CHINA AI · Command Center

Brief

the five numbers

499M

AI-native app users, June 2026

+85% YoY, deduplicated (QuestMobile)

~$74B

DeepSeek round target valuation

reported; ~$15B raised in round two

~$50B

Moonshot IPO target

confidential HK filing, 3 Sep

China AI capex, 2026e

+103% YoY, derived from RMB 932B

Top model companies

flagships, Jul–Oct 2026

Mostly open-weight MoE models with ~1M-token context and API prices well below US frontier models.

Private · STAR IPO prep

DeepSeek

Hangzhou; spun out of High-Flyer. MIT-licensed weights. R2 reasoning model still unreleased.

LatestV4.1-Flash (10 Sep), 552B MoE

AlsoV4-Pro 1.6T / 49B active (Apr)

Revenue~$1B annualized reported

Alibaba

Qwen (Tongyi)

Strongest distribution: Alibaba Cloud, Taobao, Alipay, Ele.me, Fliggy. Top China-model token share on OpenRouter in Q2.

LatestQwen3.8 Max, 2.4T / 95B active (3 Aug)

OpenQwen3.8-27B, Apache 2.0

OpenRouter share13.9% (Q2)

Private · HK IPO filed

Moonshot AI (Kimi)

Long-context, agentic coding, multi-agent "swarms". Alibaba is the largest outside holder (~36%).

LatestKimi K3, 2.8T / 104B active (16 Jul)

ARR>$300M reported

Valuation~$31.5–35B; IPO target ~$50B

HKEX · listed 8 Jan 2026

Zhipu / Z.ai

Tsinghua spin-out, first "Six Tiger" to list. Enterprise and government focus; API now the main growth driver.

LatestGLM-5.3, 744B / 40B active (Aug)

Market cappassed on 22 Jun

ChipsInference on Ascend + domestic

HKEX · listed 9 Jan 2026

MiniMax

Consumer-first: Talkie companion app, Hailuo video. Reported 25.4% gross margin on FY2025.

LatestM3 (Jun), ~428B / 23B active

VideoMiniMax-H3

BackersAlibaba, miHoYo, Tencent-linked

Big tech & others

ByteDance · Tencent · Baidu · Xiaomi · StepFun

ByteDance Doubao/Seed, 2026 capex plan . Tencent Hunyuan, Yuanbao in WeChat. Baidu Ernie. Xiaomi MiMo. StepFun raised ~$2.5B in May ahead of a HK listing.

Investors & capital

public markets opened in January

Company	Status	2026 capital events (reported)	Named investors
DeepSeek	Private → STAR IPO (filing possible late 2026, debut 2027)	June: ~$7.4B at ~$52B post-money via an LP controlled by founder Liang Wenfeng (5-yr lock-up, no voting). Round two: ~$15B reported, targeting ~$74B	CATL, Tencent, Meituan, Geely, GigaDevice; CITIC Securities, CICC among underwriters
Moonshot AI	Confidential HK IPO filing, 3 Sep; Q1 2027 targeted	Jan: ~$500M at ~$4.3B. Spring: ~$2B+ led by Meituan (~$20B). July: $3.5B at ~$31.5B. Pre-IPO target ~$50B; IPO seeks ~$3B+	Alibaba (~36%), Tencent, IDG, HSG (ex-Sequoia China), Meituan, China Mobile, CPE, national AI fund, SMIC Capital, Boyu, Hefei state funds
Zhipu / Z.ai	Listed HKEX	Jan IPO ~$558M at ~$7.1B; follow-on placements in July and September ( in September alone)	Public investors
MiniMax	Listed HKEX; STAR second listing filed	Jan IPO ; shares roughly doubled on debut	Alibaba, miHoYo, Tencent-backed entity
StepFun	Pre-IPO (HK)	~$2.5B raised in May	Not detailed in sources reviewed

◈ The pattern. Strategic money dominates. Alibaba, Tencent, Meituan, CATL, Geely and China Mobile all hold stakes, tying labs to cloud, delivery, battery, auto and telecom distribution as much as to venture capital.

Consumers

distribution is the moat

ChatGPT is not available in the mainland. Doubao rides Douyin, Yuanbao rides WeChat's 1.3B users, Qwen rides Taobao and Alipay, China Mobile's Lingxi passed 130M monthly users.

Monthly active users, AI apps in China

Sep 2026 (Doubao, Qwen, DeepSeek) and Jun 2026 (Yuanbao) · QuestMobile · millions

Yuanbao is June data from a weaker source.

181M

Doubao daily actives (30 Sep)

vs 29M for Qwen, ~6x

132 min

Doubao time per user per month

72 opens; DeepSeek users 137 min

−9M

Qwen MAU, Jun→Sep

DeepSeek −6M; Doubao +37M

Paid tiers

Doubao subscriptions

free tier stays; three paid packages prepared

Digital economy at a glance

official data

Digital economy value added, 2025

~35% of GDP (National Data Administration)

Core digital industries, 2026e

>10.5% of GDP

89%

Mobile payment adoption

Global Payments Report 2026

1.3B

WeChat monthly users

messaging + payments + commerce

Published GDP shares range from ~35% to ~42% depending on scope. The 15th Five-Year Plan (2026–30) makes digital empowerment of the real economy a core goal.

