AI Expenditure Hits a Standstill as Anthropic Approaches IPO, According to Ramp Data

While the adoption of AI technology continues to rise, expenditure on it has remained stagnant.

A recent analysis from Ramp indicates that AI spending remained steady through the latter part of summer, despite increasing token volumes. This trend suggests that intense competition and more informed buyers could be limiting earnings for companies like OpenAI and Anthropic. The data reveals that token expenditures grew merely 1.5% in August and September, substantially lower than the 42% increase witnessed in July.

As Anthropic prepares for its initial public offering (IPO), these findings raise questions about the sustainability of rapid growth in the AI sector.

“Most of the spending on AI comes from venture-backed technology firms aiming for high growth,” explained Ara Kharazian, Ramp’s chief economist, in an interview. “While this is manageable as long as AI spending increases, a decline could simultaneously hit all these companies, resulting in further repercussions.”

Kharazian noted a slight reduction in spending among the top-performing companies in August and September, indicating their expenses fell just short of the $10,000 per employee they had achieved earlier in the summer. This downturn could lead to an overall revenue decline, even amid rising token usage.

Data from Ramp indicates some growth in spending patterns among median firms and the top 10%, but it’s insufficient to offset the decline from peak spending levels. Median firms typically invest between $12 and $15 per employee monthly, while the top 10% utilize $600 to $1,000.

“Current use cases don’t necessarily indicate that other businesses will reach the $10,000 per employee monthly threshold that the top 1% have achieved,” Kharazian mentioned. While there’s a clear demand for AI solutions in areas like finance and software, industries such as medical offices or car dealerships may not require such high levels of investment in AI.

“The phase of rapid growth could be at an end, especially as it becomes evident that the top 1% may not yield as much potential,” he added.

Data from October indicated a slight rebound, with token spending increasing by 8.9% from September to October. However, Kharazian warned that the challenges to spending growth remain significant.

The slowdown is not attributed to a lack of demand, as AI usage is on the rise. Instead, Kharazian points to the competitive landscape as a major influence.

“The price reductions from OpenAI and Anthropic have made AI usage more affordable,” he noted. OpenAI halved the costs for its GPT-6 Sol and GPT-6 Luna APIs. In response, Anthropic also lowered its prices for Fable 5.1 and Opus 5.5, actions that are estimated to have reduced their revenue growth by as much as $600 million. Both companies have introduced enterprise discounts, aiming to secure customer spending, but these strategies have, in practice, impacted their revenue targets.

Kharazian emphasized that this scenario has transformed the market into a buyer-centric environment.

Historically, large tech firms have subsidized expensive technologies to retain customers. For example, the initial low fares of rides through services like Uber and Lyft increased significantly as competition waned. However, Kharazian believes that similar tactics in AI may falter due to the high level of competition.

“Companies like Meta and Nvidia are launching competitive enterprise solutions with affordable or open-source models,” he said. He also pointed out various emerging companies from the U.S. and China, including Jev, which offers a streamlined tool for selecting from preset options and is gaining traction according to Ramp’s data.

“Many AI applications only require a simple Yes/No response,” Kharazian explained. “By utilizing low-cost models and switching over to Jev, businesses can lower their expenses even more.”

The ability to execute tasks at a reduced cost is another contributing factor to the slowdown, driven by both the discounts offered by AI firms and improved frugality among users.

“As companies become more adept at integrating AI, they learn to manage costs more effectively,” Kharazian stated. “They tend to employ default settings that minimize reliance on expensive models for prolonged tasks. Companies are also more likely to impose spending limits on employees and transition lighter workloads to even more economical models.”

Akshay Kothari, co-founder of Notion, corroborated Ramp’s observations with internal data, succinctly capturing the essence of the situation: “tokens go brrr. costs go ahhh.”

“Many previous expectations regarding the growth trajectory of AI spending are now being challenged,” said Kharazian. “Increased usage does not necessarily equate to higher expenditure. While beneficial for users, this could spell trouble for the profit models of AI companies and raises questions about their revenue projections.”

Recent reports indicate that Anthropic’s potential valuation has dropped significantly from an early October estimate of $2 trillion to between $1.2 and $1.3 trillion. The company is also anticipated to receive a high-risk rating from credit agencies. Likewise, OpenAI has reduced its projected valuation from $1.4 trillion to $1.25 trillion, cutting its fundraising expectations by approximately $10 billion.

Kharazian suggests that the data available to Ramp may reflect broader investor anxieties. Yet there remains a speculative glimmer of hope for future spending trends.

The premise of continued high growth relies on the emergence of new use cases that could invigorate the market. AI coding capabilities emerged as a prominent commercial product just 18 months ago, propelling both Anthropic and OpenAI to trillion-dollar valuations. Investors are hopeful that a similar breakthrough will spark further expenditure.

Some theorize that personal agents could replicate the success of coding solutions. Companies like OpenAI, SpaceXAI, and Meta have introduced their versions, alongside startups like Instinct. However, Kharazian expresses skepticism that personal agents, being less economically productive than coding agents, will provide the necessary impetus for sustained industry growth.

“The expectation is that new products will emerge to stimulate additional growth,” Kharazian noted. “Nonetheless, based on present observations and the commercially advanced solutions available, AI spending appears to be entering a slowdown phase.”