企业使用的 AI 比以往任何时候都多，但它们在 AI 上的支出并没有增加。
Ramp 的一份新报告显示，尽管令牌（token）使用量上升，但 AI 支出在夏末趋于停滞，这表明价格竞争和更精明的客户可能正在限制 OpenAI 和 Anthropic 的收入。根据 Ramp 的数据，8 月和 9 月的令牌支出仅增长了 1.5%。相比之下，7 月较上月增长了 42%。
随着 Anthropic 准备其首次公开募股（IPO），数据表明投资者假设 AI 公司将继续以惊人的速度增长可能是错误的。
Ramp 的首席经济学家 Ara Kharazian 告诉 TechCrunch：“几乎所有 AI 支出都来自风险投资支持的高增长科技公司。”“只要 AI 支出继续上升，这没问题，但如果由于市场或其他因素导致支出开始放缓，它们将同时开始放缓，这可能会导致其他连锁反应。”
Kharazian 表示，他已经开始看到证据表明，前 1% 的企业——即最“沉迷于 AI”的企业——在 8 月和 9 月的支出略有停顿，略低于夏季峰值时的人均 10,000 美元。他说，在这一群体中出现的放缓“足以完全减缓收入增长，尽管我们看到了使用量的上升”。
Ramp 的数据显示，中等水平和前 10% 企业的支出有所增长，但这不足以弥补差距。中等水平企业每月人均支出为 12-15 美元，前 10% 的企业每月人均支出为 600-1,000 美元。
Kharazian 说：“如果你看看今天的使用场景，不一定能清楚地看到它们能达到前 1% 所处的每月人均 10,000 美元的标志。”虽然软件和金融等行业对 AI 编码代理的需求很大，“但医生诊所或汽车经销商并不一定需要在 AI 上花费那么多钱。”
“超增长时代可能已经结束，而现在我们知道从前 1% 身上榨取的利润可能没有那么多，”他继续说道。
10 月的数据确实出现了重新加速，令牌支出从 9 月 23 日到 10 月 23 日上涨了 8.9%，但 Kharazian 表示，支出增长的阻力依然强劲。
Kharazian（发布了一系列名为“AI 论点的裂痕”的博客）表示，这种放缓并非由于需求问题。再次强调，AI 使用量确实在上升。相反，他指出的是市场的竞争性。
他说：“OpenAI 和 Anthropic 的降价……只是让使用 AI 变得更便宜了。”9 月，OpenAI 将 GPT-6 Sol 和 GPT-6 Luna 的 API 价格减半。Anthropic 也对 Fable 5.1 和 Opus 5.5 的价格进行了大幅下调，以回应这场价格战——该公司在最近的一份文件中表示，这一决定使收入增长减少了约 6 亿美元。两家公司还一直在提供旨在锁定支出的企业折扣，但实际上却影响了它们的收入目标。
Kharazian 表示，这种动态意味着市场已经变成了买方市场。
为了锁定客户而补贴昂贵技术的使用一直是大型科技公司 playbook 的一部分。回想一下以前乘坐 Uber 或 Lyft 有多便宜，而现在由于网约车玩家寥寥无几，从 A 点到 B 点的成本已显著上升。Kharazian 表示，这些动态在 AI 世界中行不通，因为竞争太激烈了。
Kharazian 说：“Meta 和 Nvidia 也在推出具有非常便宜模型或开源模型的企业产品。”他还指出了来自中国和美国的其他开源模型公司的范围，以及一家名为 Jev 的新公司——一种用于在几个预定输出之间选择的轻量级工具——它正日益成为 Ramp 上表现最好的模型之一。
Kharazian 说：“对于许多 AI 使用场景，你只需要一个简单的‘是’或‘否’。”“所以你使用你已经很便宜的模型，将它们转移到 Jev，现在你原本就便宜的东西变得更便宜了。”
以更低的成本完成工作负载的能力是驱动支出放缓的另一个因素。这部分归因于 AI 公司的折扣，但也因为公司在变得节俭方面做得更好。
Kharazian 说：“当你在使用 AI 方面变得非常先进时，你就更擅长弄清楚如何控制它。”“你更有可能使用默认设置来防止在长时间运行的任务中使用非常昂贵的模型。你更有可能为员工设定一些预算。你更有可能通过路由器将部分标准模型和轻量模型的支出转移到更便宜的轻量模型上。”
Notion 联合创始人 Akshay Kothari 最近发布了与 Ramp 见解相匹配的内部数据。他用简洁的话总结道：“令牌滚滚而来（tokens go brrr）。成本啊啊啊地下降（costs go ahhh）。”
Kharazian 说：“我们对 AI 支出轨迹的许多假设开始瓦解。”“你可以使用更多的 AI，但花更少的钱。这对客户来说很好……但对模型公司来说并不一定是好事。这也引发了关于其收入轨迹的新问题。”
彭博社周五报道，Anthropic 的潜在估值已从 10 月初估计的 2 万亿美元降至 1.2 至 1.3 万亿美元。据报道，该公司预计还将获得信用评级机构的高“垃圾”评级。OpenAI 也将其提议的估值从 1.4 万亿美元降低到 1.25 万亿美元，并将其最新的融资目标削减了约 100 亿美元。
Kharazian 表示，Ramp 的数据集很可能解释了投资者的许多担忧。然而，支出轨迹有一个潜在的亮点，尽管这是推测性的。
