Zhipu AI Launches Token Sales on Tmall as AI Companies Begin Limiting Usage Despite Falling Costs
- Zhipu AI has opened a flagship store on Tmall selling its GLM Coding Plan with three tiers (Lite at 118 yuan/month, Pro at 538 yuan/month, Max at 1078 yuan/month), each with different token quotas.
- Despite token prices falling, AI companies including Kimi, Alibaba Cloud, and Tencent Cloud have begun limiting user access and adjusting pricing upward, with Tencent Cloud raising HY2.0 input prices by over 460%.
- Zhipu AI's unit token inference cost has dropped 80% compared to the beginning of the year, yet token consumption rates are growing faster than cost reductions.
- Zhipu AI's business model has shifted from local deployment (84.8% of revenue) to cloud-based MaaS API calls (86.5% of revenue) within one year, with token call volume increasing over 40 times.
- Despite falling inference costs, Zhipu AI's average API selling price increased by approximately 101% and MaaS gross margin improved to 24.6%, indicating a shift from price competition to value-based pricing.
Zhipu AI has launched a flagship store on Tmall offering its GLM Coding Plan at three subscription tiers, making AI token quotas purchasable like mobile data packages. This move reflects a broader industry shift where AI companies are increasingly concerned with limiting token consumption rather than competing on price alone.
Despite significant cost reductions—Zhipu AI's unit token inference cost has fallen 80% since the beginning of the year—companies including Kimi, Alibaba Cloud, and Tencent Cloud have begun restricting user access and raising prices. Tencent Cloud increased HY2.0 input prices by over 460%, while Kimi suspended new user subscriptions due to computing power constraints. The core issue is that token consumption growth is outpacing cost reductions, as tokens represent metered access to GPU computing resources, memory, networking, and data center operations.
Zhipu AI's financial performance illustrates this transformation. The company's business model has shifted dramatically from local deployment (which represented 84.8% of revenue a year ago) to cloud-based MaaS API services (now 86.5% of revenue). Token call volume has increased over 40 times, yet the average API selling price has risen approximately 101%, while MaaS gross margins improved to 24.6%. This indicates AI companies are moving beyond simple price competition toward value-based pricing, as models become capable of handling complex tasks like code generation and tool invocation, shifting customer perception from buying a model to purchasing computational work completion.
Market Impact: This pricing strategy shift is neutral to slightly bullish for AI service providers' profitability, as rising average selling prices and improved margins offset cost pressures, though it may create headwinds for price-sensitive consumer segments.
智谱AI在天猫售卖Token：成本下降反而限制使用，AI公司为何开始控制供应？
智谱AI在天猫开设旗舰店，推出GLM编码计划三个订阅等级，使AI token额度像移动数据套餐一样可购买。这一举措反映了更广泛的行业转变，AI公司越来越关注限制token消耗，而不仅仅是价格竞争。
尽管成本大幅下降——智谱AI的单位token推理成本自年初下降80%——包括Kimi、阿里云和腾讯云在内的公司已开始限制用户访问并提价。腾讯云将HY2.0输入价格上调460%以上，Kimi因算力压力暂停新用户订阅。核心问题是token消耗增长速度超过成本下降，因为token代表对GPU计算资源、内存、网络和数据中心运营的计量访问。
智谱AI的财务表现反映了这一转变。该公司的业务模式从本地部署（一年前占收入84.8%）大幅转向云端MaaS API服务（现占86.5%）。Token调用量增长超过40倍，但平均API售价上升约101%，MaaS业务毛利率提升至24.6%。这表明AI公司正从简单的价格竞争转向价值定价，因为模型变得能够处理代码生成和工具调用等复杂任务，改变了客户认知——从购买模型转向购买计算工作完成。
市场影响：这一定价策略转变对AI服务提供商的盈利能力呈中性至略微利好，因为售价上升和毛利率改善抵消了成本压力，但可能对价格敏感的消费者部分造成压力。