US companies’ AI spending hits a record, but growth slows

US companies’ spending on artificial intelligence reached a record level in October 2026, rising 1.65% over the month. According to the Ramp AI Index, the top 1% of companies by spending paid an average of $10,200 per employee per month, while a typical company spent $16.32. Software development teams accounted for 72% of all tracked AI spending. Meanwhile, OpenAI and Anthropic’s revenue growth still depends on a narrow group of customers: Ramp estimates that 1% of companies account for 80% of both companies’ revenue.

US companies’ AI spending hits a record, but growth slows

The Ramp AI Index tracks spending by customers of the financial services provider Ramp. Its dataset includes payment information from more than 85,000 companies in the US, but the index may not reflect the entire market: free AI tools and services used through personal accounts may not be captured.

Spending per employee among companies in the top 1% rose 9% over the month, reaching a median of $10,200. A typical company spent $16.32 — less than a dollar more than the month before. Separate figures on token volumes and related spending are available only for companies using Ramp Token Spend Management, and do not show spending across the platform as a whole. Ramp’s customer base is more likely than the overall US business population to include fast-growing, tech-savvy and larger companies, so the index may overstate AI adoption.

OpenAI grows faster than Anthropic

Business customers’ spending on OpenAI increased 2.1% over the month, while spending on Anthropic rose 0.5%. Ramp chief economist Ara Kharazian noted that the structural risks he had described earlier remain: 1% of companies account for 80% of OpenAI and Anthropic revenue.

This group consists mainly of venture-backed companies in technology, finance and technical services. Outside these industries, AI adoption remains limited.

At the same time, data processing is getting cheaper: the average price per million text tokens fell 21% over the past month. Usage is growing faster than spending, making AI cheaper. Ramp attributes this to competition between OpenAI and Anthropic, as well as open models from China.

SpacexSI grew 2.2%, becoming the fastest-growing software company in Ramp’s entire dataset. However, this result likely says more about sampling bias than about Grok’s success. The company’s models do not stand out from competitors, and the growth is believed to be linked to Cursor, a programming tool owned by SpacexSI. The company’s share of the Ramp platform is 6.3%, but this figure likely does not reflect the market as a whole.

Anthropic’s adoption reached 44.8%, up 1.1 percentage points. OpenAI remained at 40.6%, while overall adoption of AI services rose to 56.8%.

Customer support spending climbs

In October, software development teams accounted for 72% of all tracked AI spending, although their spending fell from its summer peak. The share of spending by customer support and customer success departments grew fastest, rising from 0.8% in June to 13.3% in October. One possible reason is the introduction of chatbots in customer support. Management teams also began spending slightly more.

According to Kharazian, the rise in customer support spending is still not comparable to the increase software development teams showed earlier this year. These tasks are also more often suited to cheaper and faster models than to the most expensive frontier models.

For OpenAI and Anthropic, this could mean that new customers bring lower margins. At the same time, the share of the most expensive models — Opus, Fable, Sol and Astra, in Kharazian’s classification — edged up in October, reaching 69.5% of all spending.

Dataist’s take: Record spending does not yet mean that AI is widely adopted by US companies. Most of OpenAI and Anthropic’s revenue depends on a small group of customers, and the Ramp index may overstate adoption. At the same time, spending growth is shifting toward customer support, where companies are more likely to choose cheaper models.