AI 产业增长依然强劲,但增速正在放缓

由 THE DECODER 提供提示词的 Midjourney 生成图
美国企业正以前所未有的强度使用 AI 模型,同时为此支付的费用却在下降。
最新一期的 Ramp AI Index 基于超过 85,000 家美国企业的交易数据,给出了具体数字。据 Ramp 首席经济学家 Ara Kharazian 称,Token——即 AI 提供商对模型使用量计费的计量单位——的价格在六周内下降了 21.2%。与此同时,Token 用量上升 1.5%,创下历史新高。
Token 支出,也就是企业实际为 AI API 支付的费用,较两周前的峰值下降了 6.3%。环比来看(截至 11 月 15 日当周对比截至 10 月 18 日当周),Token 总支出仍增长 12%,但同期 Token 用量跃升 27.2%。使用量与收入之间的差距持续扩大。
据 Ramp 数据,OpenAI 与 Anthropic 合计的每百万 Token 平均价格降至 0.45 美元,较上月下降 11.2%。OpenAI 更便宜,每百万 Token 为 0.38 美元,Anthropic 则为 0.50 美元。
OpenAI、Anthropic 与中国厂商之间的价格战正在显现效果:Token 变得越来越便宜。11 月中旬,每百万 Token 的平均价格为 0.45 美元,其中 OpenAI 为 0.38 美元,Anthropic 为 0.50 美元。|图片来源:Ramp AI Index
API 路由平台 OpenRouter 也呈现出类似趋势。根据 Peter Walker 在 X 上发布的一项分析,自 2026 年初以来,OpenAI、Anthropic 和 Google 的封闭权重模型每周 Token 用量增长了 1,215%。同期每周支出增长 455%,增速要慢得多。根据 OpenRouter 的数据(统计截至 11 月 9 日当周),支出峰值出现在 10 月 5 日当周。自那之后,支出一直在下滑。
在 OpenRouter 上,自 2026 年初以来,每周 Token 量增长了 1,215%,而每周支出增长了 455%。支出自 10 月初的峰值以来已有所下降。该图表仅涵盖来自 OpenAI、Google 和 Anthropic 的封闭权重模型。| 图片来源:OpenRouter
Ramp 的 Token 指标依赖于从 AI 提供商导入的使用量与计费数据,而 OpenRouter 则通过其 API 路由平台衡量流量。两个来源都显示支出在 10 月达到峰值,随后支出下降而用量上升,这一事实强化了该趋势,不过这两项服务都不能代表整体市场。尤其是 OpenRouter,它只能提供一个观察极为特定用户群体的狭窄窗口。
Anthropic 在 IPO 当日主导 Token 市场
根据 Ramp AI Index,在截至 2026 年 11 月 15 日的一周里,Anthropic 占全部 Token 支出的 58.7%,较上月上升 1.7 个百分点。其 Token 用量占比更高,达 59.4%。OpenAI 降至支出的 37.2% 和用量的 37.7%。Google 为 1.4%,xAI(SpaceXAI)为 2.4%。
Kharazian 在 X 上写道,Anthropic 在 IPO 当日占据这一份额。不过,在 10 月,Anthropic 的 Token 支出环比仅增长 0.5%,而 OpenAI 增长了 2.1%。
根据 Ramp 的数据,2026 年 10 月,所有被追踪的美国公司中有 56.8% 为 AI 产品或服务付费,较 9 月上升 1 个百分点,比一年前高出 13.3 个百分点。Anthropic 以 44.8% 的采用率领先,OpenAI 为 40.6%。
Stripe 的数据也显示 AI 增长强劲,并出现降温迹象
Ramp AI Index 追踪的是需求侧,即企业在 AI 上的支出。支付处理商 Stripe 发布了自己的 AI Index,衡量的是供给侧:约 2 万家使用 Stripe 的 AI 公司的收款情况。
Stripe AI Index 显示,自 2022 年以来 AI 公司的收款持续上升。2025 年起增幅进一步加大,该指数在 2026 年突破 600 点。|图片:Stripe
根据 Stripe AI Index,在截至 11 月 8 日的 52 周内,AI 公司的营收增长了 103%。不过,典型(中位数)AI 公司仅增长 41%,这表明少数大型提供商拉高了整体数据。
在更短时间窗口内的放缓尤为明显。按 13 周年化计算,增长率约为 80%,而就在几个月前还是 140%。美国以 104% 的同比增长领先,欧盟则为 67%。Stripe 表示,AI 公司的增长速度远快于传统软件公司:103% 对比 SaaS Index 的 33%。
在截至 2026 年 11 月 8 日的 52 周内,受追踪的 AI 公司在美国的支付活动增长了约 104%。英国以 83% 紧随其后,世界其他地区为 76%,欧盟增长 67%。|图片:Stripe
两个指数描绘出相似的图景。AI 仍在快速增长,但势头正在降温。在 Ramp,这体现为价格下降和需求集中;在 Stripe,则体现为供给侧增长率的放缓。
新的增长可能来自能力更强或更可靠的模型,也可能仅仅来自将现有模型更好地整合进业务工作流。考虑到 AI 在短短几个月内颠覆了整个编程市场,这种潜力是存在的。不过,AI 行业若想收回其计划中的大规模产能建设投入,仍然依赖异常强劲的增长。
没有炒作的 AI 新闻——由人工筛选
可完整访问 THE DECODER 上的每一篇文章
参与评论和社区讨论
每周通过邮件发送的 AI 新闻回顾
每年 6 次:“AI Radar”——深入探讨最重要的 AI 议题
每日 AI 新闻,始终紧跟最新动态
我们完整的十年存档
由一支拥有 10 年以上 AI 经验的团队负责
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(AI industry growth remains strong, but the pace is slowing)
Midjourney prompted by THE DECODER
US companies are using AI models more intensively than ever and paying less for the privilege.
