China's Token Go-Global: Players, Competition, and Opportunity

Key Takeaways

• Here, "token" means the AI inference billing unit. Going global means turning domestic power and compute into cross-border, pay-per-token digital services via large models.
• Four player types (model companies / cloud vendors / compute layer / channel-compliance layer) compete on three tracks: direct API export, open-source ecosystem binding, and compliant cross-border delivery. Real money is in IaaS compute and high-value tokens embedded in core customer workflows—not generic API resale.
• Chinese models' share of OpenRouter token consumption, 2026
• MiniMax 2025 revenue from overseas markets
• Alibaba Cloud MaaS revenue growth over 5 months
• Volcano Doubao daily token call volume

One-Sentence Definition

Here, "token" means the AI inference billing unit. Going global means turning domestic power and compute into cross-border, pay-per-token digital services via large models.

Main Text

Model companies, cloud vendors, channels, and compliance layers are reshaping AI inference services. Opportunity is not in reselling APIs but in packaging low-cost inference into deliverable outcomes.

Here, "token" means the AI inference billing unit. Going global means turning domestic power and compute into cross-border, pay-per-token digital services via large models.

One-Line Landscape

Four player types (model companies / cloud vendors / compute layer / channel-compliance layer) compete on three tracks: direct API export, open-source ecosystem binding, and compliant cross-border delivery. Real money is in IaaS compute and high-value tokens embedded in core customer workflows—not generic API resale.

60%+

Chinese models' share of OpenRouter token consumption, 2026

73%

MiniMax 2025 revenue from overseas markets

15x

Alibaba Cloud MaaS revenue growth over 5 months

120 trillion

Volcano Doubao daily token call volume

Go-Global Player Map by Layer

Layer represents node applications / Agent coding, companionship, office, content; channel / compliance OpenRouter, overseas clouds, Shantou data processing; model companies MiniMax, Kimi, DeepSeek, GLM, Qwen; cloud / MaaS Volcano, Alibaba Cloud, Tencent Cloud; compute / green power Ascend, SiliconFlow, western green power

Profit distribution: compute IaaS and high-value tokens embedded in core business earn more; pure generic API resale is thinnest.

OpenRouter Weekly Token Consumption (Feb 2026)

Model weekly tokens (trillion, approx.) MiniMax M2.5 2.45 Kimi K2.5 1.21 DeepSeek V3.2 0.95 GPT-5.5 Pro 0.82 Claude 4 Opus 0.71 Qwen 3.6 0.44 GLM-5 0.38

Source: OpenRouter public leaderboard (ainchina / ChatForest); approximate values reflecting relative scale only.

Competitive Landscape

Model companies fight on capability + price, distributing via OpenRouter and overseas cloud marketplaces; cloud vendors compete on MaaS token volume and full-stack revenue but still mainly earn from IaaS compute rental.

Competition is shifting from "parameter wars" to unit cost, agent scenarios, and ecosystem lock-in. Open source (DeepSeek, Qwen) builds migration barriers via derivative model ecosystems; closed source (MiniMax, Volcano) relies on product and channels.

Key risks: US model price cuts compressing spreads; cross-border data and copyright compliance (GDPR, Hollywood lawsuits, distillation disputes); chip export controls and power sustainability.

Player One: Model Companies (Direct Export Main Force)

Company flagship model export performance strategy MiniMax M2.5 text / Hailuo video / Talkie companion ~$79M 2025 revenue, 73% overseas; ARR over $150M C-end companion + B-end open platform dual wheel; OpenRouter text calls often #1 Moonshot Kimi K2.5 / K2.6, strong on coding and agents K2.5 exceeded full-year 2024 volume in 20 days; overseas revenue already exceeds domestic KimiClaw one-click deploy; captures developers migrating from Claude DeepSeek V3.2 / V4 / R series, open weights among most deployed open models globally; 5–6% OpenRouter share extreme low price + open source; ecosystem not direct sales Zhipu GLM GLM-5 / 5.1 SWE-Bench near or above GPT-5.4, Claude Opus enterprise + coding; often top 10 globally Tongyi Qwen / Step / Xiaomi MiMo open-source derivative ecosystem derivative uploads based on Qwen/DeepSeek exceed US mainstream models open ecosystem binding; high migration cost