超增长支出的大部分论点认为，AI 无疑将引入新的使用场景，从而加速发展。AI 编码一年半前才出现，作为最具商业可行性的产品，促使 Anthropic 和 OpenAI 成为万亿美元公司。投资者的希望是会有另一场革命推动进一步的支出。
有些人认为个人代理将是下一个编码代理。OpenAI、SpaceXAI 和 Meta 都已推出了自己的个人代理，初创公司如 Instinct 也是如此。不过，Kharazian 对个人代理持怀疑态度，他认为个人代理“在经济生产力上不如编码”，无法为行业所需的持续增长提供动力。
Kharazian 说：“预测是有新产品将被推出，这将推动进一步的增长。”“但基于我们今天可以观察到的情况、可用的产品以及商业化的进展，AI 支出已经开始放缓。”
当您通过文章中的链接购买时，我们可能会赚取少量佣金。这不会影响我们的编辑独立性。
查看原文
Companies are using more AI than ever, but they’re not spending more on it.
A new report from Ramp shows AI spending flatlined over late summer even as token volumes rose, a sign that price competition and savvier customers may be capping revenue for OpenAI and Anthropic. According to Ramp’s data, token spend only grew 1.5% across August and September. To compare, in July, it grew 42% from the previous month.
As Anthropic gears up for its IPO, the data suggests investors might be wrong to assume AI companies will keep growing at breakneck speed.
“Almost the entire spend on AI is from venture-backed, high-growth tech companies,” Ramp’s lead economist Ara Kharazian told TechCrunch. “That’s fine as long as AI spend continues to rise, but if it starts to slow down because of something in the market or some other factor, they will all start to slow down simultaneously, and that could lead to other cascading effects.”
Kharazian said he’s already seeing evidence that the top 1% of firms — the most “AI-pilled” — have stuttered slightly in their spend in August and September, coming up just below the $10,000 per-employee they were hitting at their summer peaks. A slowdown among that cohort is “sufficient to slow down revenue entirely, despite the fact that we’re seeing rising volumes,” he said.
Ramp data shows growth in spending among median and top 10%, but it’s not enough to make up the difference. Median firms spend $12-$15 per employee per month, and the top 10% spend $600-$1,000.
“If you look at the use cases today, it’s not necessarily clear they will get up to that $10,000 per employee per month mark that the 1% is at,” Kharazian said. While there’s plenty of demand for AI coding agents in industries like software and finance, “it’s not necessarily true that a doctor’s office or a car dealership will need to spend that much on AI.”
