AI subscriptions are a ticking time bomb for enterprise
AI subscriptions are a ticking time bomb for enterprise
AI 订阅对企业而言是一枚定时炸弹
Every AI lab is losing money serving your company right now. They know it. And they are doing it on purpose. 目前，每一家人工智能实验室在为贵公司提供服务时都在亏钱。他们心知肚明，而且这是他们刻意为之。
OpenAI, Anthropic, Google, and the rest are running an industry-wide loss-leader program at a scale that has no precedent. They are selling enterprises filet mignon at gas station hot dog prices and calling it a business model. The gap between what your company pays for AI subscriptions and what it actually costs to serve those seats is not a rounding error. It is a gulf. And every organization that has built workflows, products, or entire business units on top of these subsidized prices is standing right on the edge of it. OpenAI、Anthropic、Google 以及其他公司正在进行一场史无前例的行业性“亏本赚吆喝”计划。他们以加油站热狗的价格向企业兜售菲力牛排，并将其美其名曰“商业模式”。贵公司支付的 AI 订阅费与实际服务成本之间的差距，绝非四舍五入的零头，而是一道巨大的鸿沟。任何将工作流、产品甚至整个业务部门建立在这些补贴价格之上的组织，都正站在悬崖边缘。
This should be front of mind for every CTO, CFO, and head of operations reading this. Because when the pricing corrects, and it will, the companies that treated AI as a permanently cheap utility are going to wake up to bills that make their current SaaS spend look quaint. 每一位阅读本文的首席技术官（CTO）、首席财务官（CFO）和运营主管都应高度警惕。因为当价格回归理性（这必然会发生）时，那些将 AI 视为廉价公用事业的公司将会发现，未来的账单将让现在的 SaaS 开支显得微不足道。
The Math Your Finance Team Has Not Done
你的财务团队尚未计算的账目
Pull out the napkin. This matters. Claude Pro costs $20 a month. For that, you get access to Sonnet 4.6, Opus 4.6, web search, code execution, file creation, and roughly 5x the usage of the free tier. On the API side, Sonnet 4.6 costs $3 per million input tokens and $15 per million output tokens. Opus 4.6 runs $5 input and $25 output per million tokens. 拿张餐巾纸算算吧，这很重要。Claude Pro 每月收费 20 美元。为此，你可以使用 Sonnet 4.6、Opus 4.6、网络搜索、代码执行、文件创建功能，且使用额度约为免费版的 5 倍。在 API 端，Sonnet 4.6 的价格为每百万输入 Token 3 美元，每百万输出 Token 15 美元；Opus 4.6 则分别为 5 美元和 25 美元。
A knowledge worker running a few hours of Claude daily, uploading documents, drafting reports, analyzing data, can easily burn through several million tokens per week. At API rates, that same workload runs somewhere between $200 and $400 a month per seat. Some power users push well beyond that. But on a Pro subscription, the company is paying $20 per head. 一名知识工作者如果每天使用几小时 Claude，上传文档、起草报告、分析数据，每周很容易消耗数百万个 Token。按 API 费率计算，同样的工作量每月每个席位的成本在 200 到 400 美元之间。一些重度用户甚至远超此数。但在 Pro 订阅模式下，公司每人每月只需支付 20 美元。
Anthropic is not the only one eating this cost. Microsoft was reportedly losing over $20 per user per month on GitHub Copilot. For power users, the compute burn was hitting $80 a month on a $10 subscription. One widely cited analysis found that Anthropic users were consuming upwards of $8 in compute for every $1 of subscription revenue. OpenAI’s own VP of Product, Nick Turley, has described their subscription pricing as something they “stumbled into” and has floated the idea of phasing out unlimited plans entirely, comparing them to “unlimited electricity.” Anthropic 并非唯一一家承担此成本的公司。据报道，微软在 GitHub Copilot 上每位用户每月亏损超过 20 美元。对于重度用户，在 10 美元的订阅费下，计算成本消耗高达每月 80 美元。一项被广泛引用的分析显示，Anthropic 用户每贡献 1 美元的订阅收入，就会消耗超过 8 美元的计算资源。OpenAI 产品副总裁 Nick Turley 曾形容他们的订阅定价是“误打误撞”的结果，并提出了完全取消无限量计划的想法，将其比作“无限量电力”。
ChatGPT Plus has been $20 a month for three years. In that time, the models got dramatically more capable. The features multiplied. Image generation, code interpretation, voice mode, agentic reasoning, web search. And the price never moved. For enterprise buyers who locked in team or business rates during this window, the question is not whether they got a good deal. The question is how long that deal survives. ChatGPT Plus 的价格在三年里一直保持每月 20 美元。在此期间，模型能力大幅提升，功能成倍增加：图像生成、代码解释、语音模式、智能体推理、网络搜索……而价格从未变动。对于在此期间锁定团队或企业费率的买家来说，问题不在于他们是否占了便宜，而在于这种优惠能持续多久。
This Is Not One Company’s Problem
这并非一家公司的问题
