Gunjo · Business Intelligence for the AI Era
← Sticker Wall MODEL · DETAIL

LLM API Reselling and Token Retail Ecosystem

1) Metered billing based on token usage, charging developers and enterprises for model API calls; 2) Exploring direct sa

MODEL

Key Fields

FIELD STAMPS
IndustryAI / LLM
RegionChina
ScaleGiant
ChannelOnline

📌 Background

In 2026, the commercialization of large language models in China reached an inflection point. According to Zhipu's mid-2026 financial report, H1 revenue reached 954 million RMB (a 399.7% year-over-year increase), with open platform and API revenue hitting 825 million RMB (up 2,735.7% YoY), growing from 15.2% to 86.5% of total revenue. As of late August, the MaaS platform ARR reached 1.6 billion USD, and token call volume surged over 40 times compared to the beginning of the year. API gross margin turned positive, rising from -0.4% to 24.6%.

👤 Target Customers

Developers needing AI capabilities, SMB SaaS vendors, internet platforms, and enterprise clients purchasing token packages through retail channels like Tmall.

💰 Revenue Streams

1) Metered billing based on token usage, charging developers and enterprises for model API calls; 2) Exploring direct sales of token packages through online retail channels like Tmall to lower the barrier to entry for small and medium-sized clients; 3) Capacity expansion: Tiered overage fees for token usage exceeding the package limit, plus dedicated capacity channel expansion fees.

🧮 Cost Structure

GPU computing power costs for model training and inference, data center operations, API gateway and billing system development and maintenance, and ongoing R&D investment for model iteration.

🛡️ Moat

Technical barriers of Zhipu's proprietary GLM series models, computing cost dilution advantages driven by economies of scale in API revenue, and developer ecosystem and customer switching costs accumulated on the MaaS platform.

🔑 Keys to Success

  • Continuously lower model inference costs to maintain API gross margins
  • Expand novel retail channels like Tmall to reach long-tail customers
  • Enhance customer stickiness and call frequency through multi-modal and agent capabilities

⚠️ Risks

  • Rising inference computing costs squeezing profit margins
  • Competitors' low-price strategies triggering price wars and slowing revenue growth

🏢 Cases

  • Zhipu Open Platform API revenue increased by 2736% year-over-year in the first half
  • Zhipu launched token packages for sale on its Tmall flagship store
  • SenseTime advanced its token factory and task delivery model to achieve profitability in its semi-annual report

📊 SWOT Analysis

Strengths

  • API revenue has formed a stable cash flow with a clear commercialization path
  • Deep technical accumulation in proprietary models supported by domestic substitution demand

Weaknesses

  • API gross margin is highly vulnerable to fluctuations in computing power prices
  • High dependency on top-tier clients with a single-stream revenue structure

Opportunities

  • The token retail market for small-to-medium enterprises and individual developers remains under-exploited
  • Upgrades in multi-modal and agent capabilities can drive higher per-customer call volumes

Threats

  • Aggressive API pricing by competitors such as DeepSeek, carrying price war risks
  • Lower barriers to open-source model deployment leading some clients to build in-house inference