The latest update of the Ramp AI Index, which draws on transaction data from more than 85,000 US companies, lays out the numbers. Tokens, the units of computation in which AI providers bill model usage, got 21.2 percent cheaper in six weeks, according to Ramp lead economist Ara Kharazian. At the same time, token volume rose 1.5 percent to a record high.
Token spending, meaning what companies actually pay for AI APIs, fell 6.3 percent from its peak two weeks earlier. Month over month (the week ending November 15 vs. the week ending October 18), total token spending still rose 12 percent, but token volume jumped 27.2 percent over the same span. The gap between usage and revenue keeps widening.
The average price per million tokens for OpenAI and Anthropic combined fell to $0.45, according to Ramp, down 11.2 percent from the prior month. OpenAI is cheaper at $0.38 per million tokens compared with $0.50 for Anthropic.
The price war between OpenAI, Anthropic and Chinese providers is having an effect: tokens keep getting cheaper. In mid-November, the average price per million tokens sits at $0.45, with OpenAI at $0.38 and Anthropic at $0.50. | Image: Ramp AI Index
API routing platform OpenRouter shows a similar pattern. According to an analysis Peter Walker published on X, weekly token volume for closed-weight models from OpenAI, Anthropic and Google grew 1,215 percent since the start of 2026. Weekly spending rose 455 percent over the same period, far more slowly. The spending peak came in the week of October 5, according to the OpenRouter data, which runs through the week of November 9. Spending has been declining since then.
On OpenRouter, weekly token volume rose 1,215 percent since early 2026, while weekly spending grew 455 percent. Spending has fallen since its early-October peak. The chart covers only closed-weight models from OpenAI, Google and Anthropic. | Image: OpenRouter
Ramp's token metrics rely on usage and billing data imported from AI providers, while OpenRouter measures traffic through its API routing platform. The fact that both sources show a spending peak in October followed by declining outlays and rising volume reinforces the trend, though neither service is representative of the overall market. OpenRouter in particular offers only a narrow window into a very specific user base.
Anthropic dominates the token market on IPO day
In the week ending November 15, 2026, Anthropic accounted for 58.7 percent of all token spending, according to the Ramp AI Index, up 1.7 percentage points from the prior month. Its share of token volume was even higher at 59.4 percent. OpenAI fell to 37.2 percent of spending and 37.7 percent of volume. Google came in at 1.4 percent and xAI (SpaceXAI) at 2.4 percent.
Kharazian wrote on X that Anthropic holds this share on IPO day. In October, however, Anthropic's token spending grew only 0.5 percent month over month, while OpenAI's grew 2.1 percent.
In October 2026, 56.8 percent of all tracked US companies paid for AI products or services, according to Ramp, up one percentage point from September and 13.3 percentage points more than a year earlier. Anthropic leads with an adoption rate of 44.8 percent, while OpenAI sits at 40.6 percent.
Stripe data also shows strong AI growth and signs of cooling
The Ramp AI Index tracks the demand side, what companies spend on AI. Payment processor Stripe has published its own AI Index that measures the supply side: incoming payments at roughly 20,000 AI companies that use Stripe.
The Stripe AI Index shows rising incoming payments for AI companies since 2022. The increase steepens starting in 2025, and the index crosses 600 points in 2026. | Image: Stripe
According to the Stripe AI Index, revenue at AI companies grew 103 percent over the 52 weeks ending November 8. The typical (median) AI company grew only 41 percent, though, suggesting that a few large providers are skewing the overall picture.
The slowdown over shorter time frames is pronounced. Annualized over 13 weeks, growth stood at about 80 percent, down from 140 percent just a few months ago. The US leads with 104 percent year-over-year growth, while the EU sits at 67 percent. AI companies are growing far faster than traditional software firms, according to Stripe: 103 percent compared with 33 percent in the SaaS Index.
Payment activity among tracked AI companies rose about 104 percent in the US over the 52 weeks ending November 8, 2026. The UK followed at 83 percent, the rest of the world at 76 percent and the EU at 67 percent growth. | Image: Stripe
Both indexes paint a similar picture. AI is still growing fast, but momentum is cooling. At Ramp, that shows up through falling prices and concentrated demand. At Stripe, it shows up through slowing growth rates on the provider side.
New growth could emerge from more capable or more reliable models, or simply from better integration of current models into business workflows. The potential is there, considering how AI has upended the entire coding market in a matter of months. The AI industry still depends on extraordinarily strong growth, though, if it wants to recoup the massive capacity buildout it has planned.
AI News Without the Hype – Curated by Humans
Full access to every article on THE DECODER
Join the comments and community discussions
A weekly AI news recap via mail
6x/year: "AI Radar" — deep dives on the AI topics that matter most
Daily AI news, always up to date
Our full ten-year archive
Covered by a team with 10+ years in AI
原文为英文,正文由自动抓取并保留原文。
原文发布于 The Decoder(AI industry growth remains strong, but the pace is slowing),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。