Ramp Data: 1% of Companies Account for 80% of OpenAI and Anthropic Commercial Spending, Token Usage Rises While Payments Stall

The commercialization narrative of AI large models is facing challenges at the data level. After tracking payment data from more than 70,000 U.S. businesses, corporate card giant Ramp found that just 1% of companies account for 80% of commercial spending on OpenAI and Anthropic, and this highly concentrated demand is hitting a ceiling—token usage continues to climb, but actual dollar spending growth has nearly stalled, and paid-user penetration is flattening.

In his October AI index report, Ramp chief economist Ara Kharazian noted that token spending saw "almost zero growth" for two consecutive months in August and September, and although it barely set a record high in early October, it was only 1.5% above the August peak. The core driver of the spending stagnation is price deflation: frontier model prices keep falling, and combined with improving model efficiency, the blended average price per million tokens keeps declining, making the divergence between volume and price increasingly evident.

This signal has already alarmed mainstream institutions. Goldman Sachs warned that token demand growth must outpace price declines to support sustained computing investment; Apollo chief economist Torsten Slok, citing Ramp data, noted that the computing demand curve is "flat and highly price-sensitive," with enterprise buyers cutting usage or switching to cheaper open-source alternatives at the slightest price movement. Meanwhile, Citadel founder Ken Griffin said publicly that the market is "almost certainly" in an AI bubble.

Demand is highly concentrated, with whale customers propping up revenue

AI enterprise spending is the most concentrated of any software category. Ramp data shows that venture capital-backed startups spend $76 per person per month on AI, more than 10 times that of private equity-backed companies.

This top-heavy pattern is especially pronounced at Anthropic. According to YipitData, Anthropic's top 1% of customers contributed 46% of revenue in August, nearly double the 25% a year earlier. Anthropic's prospectus shows that just two undisclosed customers accounted for nearly a quarter of its 2025 revenue.

Capital flows reveal the internal circulation of this ecosystem: Cursor parent company Anysphere completed a $2.3 billion funding round last November at a $29.3 billion valuation, co-led by Accel and Coatue, with participation from Thrive Capital, Andreessen Horowitz, Nvidia and Google; three months later, Coatue and Singapore's GIC co-led Anthropic's $30 billion Series G, with Nvidia investing again.

Coatue's capital flows into Cursor, Cursor converts it into Claude token consumption, token consumption is counted as Anthropic revenue, and Anthropic revenue in turn supports the valuation premium on the Anthropic equity held by Coatue.

Price war intensifies, Anthropic's effective token price halves

Model prices are collapsing rapidly. The Ornn Anthropic Token Price Index—which measures the actual per-token cost paid by customers—has fallen from more than $1.70 in mid-September to $0.83, a record low since the benchmark was launched in June.

Even so, Anthropic remains more expensive than OpenAI. OpenAI's corresponding index currently stands at $0.45. OpenAI's effective pricing collapsed sharply at the end of July—when it cut prices for its GPT-5.6 Luna and Terra models by up to 80%, just three days after Chinese large-model company Moonshot released the open-weight Kimi K3, which is believed to have been the trigger that sent chip stocks into a bear market that month. OpenAI has since leaned toward postponing its IPO to 2027.

The "Jevons paradox" has failed to materialize

In a Sept. 25 report, Goldman Sachs noted, based on Anera and OpenRouter data, that total token demand has grown about 18-fold since last December, but demand for high-margin frontier models has grown at only half that rate. The margin compression caused by falling prices has already fed through to the data center investment layer. Goldman Sachs estimates that the realized return on invested capital (ROIC) of five companies—Amazon, Alphabet, Meta, Microsoft and Oracle—has fallen from 40% to 25%, the lowest level in its dataset since 2016.

Torsten Slok was blunt about this: "The premise in Silicon Valley is that computing demand is unlimited, but the recent slowdown in token spending growth suggests that such demand may exist only when computing is nearly free."

He warned that the core variable determining whether the data center construction model is viable is neither physical capacity nor power supply, but "the price users are actually willing to pay for computing"—and support for that price depends on whether companies can convert token consumption into revenue growth, margin improvement or substantive cost savings. "At present these benefits have not yet shown up at scale. They may in the future, or they may not."

Paid penetration is saturating, and growth must come from depth rather than breadth

Beyond the slowdown in top-tier spending, expansion of the paid user base has also hit a bottleneck. Ramp data shows that in September, paid AI adoption across the corporate network rose only 0.62 percentage points, to 55.9%.

Kharazian wrote in the report: "It feels like we are already at the top of the S-curve, and market growth will have to rely on deeper usage rather than more paying users."

This stands in sharp contrast to comments by Ramp product director Adam Sommer in June this year—he said at a forum in New York that "our largest AI-spending companies on the Ramp platform have doubled their spending every month over the past few months." Three months later, per-person spending among the top 1% of companies has fallen below its July peak. Risk disclosure and disclaimer

Markets carry risk, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation or needs of individual users. Users should consider whether any opinions, views or conclusions in this article are suitable for their particular circumstances. Invest at your own risk.