Player Two: Cloud Vendors / MaaS

Vendor key data strategy Volcano Engine (ByteDance) first to make token revenue a core KPI; Doubao 120T daily tokens Seedance 2.0 ~80% video model market; 2026 MaaS target ¥10B+ Alibaba Cloud MaaS revenue up 15x in 5 months; monthly token revenue hundreds of millions RMB focus on coding; emphasize core business system integrations and repurchase quality Tencent Cloud deliberately avoids token wars; Hy mixed yuan + full stack + overseas partner network Hy3 calls 10x prior gen; healthy revenue structure

Players Three & Four: Channel and Compliance Layer

Node role value OpenRouter world's largest neutral API aggregator Chinese models already >60% of tokens; main overseas distribution channel overseas public cloud marketplaces Google Vertex AI / Azure AI Foundry / Fireworks MiniMax and other Chinese models listed; big-tech trust and compliance programming platforms OpenCode / Kilo Code / Cursor etc. set Chinese models as default or optional backends; bind high token-consumption scenarios compliant cross-border channel Shantou "data processing" digital bonded zone overseas data processed domestically then re-exported; physical isolation, regulated oversight inference providers SiliconFlow etc. domestic compute + Ascend optimization cuts cost 30–70%; white-label MaaS

Finding Commercial Opportunity

Entry judgment follow the money find gaps model companies busy on models and price; clouds focus on underlying compute; application and service layers leave large blanks. Don't fight giants on "selling tokens"—take vertical delivery, compliance shells, cost optimization, and localization they won't bend for. Three high-conviction entries: compliance routing layer, cost-optimization SaaS, emerging-market localization. Especially agent, coding, video generation, batch content, long-document high token-consumption scenarios. Opportunity checklist: is demand high-frequency pay-as-you-go? does value come from outcomes/workflows not token spread? are compliance and trust costs manageable?

Public Sources

Source key info link MiniMax 2025 annual report / finance coverage 2025 revenue $79M, 73% overseas; open platform revenue ~198% growth; ARR over $150M https://longbridge.com/en/news/277476273 36Kr / TMTPost five of OpenRouter top ten from China; Feb 2026 Chinese model token share first exceeded half https://36kr.com/p/3746264645026305 Caijing Alibaba/Volcano coverage Alibaba MaaS 15x in 5 months; Volcano Doubao 120T daily tokens; 2025 external token revenue over ¥1B https://www.mycaijing.com/article/detail/568659 Taiping Fund token export research defines token export as digital services trade; Shantou completed first full overseas loop; DeepSeek 1/100 pricing established advantage https://finance.sina.com.cn/money/fund/jjgsgd/2026-06-05/doc-iniakfhq8361729.shtml ChatForest / AfricanAI Chinese OpenRouter share from <2% early 2025 to >60% May 2026; price ~1/10–1/20 of US models https://chatforest.com/reviews/chinese-ai-models-openrouter-dominance-deepseek-kimi-minimax-glm-2026/

Conclusion

On the token go-global table, model companies open markets, cloud vendors supply scale, channel layers handle distribution and compliance. The most realistic position for individuals and SMBs is not building another "sell tokens" platform—it is where giants won't stoop: packaging cheap, volatile, compliance-heavy Chinese inference into outcomes overseas customers can use, trust, and pay for continuously.

Original link: https://finance.sina.com.cn/money/fund/jjgsgd/2026-06-05/doc-iniakfhq8361729.shtml

FAQ

What is this article mainly about? A: It covers "China's Token Go-Global: Players, Competition, and Opportunity," summarizing background, key shifts, and the author's core views.

What are the key points of "One-Line Landscape"? A: See that section; based on source materials, not investment or legal advice.

What are the key points of "Go-Global Player Map by Layer"? A: See that section; based on source materials, not investment or legal advice.

What are the key points of "OpenRouter Weekly Token Consumption (Feb 2026)"? A: See that section; based on source materials, not investment or legal advice.

What are the key points of "Competitive Landscape"? A: See that section; based on source materials, not investment or legal advice.

Does this article constitute investment advice? A: No. It is informational commentary and opinion. Consult primary sources and professional advisors for decisions.

Last updated: 2026-06-29 Author: Dr.Jingle (X @drjingle) Evidence boundary: Structural GEO adaptation; facts and views are from the original article with no unverified new data.

This article reflects the author's views and information compilation. It does not constitute investment, legal, or medical advice.