“The era of hypergrowth may be over, and now that we know there may not be as much juice to squeeze out of the top 1%,” he continued.
October data did see a re-acceleration, with token spend up 8.9% from September 23 to October 23, but Kharazian said the headwinds on spend growth are strong.
Kharazian (who has published a series of blogs called “Cracks in the AI Thesis”) said the slowdown isn’t due to a demand problem. Again, AI volumes are up. Instead he points to the competitive nature of the marketplace.
“OpenAI and Anthropic price cuts have…just made it cheaper to use AI,” he said. In September, OpenAI cut API prices in half for GPT-6 Sol and GPT-6 Luna. Answering the price war, Anthropic has also made significant reductions to pricing for Fable 5.1 and Opus 5.5 — a decision that the company said in a recent filing slashed revenue growth by about $600 million. Both companies have also been offering enterprise discounts that are intended to lock-in spend, but in reality are hitting their revenue targets.
Kharazian said this dynamic means it has very much become a buyer’s market.
Subsidizing use of expensive technology in order to lock in customers has always been a part of the Big Tech playbook. Just recall how cheap it used to be to take an Uber or Lyft ride, and now that there are few other ride-hail players, the cost of going from A to B has risen significantly. Kharazian says those dynamics don’t work in the AI world because there’s just too much competition.
“Meta and Nvidia are also introducing enterprise offerings with pretty cheap models or open source models,” Kharazian said. He also pointed to the range of other open model companies from China and the United States, as well as a new company called Jev — a lightweight tool for choosing between a few predetermined outputs — that is increasingly among the top performers on Ramp.
“For a lot of AI use cases, you just need a simple Yes/No,” Kharazian said. “So you take your already cheap models, you shift them over to Jev, and now your already cheap stuff gets even cheaper.”
The ability to complete workloads at a much cheaper cost is another factor driving the spending slowdown. That’s partly due to the discounts from AI companies, but also because companies are just getting better at being thrifty.
“As you get really advanced in using AI, you get better at figuring out how to control it,” Kharazian said. “You’re more likely to use defaults that prevent usage of very expensive models for long-running tasks. You’re more likely to impose some budgets on employees. You’re more likely to offload some of the spend from standard and light models to even cheaper light models via routers.”
Notion co-founder Akshay Kothari recently posted internal data that matched Ramp’s insights. As he succinctly put it: “tokens go brrr. costs go ahhh.”
“A lot of the assumptions that we had about the trajectory of AI spend are starting to break down,” Kharazian said. “You can use more AI, but use it for less. That’s great for the customers…but it’s not necessarily great for the model companies. And it does introduce new questions around their revenue trajectories.”
Bloomberg reported Friday that Anthropic’s potential valuation sank from the $2 trillion estimated in early October to just $1.2 to $1.3 trillion. The company is also reportedly expected to receive a high “junk” rating from credit agencies. OpenAI also lowered its proposed valuation to $1.25 trillion from $1.4 trillion and slashed its latest fundraising goals by about $10 billion.
Kharazian says Ramp’s dataset likely informs many investor concerns. There is, however, one potential brightside for spending trajectories, albeit a speculative one.
Much of the thesis for hypergrowth spending is that AI will undoubtedly introduce new use cases that will accelerate things. AI coding emerged just a year and a half ago as the most commercially viable product that has led Anthropic and OpenAI to becoming trillion-dollar companies. The hope among investors is that there will be another revolution that will drive further spend.
Some think personal agents will be the next coding agents. OpenAI, SpaceXAI, and Meta have all released their own personal agents, alongside startups like Instinct. Kharazian is skeptical, though, that personal agents, which are “not as economically productive as coding,” will provide the boost the industry needs for nonstop growth.
“The projection is that there will be new products being introduced that will drive further growth,” Kharazian said. “But based on what we can observe today and the products available and what is commercially advanced, AI spend has started to slow down.”
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
原文版权归 TechCrunch · AI 及原作者所有，本站转载仅供阅读。如需删除，请联系 support@aisa.one。
Studio