---
id: "2026-10-06-epoch-chinese-ai-companies-revenue"
url: "https://postcutoff.com/e/2026-10-06-epoch-chinese-ai-companies-revenue/"
as_of: "2026-10-13T07:38:00+02:00"
date: "2026-10-06"
date_precision: day
category: business
importance: 3
confidence: medium
status: [Partly confirmed]
sources: 2
editor: Adam Bicz
human_review: null
version: null
---

As of: 2026-10-13 07:38 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code). Human editor: Adam Bicz.
Canonical page: https://postcutoff.com/e/2026-10-06-epoch-chinese-ai-companies-revenue/

# Epoch AI: China's six leading AI firms earn about 10% of OpenAI and Anthropic's combined AI revenue; China is 2.7× more exposed to a chip supply shock

Two Epoch AI reports of Oct 6, 2026. "How do Chinese AI companies make money?" (Cheryl Wu, Anson Ho) puts annualized AI revenue at $4.0B for ByteDance and $2.4B for Alibaba (model-as-a-service only), $1.8B for Z.ai, $1.0B each for Moonshot and DeepSeek and $0.8B for MiniMax, about 10% of Anthropic ($65B, July) plus OpenAI ($40B, August) as charted by Epoch. A second report (Daniel Carey) finds China about 2.7× as exposed as the US to semiconductor supply disruptions.

## Key facts
- Annualized AI revenue in Epoch's chart (US$B): Anthropic 65 (July), OpenAI 40 (August), ByteDance 4.0 (July, MaaS only), Alibaba 2.4 (August, MaaS only), Z.ai 1.8 (September), Moonshot 1.0 (August), DeepSeek 1.0 (September), MiniMax 0.8 (August); months and definitions differ, so the figures are not strictly comparable
- DeepSeek: mainly API sales (est. $1B), estimated API gross margin 70–80%; about 8% of OpenRouter spending on its models goes to DeepSeek itself
- Z.ai (Zhipu): estimated 2026 API revenue $805M; on-premises share of revenue fell from 73.7% (2025) to 13.5% (H1 2026); API gross margin 18.9% (2025) → 24.6% (H1 2026); management's September ARR of $1.8B differs from interim-report figures
- MiniMax: AI-native app gross margin 4.7% (first nine months of 2025) vs 69.4% for enterprise services; Moonshot: about 70% of revenue from model access as of June 2026
- ByteDance: Volcano Engine handled 49.5% of China's public-cloud large-model token calls in 2025; 2025 MaaS revenue about $221M vs an estimated ~$4B ARR in 2026. Alibaba: AI cloud and compute revenue $7.1B in Q2 2026 (+45% YoY)
- Epoch: 'consumer-facing AI apps are not highly profitable'; Liang Wenfeng is quoted saying DeepSeek 'doesn't have that much compute'
- Chip-shock report (Daniel Carey): in 2022 every $1,000 of Chinese final demand generated $15.2 of semiconductor revenue vs $5.7 for the US (2.7×); China's domestic producers met 23% of its demand in 2022 (12% in 2016)
- Modelled Taiwan collapse plus China–West decoupling (2–3 year horizon): China's advanced-processor prices rise 17-fold and real expenditure falls 3%; US prices rise about 20% and expenditure falls 0.6%; price effects are much less robust than the expenditure effects (range: +6% to 100-fold)
- Caveats: revenue from disclosures, press and run-rate statements; trade tables end in 2022 and miss the US AI-server boom (a 2024 check gives 2.2×)

## What happened

Epoch AI collected company disclosures and press figures for the six Chinese firms that lead in models. Together they run at roughly
a tenth of the two US leaders' revenue. Start-ups (DeepSeek, Moonshot, Z.ai) live mostly on API sales with thin to good margins;
ByteDance and Alibaba earn mostly indirectly, through cloud, ads and commerce. Consumer apps earn little.

The companion trade-model report measures who depends more on imported chips. China does, by a factor of about 2.7 in 2022, and in
every scenario tested its losses from a disruption exceed those of the US.

## Why it matters

Chinese open-weight models are close to the frontier on many benchmarks while their makers earn an order of magnitude less, which
limits what they can spend on compute. The trade result gives a number to China's chip dependence at a time of export controls and
Taiwan risk. Both reports rest on estimates, as Epoch states.

## Your AI and this story

- GPT-6 Astra (training cutoff April 2026): 159 days after its cutoff
- Claude Opus 5.5 (training cutoff June 2026): 98 days after its cutoff
- Gemini 3.8 Flash (training cutoff March 2026): 189 days after its cutoff
- Grok 4.7 (training cutoff May 2026): 128 days after its cutoff

## Sources

1. [Epoch AI: How do Chinese AI companies make money? (Oct 6, 2026)](https://epoch.ai/publications/how-do-chinese-ai-companies-make-money) (epoch.ai, official)
2. [Epoch AI: Who is most exposed to a chip supply shock? (Oct 6, 2026)](https://epoch.ai/publications/china-us-semiconductor-supply-chain-exposure) (epoch.ai, official)

## Changes

- 2026-10-13 (filed): Created (both reports read directly)

## Related

- 2026-10-06: [DeepSeek nears $12B+ funding round led by Tencent and CATL ahead of an early-2027 IPO](https://postcutoff.com/e/2026-10-06-deepseek-12b-funding-round/index.md)
- 2026-10-02: [Epoch AI: chips shipped through 2027 could run 33–171 million frontier-model agents at once](https://postcutoff.com/e/2026-10-02-epoch-ai-agent-population-estimate/index.md)
- People: [Liang Wenfeng](https://postcutoff.com/person/liang-wenfeng/)