Skip to content
UniqueCapital

UNIQUE RESEARCH / ENGLISH ARTICLE

Industry Report | April 2023 Global AIGC Monthly

Unique Research · Insight / industry monthly · May 9, 2023 · Full rewrite — not a stub

April flash: Global AIGC financing ¥2.642B / 14 deals; B+ (incl. strategic) = 96.37% of amount; largest OpenAI $300M (Tiger, Sequoia, a16z, Thrive, K2, Founders Fund…). China: ¥580M / 10 deals; largest Baichuan Intelligence $50M; China avg ~¥57.6M. Only 3 global rounds >¥100M (25%).

01.1 Global headlines

  • Microsoft (Apr 12): open-sources DeepSpeed Chat — train ChatGPT-class LMs with RLHF; claimed >15× train speed at much lower cost.
  • Alphabet (Apr 20): merges Google Brain + DeepMind → Google DeepMind. Also: on-device Stable Diffusion in ~12s via GPU-optimized diffusion inference.
  • OpenAI: ChatGPT Business (enterprise) in coming months — more data control; default: end-user data not used to train models (API policy).
  • AWS (Apr 13–14): genAI “full stack” + Amazon Bedrock API to AI21 / Anthropic / Stability / Amazon FMs.
  • Adobe: Premiere Pro text-based video edit (May target) — edit video like text.
  • Apple: engineers propose ChatGPT-class tech into Siri (approval uncertain given 12yr caution).
  • Meta: Zuckerberg — genAI “will touch every product”; more open-source path vs Google/MSFT/AWS.
  • PwC US: $1B over 3yrs into genAI with MSFT/OpenAI for tax/audit/consult automation.
  • Stability AI: open-sources StableLM (Alpha 3B/7B; roadmap 15–65B); CC 4.0 like Stable Diffusion.

01.2 China headlines

  • Alibaba Cloud: MaaS / PaaS / IaaS; Tongyi + enterprise LMs + ModelScope + APIs; all Ali products to plug LMs (Tmall, DingTalk, Amap, Taobao, Youku, Freshippo…). OPPO Andes joint LM infra on Tongyi; Geely/IM/Chery NEV/Haomo/Swire Coke/Bosideng/Zhangyue exploring.
  • Baidu: ERNIE Bot dedicated Android beta app (Apr 26) — voice I/O; assistants “Laowan” / “Xiaohua”; history; not synced with web.
  • Tencent Cloud: new HCC for LM training, ~3× prior performance; Robotics X TRX-Hand / TRX-Arm dexterous manip.
  • SenseTime: “Ririxin” LM suite (NLG, text-to-image, perception labeling, model R&D) + Chinese LM app platform “Shangliang”.
  • Kunlun: “Tiangong 3.5” invite test from Apr 17 — claims near ChatGPT / GPT-3.5 class with Qidian.
  • Haomo: DriveGPT (“Xuehu·Hairuo”) — first claimed autonomous-driving gen LM using human takeover data.
  • iFlytek: pretrain LM details at May 6 product launch. Zhihu “Zhihaitu AI” Chinese LM in beta.

02 Markets

Secondary color (sample): Q1 AIGC concept index +59.24% (Tonghuashun) lifts media + compute names. ChineseAll: AI hosts >180k audio hours / ~1k titles in 2022; RESTART metaverse on The Wandering Earth. BlueFocus 2022 ¥36.683B rev (−8.47%), net −¥2.175B (impairments); Q1’23 rev ¥9.774B (+25.6%), net ¥149M (+355%) — All-in AI / BlueFocus AI / “human+AI assistant”. DeepGlint 2022 first profit (rev ¥354M +20.5%; net ¥32.6M +147.7%). Jiecheng / Opera browser AI notes in CN tables.

Primary global sample: Narrato $1M seed (AirTree); Typeface $65M A (Lightspeed/GV/M12/Menlo — note dated Feb in CN). CB Insights: ~250 ChatGPT-concept startups; 33% no external equity; 51% ≤A; unicorns OpenAI $20B, HF $2B, Lightricks $1.8B, Jasper $1.5B, Glean & Stability $1B.

Primary China sample: DeepMusic multi-ten-M USD A+ (GGV/Fengyuan); Beibei Tech ¥100M B2 (Jufeng/Qingdao Wealth); Zhixiang Future seed (Alpha Commune + USTC experts); DeepLang ~$100M valuation new round (THUNLP/BAAI); Baichuan $50M (Wang Xiaoguan + friends; co-founder Ru Liyun); ModelBest multi-ten-M angel (Zhihu lead; Zhipu follow); Yunfu near-¥10M A (Yushi); Westlake Xinchen strategic (Tom Cat lead, BlueRun) — second round in a month after Mar 27 Pre-A (Baidu Ventures et al.); Wanxiang AIGC virtual-idol A2 (Porsche) also in month tables.

Methodology: Xenniu; seed merged with angel; FX rules in CN. 5 CN images; attach on ingest.

— END —

Editorial English edition by Unique Research, published on UniqueCapital on May 9, 2023. This translation is based on an earlier Chinese article; the Chinese source is credited separately and is not treated as a line-by-line mirror.

View the Chinese source article ↗
← Back to English research