tencent, and Alibaba—have entered the large‑model arena. They leverage advantages in deployment timing, infrastructure, and application scenarios, capturing the majority of the Chinese market for general‑purpose models.

Hot applications: The rapid development of ChatGPT and other large models has produced breakout applications across dialogue systems, image generation, education, and office productivity.

Future Outlook

Content transformation: AI large models with universal, multimodal, high‑parameter, and high‑quality generation capabilities are becoming "content factories" and "assembly lines." With the emergence of GPTstore, an "App Store" for AI models is forming, expanding the commercial prospects of AIGC.

Model evolution: Multimodal models aim to emulate human brain information processing, offering more comprehensive understanding and generation. Core foundational models are the current focus, while intermediate‑layer models have yet to see significant domestic development.

AIGC Revenue Models

The industry’s revenue streams can be grouped into four categories: Model‑as‑a‑Service (MaaS), pay‑per‑output, software subscription, and custom model development fees. Pay‑per‑output dominates today, but MaaS is expected to become the primary growth driver as the underlying model ecosystem matures. According to QuantumBit forecasts, the AIGC market was about ¥170 billion in 2023, is projected to double by 2026, and could exceed ¥1 trillion by 2030.

Conclusion

AI large models are transitioning from experimental research to industrial‑scale production tools, reshaping content creation, driving multimodal innovation, and establishing new economic models that promise massive market expansion in the coming decade.