The AI Spending Boom Has a Problem: Most Companies Spend Very Little

Businesses are using more AI, so you'd expect them to be spending a lot more money on it, too. Surprisingly, the latest figures tell a different story.

According to the Ramp AI Index, AI spending among its US business customers reached another record in October 2026, but it grew just 1.65% from September.

Meanwhile, more companies are paying for AI services. Ramp found that 56.8% of businesses on its platform purchased an AI product or service in October, up one percentage point from the previous month.

So, why aren't their bills growing as quickly?

Part of the explanation is that AI is getting cheaper. Companies can process more information for less money, while much of the industry's spending still comes from a relatively small group of customers.

For OpenAI and Anthropic, the challenge is getting businesses to spend more on AI when the technology itself costs less to use.

A few companies are spending thousands on AI. Most spend far less.

The difference between what ordinary businesses and the biggest AI customers spend is pretty surprising.

In his November 11 analysis, Ramp economist Ara Kharazian found that the median business paying for AI spent just $16.32 per employee in October. That's an increase of only 93 cents from September.

Now compare that with the top 1% of AI spenders.

Those companies spent a median of approximately $10,200 per employee per month, a 9% increase. And the spending concentration goes beyond those figures.

According to Ramp, the top 1% of companies accounted for roughly 80% of the OpenAI and Anthropic revenue it observed. Many of those big spenders are venture-backed businesses in technology or finance, where AI already plays a major role in software development and other technical work.

For OpenAI and Anthropic, having customers willing to spend thousands per employee is encouraging. The concern is how much revenue depends on those customers.

If a few major buyers reduce their budgets or move to cheaper models, the effect could be considerable.

Meanwhile, spending among more typical businesses is increasing much more slowly. AI providers will need to persuade more companies to expand their usage if they want revenue growth to keep up with their investment plans.

Companies are using more AI because it's getting cheaper

Here's another finding from Ramp's report: the average price per million AI tokens fell 21% over the previous month.

Tokens are the pieces of text AI models process when they read a request or produce an answer. Businesses that access models through an API often pay according to how many tokens they use.

As those prices fall, the same budget can cover more AI activity.

Say a company uses AI to answer customer questions. If the model becomes cheaper, the business could handle more conversations while spending roughly the same amount.

Ramp's token-spending data shows something similar.

Among businesses using its Token Spend Management service, token usage rose 29% over a comparable four-week period, while spending increased 10.5%.

Those figures come from a smaller group of companies than Ramp's main AI Index, but they show how usage can grow faster than spending.

For businesses, that's welcome news. More affordable models could make AI tools accessible to companies that previously couldn't justify the expense.

The economics are trickier for AI providers.

OpenAI and Anthropic can serve more requests, but lower prices mean they may earn less from each one. They'll need customers to increase usage enough to compensate, or find additional services businesses are willing to pay for.

There's also more competition. Companies can now choose from several AI providers, including cheaper models that can handle routine tasks.

As customers compare costs, AI providers have another reason to keep prices competitive.

Coding still accounts for most AI spending. Can customer service catch up?

If AI companies want businesses to spend more, they'll need to find applications beyond software development that justify larger budgets.

For now, coding remains the biggest source of demand in Ramp's department-level data.

Engineering teams accounted for 72% of all classified AI spending in October, although spending had fallen from its summer highs. AI coding assistants have become popular among developers because they can help write code and complete repetitive programming tasks.

Customer service is beginning to account for a larger share of spending, too.

Ramp found that customer success and support departments represented 13.3% of classified AI spending in October, up from just 0.8% in June.

Kharazian suggested that AI customer service chatbots could be contributing to the increase, although the spending data doesn't confirm exactly which applications are responsible.

The opportunity is easy to understand. Companies receive plenty of routine customer questions, and AI could help answer them without requiring an employee to handle every request.

But there's a catch for AI providers.

A chatbot that answers questions about delivery schedules or return policies probably doesn't need the most expensive AI model available. Businesses may prefer cheaper models that respond quickly and keep operating costs low.

So even if customer service becomes a major source of AI usage, it may generate less revenue per request than applications that require more advanced models.

For OpenAI and Anthropic, customer support could bring in more business, although it remains unclear whether spending will grow as quickly as it did during the coding assistant boom.

What does slower spending growth mean for OpenAI and Anthropic?

Both companies are still attracting business customers, according to Ramp.

In October, Anthropic's paid adoption rate reached 44.8% among businesses on the platform, up 1.1 percentage points from September. OpenAI remained near 40.6%.

These figures measure the share of Ramp businesses paying each provider, so a company that subscribes to both can appear in both percentages. They also shouldn't be confused with worldwide market share.

Still, the figures suggest that AI adoption continues even as spending increases more slowly. Meanwhile, both providers are investing heavily in the infrastructure needed to deliver their services.

As The Neuron recently reported, Anthropic reportedly agreed to a $3.4 billion arrangement to secure gas-powered generators for future data centers. Power is only one of the expenses involved in operating large AI systems. Companies also need expensive computing equipment, and training new models requires substantial resources.

Lower prices could help attract more customers, but AI providers still need enough revenue to cover those costs. As The Neuron's AI economics explainer discussed, the amount of computing capacity available can also affect how quickly AI companies expand their services.

So, is the AI spending boom slowing down?

Ramp's latest figures suggest businesses are becoming more selective about AI spending, though lower model prices also help explain the slower growth.

There's an important limitation to keep in mind.

The AI Index uses transaction data from over 85,000 American businesses on Ramp's financial platform. Its customers tend to be more technology-oriented than the broader US business population. The figures also miss some AI activity, such as employees using free tools or personal accounts for work.

For businesses considering AI, falling model prices could be an opportunity. A company that previously couldn't afford to automate customer inquiries or other repetitive work may find those applications more affordable.

Of course, cheaper AI is only worthwhile if it performs the tasks reliably and saves enough time or money to justify the investment.

For OpenAI and Anthropic, the next challenge is convincing more businesses to increase their AI budgets beyond coding applications.

More companies are paying for AI, and they're using it more often. Whether that translates into the revenue growth AI providers need is becoming a much more interesting question.