---
title: "Industry Map | China AIGC Industry Map V3.0 (2023)"
author: "Unique Research"
sourcePublication: "Unique Research (FFCAP English library)"
originalPublishedAt: "2023-09-06T18:00:00+08:00"
canonical: "https://ffcap.cn/en/research/20230906a0937a00"
source: "https://ffcap.cn/en/research/20230906a0937a00"
language: "en"
cnSourceUrl: "https://view.inews.qq.com/a/20230906A0937A00"
---

# Industry Map | China AIGC Industry Map V3.0 (2023)

_Unique Research · Insight / industry map · September 6, 2023 · Full framework rewrite of V3.0 map layers — company grids primarily in CN images_

**Contents:** 01 Map release · 02 Deep RL paradigm (AI 3.0) · 03 Compute, LLMs & new tools · 04 Infra companies · 05 General LLMs · 06 Middle layer · 07 General apps · 08 Industry apps/models · 09 Disclaimer.

## 01 Map release

As the AIGC chain densifies, Unique Research updates from **China AIGC Industry Map V2.0** (Q1 2023 collect) to **V3.0**. Prior stack: base / middle / application. V3.0 splits the model layer into **general LLMs** vs **industry models**, and splits applications into **industry** vs **general**.

General LLMs: broad-domain, transferable across industries. Industry models: specialized for vertical needs. Industry apps: healthcare, finance, education, transport, energy, etc. General apps: text/image/video/audio generation as shared primitives.

## 02 Deep reinforcement-learning paradigm — AI 3.0

AIGC fuses GANs, pretrained LMs, and multimodality—find patterns in existing data and generalize new content. Commercially: enablement via high-quality, high-freedom, low-barrier generation for content scenes/producers.

Early decision AI assumed logical purity—unfit for real-world uncertainty. Probabilistic paradigms (empiricism + rationalism) partially fix that. Deep RL with enough data + reward/penalty models approaches human-like intelligence for high-quality automated creation. General AI aims at pan-human task competence.

## 03 Compute, LLMs & new tools — new industry shape

**Base:** data, compute, compute platforms, model train/dev platforms, supporting facilities. **Model:** bottom general LLMs (open / closed / hosting) + middle (RLHF/human feedback, adaptation, personalization). **Application:** C-facing text/image/A/V generation split into industry vs general value logics.

Industry apps: security, finance, education, healthcare, driving, logistics, transport… General apps: translation, Q&A/CS/chatbots, KG, recommenders, robot OS, data mining, smart interaction…

Players: internet majors, AI cos, ecosystem/scene chains, research labs—boundaries still blur while the industry is young. **Application layer is the best entry for startups**—hence fiercest competition there.

**Compute as core pedestal:** AI chips (CPU base; GPU/FPGA/ASIC accelerators) · AI servers (CPU+accelerator for train/infer) · data centers (cloud training). Cost driven by chips/memory/disk; exploding datasets + parameter counts drive compute demand and cost.

**LLMs as paradigm milestone:** not mere algorithm bumps—new tech paradigm with stronger generalization, penetrating NLP, CV, healthcare, finance…

**New tool layer:** continuous emergence of orchestration, fine-tune, Agents, multimodal editors. **Industry enablement trend already visible** across verticals.

## 04–08 Company layers

Detailed named company grids for infrastructure, general LLMs, middle layer, general apps, and industry apps/models live primarily as **map graphics** in the Chinese original (4 images). On ingest, attach those figures so readers can navigate the V3.0 landscape company-by-company; this English package preserves the taxonomy and thesis so the map is usable without inventing names not clearly readable from text alone.

## 09 Disclaimer

Industry information for research reference; figures as of the V3.0 collection window. Not investment advice.

_4 CN map images; attach on ingest._

— END —

---

Editorial English edition published on UniqueCapital: https://ffcap.cn/en/research/20230906a0937a00
On-site reading page: https://ffcap.cn/en/research/20230906a0937a00