Every major provider is playing the same game with the same math. Google offers Gemini Advanced at $20 a month bundled into Google One AI Premium while simultaneously charging developers real money for API access to the same models. Meta gives away Llama for free, subsidizing the compute cost of hundreds of millions of AI queries across its platforms entirely through ad revenue. xAI’s Grok undercuts everyone on API pricing at $0.20 per million input tokens, a number that only makes sense if you assume the company is willing to hemorrhage money to buy market share. 每一家主要供应商都在用同样的逻辑玩着同样的游戏。Google 以每月 20 美元的价格提供捆绑在 Google One AI Premium 中的 Gemini Advanced，同时却向开发者收取高昂的 API 访问费用。Meta 免费提供 Llama，完全通过广告收入来补贴其平台上数以亿计的 AI 查询计算成本。xAI 的 Grok 以每百万输入 Token 0.20 美元的 API 价格击败了所有人，除非你认为该公司愿意通过巨额亏损来换取市场份额，否则这个价格根本无法盈利。
The pattern is identical across the board. Price for adoption, not for economics. Lock organizations in. Make AI a load-bearing part of every team’s daily workflow. Worry about the bill later. 这种模式在整个行业如出一辙：定价是为了普及，而非为了经济效益。锁定组织，让 AI 成为每个团队日常工作流的支柱，至于账单，以后再说。
For enterprises, “later” is arriving. OpenAI is losing money on consumer subscribers and is reportedly considering a strategic pivot away from its consumer bets toward a tighter focus on enterprise, where the unit economics are slightly less ruinous. The Wall Street Journal reported that the company missed key revenue and user targets in its sprint toward an IPO. The subsidy era is not winding down gracefully. It is showing cracks everywhere. 对于企业而言，“以后”已经到来。OpenAI 在消费者订阅业务上正在亏钱，据报道，该公司正考虑战略转型，从消费者业务转向更专注于企业市场，因为那里的单位经济效益稍微没那么惨烈。《华尔街日报》报道称，该公司在冲刺 IPO 的过程中错失了关键的收入和用户目标。补贴时代并没有优雅地落幕，它正在全面崩塌。
Agents Broke the Economics
智能体（Agents）打破了经济模型
What made the subsidy math merely bad just became catastrophic. The reason is agentic AI. 原本只是“糟糕”的补贴账目，现在变得“灾难性”了。原因就是智能体 AI（Agentic AI）。
When AI was a chatbot, you ask a question, it answers, token consumption was relatively predictable. A conversation might run a few thousand tokens. Heavy use might push into the tens of thousands. That was manageable at subsidized rates. 当 AI 还是聊天机器人时，你问它答，Token 消耗相对可预测。一次对话可能消耗几千个 Token，重度使用可能达到几万个。在补贴费率下，这尚在可控范围内。
The agentic shift changes the equation completely. Claude Code sessions run autonomously for extended periods, burning through tokens at rates that dwarf conversational usage. Users have reported exhausting 5-hour rate limit windows in under 90 minutes. GitHub just announced that Copilot is moving to usage-based billing on June 1, 2026 specifically because the flat-fee model collapsed under agentic workloads. GitHub’s own announcement acknowledged that Copilot has evolved substantially and that agentic usage “is becoming the default,” which causes higher compute and inference demands. Sam Altman has said publicly that OpenAI now needs to become “an AI inference company,” an acknowledgment that agentic usage requires a fundamentally different economic model. 智能体的转变彻底改变了这一等式。Claude Code 会话可以长时间自主运行，其消耗 Token 的速度远超对话式使用。有用户反映，原本 5 小时的限额在不到 90 分钟内就被耗尽。GitHub 刚刚宣布，自 2026 年 6 月 1 日起，Copilot 将转向基于用量的计费模式，原因正是固定费用模式在智能体工作负载下已经崩溃。GitHub 的公告承认，Copilot 已大幅进化，智能体使用“正成为默认方式”，这导致了更高的计算和推理需求。Sam Altman 也公开表示，OpenAI 现在需要成为一家“AI 推理公司”，这承认了智能体使用需要一种根本不同的经济模型。
For enterprise engineering teams, the implications are concrete. Agent Teams, multiple AI instances working in parallel on a single project, multiply the burn rate dramatically. A developer running three or four concurrent coding agents is not consuming 3x or 4x the tokens of a chat conversation. It is an order of magnitude more. And the subscription price on that seat has not changed. 对于企业工程团队来说，影响是具体的。智能体团队（Agent Teams）——即多个 AI 实例在同一项目上并行工作——会成倍增加消耗速度。一名开发者同时运行三四个编码智能体，其消耗的 Token 绝非对话式使用的 3 到 4 倍，而是高出一个数量级。然而，该席位的订阅价格却纹丝未动。
The Enterprise Exposure No One Is Measuring
没人衡量的企业风险敞口
This is where it gets ugly for organizations that have not done the work. Over the past two years, thousands of companies have woven AI subscriptions deep into their operations. Marketing teams draft copy through ChatGPT Plus. Engineering teams write and review code through Claude Pro. Research teams synthesize documents… 对于那些尚未做好准备的组织来说，情况正变得棘手。在过去两年里，成千上万家公司已将 AI 订阅深度融入其运营中。营销团队通过 ChatGPT Plus 起草文案，工程团队通过 Claude Pro 编写和审查代码，研究团队合成文档……