Platform players

where AI gets embedded

Player	Digital role	AI angle
Alibaba	Taobao/Tmall, Alipay (via Ant), Ele.me, Alibaba Cloud	3-year AI and cloud capex plan; external cloud revenue +35% latest quarter; T-Head chips: 470k shipped by Feb 2026, 60%+ to external customers; Qwen app wired into shopping and payments
Tencent	WeChat, WeChat Pay, gaming, Tencent Cloud	Yuanbao inside WeChat; building agent capability natively in the super-app; chip access constrained its 2025 capex
ByteDance	Douyin e-commerce and livestream, TikTok	Doubao books tickets through Douyin commerce; 2026 capex plan , the largest marginal driver of China AI capex
Baidu	Search, Apollo Go, AI Cloud	Q2 capex (~+200% YoY); AI cloud revenue +21%; own chips ~5% of domestic accelerators projected
PDD · JD · Meituan	Low-price marketplace; first-party retail and logistics; local delivery and travel	JD logistics revenue +29% YoY to ; Meituan is lead investor in Moonshot and a DeepSeek backer
Ant · WeBank	Alipay and MYbank; WeChat Pay and WeBank	Ant's Afu assistant is in the top-10 AI apps; payments rails give agents a native checkout

◈ Agentic commerce. Qwen, JD's AI Shopping and Doubao now do conversational shopping, comparison and checkout inside their own ecosystems. Whoever owns payments plus an agent owns the transaction.

Compute

capex and chips

China AI infrastructure capex

USD billions · Rhodium Group survey of 13 listed firms · 2026–27 projected

Roughly 15–20% of US AI investment. 2025 is implied from +103% growth in 2026.

Domestic AI accelerator share, 2026

Percent · sources use different bases (revenue vs units), so they disagree

Huawei: Morgan Stanley projects 62% and ~$12B AI chip revenue. IDC units at June 2026: Huawei 28%, Nvidia 34%. Bernstein puts Nvidia near 8%.

◈ The catch. Chinese hyperscalers build from a far smaller revenue base than US peers, and AI ROIC is estimated at 13–20% versus 25–50% in North America. Capex is outrunning monetization.

The state-owned big three

pipe → compute → tokens

Subscriber growth continues but revenue is flat to falling, so growth is shifting to cloud, data centres and AI tokens.

Largest operator

China Mobile

World's largest mobile operator. Focus: communications, computing, AI.

Mobile subscribers (Sep)1.014B

5G subscribers702M (69.2%)

IoT connections1.497B

9M revenue (−1.8%)

9M net profit (−7.9%)

H1 AI services revenue

Lingxi AI assistant130M+ monthly users

Fixed + mobile + cloud

China Telecom

Highest 5G penetration of the three; most vertically integrated AI offer (own TeleChat LLM).

Mobile subscribers (Jun)442.4M

5G subscribers322.5M (~73%)

Wireline broadband subscribers201.3M

H1 revenue

Interim profit−15%; dividend cut 11%

Telecom Cloud (+7.8%)

Blended ARPU (integrated)/month

Compute & IoT heavy

China Unicom

Largest IoT base; positions as an aggregator of AI tokens and an operator of "high-value tokens".

Total connections (Sep)>1.3B (incl. 789.7M IoT)

5G subscribers255.6M

9M revenue (+0.6%)

9M net profit (−33.6%)

Computing revenue+12.2% (IDC +10.4%)

2026 capex plan, 35%+ to compute

Cloud-AI paying users123.5M

Scale

subscribers

5G network subscribers

Millions · China Mobile, China Unicom Sep 2026; China Telecom Jun 2026

Mobile subscribers

Millions · China Mobile Sep 2026; China Telecom Jun 2026

China Unicom excluded: its headline connection count mixes in hundreds of millions of IoT terminals.

What is changing

the token economy

AI token subscriptions

Since May all three sell tokenised AI usage. China Telecom: three national tiers (consumers, developers and small businesses, ecosystem partners) bundled with TeleChat and DeepSeek. China Mobile: marketplace-style trials in several provinces. China Unicom: aggregator packages, starting in Shanghai.

Core under pressure

Weak consumer spending and a tax hike weigh on all three. China Mobile's Q3 profit fell 12.5%; the pivot is from "communications operator" to "technology services enterprise".

Network base

4.83M 5G base stations and 1.19B 5G subscribers nationally by Nov 2025 (65.3% of mobile). All three hold satellite mobile licences; MIIT is advancing 6G projects.

◈ The CEO line. The pipe is a mature utility. The growth case is selling compute and metered AI to subscribers who already pay every month. Try it in the Class Model tab.

Class model

stress the numbers

Three live models built from the data on the other tabs. Inputs marked illustrative are assumptions, not reported figures.

A · Consumer subscription revenue

illustrative

What if a top AI app converts its users to paid, as Doubao is preparing to?

App Paid conversion Price per month (USD)

paying users

annual revenue

vs DeepSeek ~$1B ARR

B · Lab valuation multiple

illustrative growth

Reported valuation targets against reported ARR. How fast must revenue grow to look reasonable?

Next-12-month ARR growth Multiple investors would accept (forward ARR)

C · Telco AI-token revenue

illustrative adoption

How much new revenue can a big-three operator add by selling metered AI to its existing subscribers?

Operator Token plan adoption Added spend per adopter per month (USD)

base subscribers

added revenue / yr

annualized revenue

uplift on revenue

◈ The catch. Models A and C turn on conversion and price assumptions that no operator or lab has disclosed. Treat outputs as sensitivity, not forecast.