Ramp data shows AI spending slows amid Anthropic's IPO preparations

icon币界网
Share
AI summary iconSummary
On-chain data from Ramp shows AI spending slowed in August and September, rising just 1.5% compared to 42% in July. Ara Kharazian noted that most spending comes from high-growth, venture-backed firms. The decline in the top 1% segment outweighs gains from mid-tier companies. Price cuts by OpenAI and Anthropic are driving users toward cost efficiency. Inflation data from Ramp highlights growing concerns about the sustainability of AI firms, especially as Anthropic plans an IPO.
CoinMarketCap reports:

Companies are using AI more frequently than ever, but their spending on it has not increased proportionally.

A new report from Ramp shows that, despite rising token usage, AI spending plateaued in late summer, suggesting that price competition and more savvy customers may be limiting revenue for OpenAI and Anthropic. According to Ramp’s data, token spending in August and September increased by only 1.5%. In contrast, spending in July rose 42% from the previous month.

As Anthropic prepares for its IPO, this data suggests that investors should not assume AI companies will continue to grow at such an astonishing rate.

Ara Kharazian, Chief Economist at Ramp, told TechCrunch: “Virtually all AI spending comes from venture-backed high-growth tech companies. That’s fine while AI spending continues to rise, but if it begins to slow due to factors in the market or otherwise, they’ll all slow down simultaneously, which could trigger other ripple effects.”

Kharazian said he has seen evidence that the top 1% of companies—the most "AI-obsessed"—experienced a slight slowdown in spending in August and September, falling slightly below the peak summer level of $10,000 per employee. He noted that this group’s slowdown is “enough to slow overall revenue growth, even though we’re seeing usage rise.”

Ramp data shows that spending among median and top 10% companies continues to rise, but not enough to close the gap. Median companies spend $12 to $15 per employee per month, while top 10% companies spend $600 to $1,000 per employee per month.

Kharazian said, “Looking at today’s use cases, it’s not clear they will reach the level of $10,000 per employee per month that today’s top 1% of companies are achieving.” He noted that industries like software and finance have significant demand for AI coding agents, but “a doctor’s office or car dealership may not need to spend as much on AI.”

He continued, "The era of supercharged growth may be over; now we know that the potential for extracting further growth from the top 1% of companies may be more limited than previously thought."

The October data indeed shows a renewed acceleration, with token spending from October 23 to October 23 increasing by 8.9% compared to the period from September 23 to October 23; however, Kharazian noted that headwinds facing spending growth remain strong.

Kharazian published a series of blogs titled "Cracks in the AI Thesis," stating that the slowdown is not due to demand issues—AI usage continues to rise. Instead, he attributes the cause to market competition.

“The price cuts from OpenAI and Anthropic... are simply making AI usage cheaper,” he said. In September, OpenAI halved the API prices for GPT-6 Sol and GPT-6 Luna. In response to the price war, Anthropic also significantly lowered the pricing for Fable 5.1 and Opus 5.5—the company stated in a recent filing that this decision reduced revenue growth by approximately $600 million. Both companies are also offering enterprise discounts aimed at locking in spending, but are effectively impacting their revenue targets.

Kharazian stated that this dynamic means the market has clearly become a buyer's market.

Subsidizing the use of expensive technology to lock in customers has long been a common strategy among big tech companies. Consider how cheap Uber or Lyft used to be; now, with fewer ride-hailing platform players, the cost of getting from point A to point B has risen significantly. Kharazian says this dynamic won’t work in AI because the competition is too fierce.

“Meta and Nvidia are also launching enterprise products that offer relatively inexpensive models or open-source models,” said Kharazian. He also mentioned other open-source model companies from China and the United States, as well as a new company called Jev—a lightweight tool for selecting among a few preset outputs—that is increasingly appearing on Ramp’s high-performance list.

“For many AI use cases, you only need a simple yes or no,” Kharazian said. “So you connect your already inexpensive model to Jev, and now what was already cheap becomes even cheaper.”

The ability to accomplish workloads at a lower cost is another factor driving slower spending. This stems partly from discounts offered by AI companies, but also because businesses themselves are becoming more cost-conscious.

“When you become very skilled at using AI,” Kharazian said, “you’ll be better at controlling it. You’re more likely to stick with default settings and avoid running long tasks on very expensive models. You’re more likely to set budgets for your team. You’re also more likely to route some standard and lightweight workloads to cheaper, lighter models through a router.”

Notion co-founder Akshay Kothari recently shared internal data that aligns with Ramp’s observations. He succinctly wrote: “Tokens go brr. Costs go ahhh.”

Kharazian said: “Many of our past assumptions about AI spending trajectories are beginning to break down. You can use more AI while spending less. This is great for customers... but not necessarily good for model companies. It also introduces new challenges for their revenue trajectories.”

Bloomberg reported on Friday that Anthropic’s potential valuation has dropped from an estimated $2 trillion in early October to between $1.2 trillion and $1.3 trillion. The company is also said to be receiving a higher “junk” rating from credit rating agencies. OpenAI has similarly reduced its proposed valuation from $1.4 trillion to $1.25 trillion and cut its latest fundraising target by approximately $10 billion.

Kharazian said that Ramp's dataset likely influenced many investors' concerns. However, there is also a potential bullish aspect to the spending trajectory, though this is merely speculative.

One theoretical basis for rapid spending growth is that AI will undoubtedly create new use cases, further driving growth. AI coding emerged as the most commercially viable product one and a half years ago, propelling Anthropic and OpenAI to become trillion-dollar companies. Investors hope another revolution will emerge in the future to spur even more spending.

Some believe personal agents will be the next coding agents. OpenAI, SpaceX AI, and Meta have all released their own personal agents, with startups like Instinct also entering the space. However, Kharazian is skeptical whether personal agents—“less economically productive than coding”—can provide the sustained momentum the industry needs.

“Our prediction is that new product launches will drive further growth,” said Kharazian. “However, based on what we can observe today, along with the current state of our products and their commercialization, AI spending has already begun to slow.”

Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information. Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.