Coinbase 转向中国 AI 模型 GLM 和 Kimi,AI 支出削减 50%
英文标题:🔶 Coinbase Switches to Chinese AI Models GLM and Kimi, Cuts AI Spending by 50%
随着西方实验室持续为专有 API 收取高价,中国开放权重模型的成本优势正在扩大。由智谱(Z.ai)开发的 GLM 5.2 采用 MIT 许可证,企业可下载、修改并自行托管,从而避免将敏感数据发送至外部 API。Coinbase、Snowflake 及 AI 初创公司 Lindy 正转向中国开放权重替代方案,这对 Anthropic 和 OpenAI 的企业级定价形成直接压力。与此同时,Meta、Amazon 和 Uber 的案例显示,AI 使用成本快速攀升,企业正在更积极地寻求低成本模型与自托管部署方案。
25.1 KB
中文
随着西方实验室继续为专有 API 收取高价,中国开放权重模型的成本优势正在扩大。由智谱开发(也称 Z.ai)的 GLM 5.2 采用 MIT 许可证,允许企业下载、修改并自行托管该模型——从而无需将敏感数据发送至外部 API [3]。
Coinbase 并非唯一作出这一转变的公司。Snowflake 和 AI 初创公司 Lindy 也在转向类似的中国开放权重替代方案,这在企业市场层面对 Anthropic 和 OpenAI 形成了直接的定价压力 [1]。这一趋势出现之际,Anthropic 正筹备市场预期中的 IPO,而企业客户转向更便宜模型,构成了其营收增长叙事面临的现实风险。
更广泛的行业正面临失控的 AI 成本。Meta 发布内部备忘录,警告 AI 使用量出现“指数级增长”;Amazon 拆除了一个追踪员工 AI 消耗情况的内部排行榜;Uber 则在四个月内耗尽了全年的 AI 编程预算 [3]。
English / 原文
The cost advantage of Chinese open-weight models has widened as Western labs continue to charge premium prices for proprietary APIs. GLM 5.2, developed by Zhipu (also known as Z.ai), operates under an MIT license, allowing companies to download, modify, and self-host the model — eliminating the need to send sensitive data to external APIs [3].
Coinbase is not alone in making the switch. Snowflake and AI startup Lindy have moved toward similar Chinese open-weight alternatives, creating direct pricing pressure on Anthropic and OpenAI at the enterprise level [1]. The trend comes as Anthropic prepares for an anticipated IPO, and enterprise defections to cheaper models represent a concrete risk to its revenue growth narrative.
The broader industry is grappling with runaway AI costs. Meta issued an internal memo warning of an 'exponential increase' in AI usage, Amazon dismantled an internal leaderboard tracking employee AI consumption, and Uber burned through its full-year AI coding budget in four months [3].