Ramp data reveals decline in per-employee AI spending at top firms amid lower token costs
Article Summary
Business adoption of AI tools experienced a slowdown in August, with only a 0.4% increase in paying customers, reaching 56% of Ramp clients, a metric that has caused concern given the significant investments in AI infrastructure. Spending per employee in top-tier firms decreased by nearly 10% to $7,205, coinciding with a drop in average token costs to $0.68 per million tokens due to price cuts by major AI providers like OpenAI and Anthropic, which may deter the adoption of newer, more expensive models. Despite competition driving down prices, the anticipated revenue growth from increased usage has not materialized as expected, impacting the financial models of AI labs and hyperscalers heavily invested in chip orders.
AI Analysis
## Core Assessment
Ramp’s latest spending data indicates a plateau in business AI adoption, with paid usage among its 70,000 clients growing by a negligible 0.4% in August. While seasonal vacations likely account for some of this lull, the underlying metrics reveal a structural shift: spending per employee at top-tier firms dropped nearly 10% to $7,205. This decline aligns with a significant reduction in average token costs—down to $0.68 per million from a March peak of $1.15. The data suggests that price competition between frontier labs is currently outpacing volume growth, forcing enterprises to shift toward cheaper, older models rather than premium-priced frontier releases. For hyperscalers and model labs banking on a "build it and they will pay" revenue model, the lack of a corresponding volume surge represents a genuine financial friction point.
## Impact and Advice
* **Enterprise Procurement Officers:** Audit your current AI spend; with token prices dropping, you may be overpaying for "frontier" tiers when older, cheaper models now handle most routine tasks with sufficient capability.
* **AI Lab Financial Planners:** Reassess revenue projections; the data suggests that lower unit costs are currently cannibalizing top-line growth rather than stimulating enough volume to offset the price drops.
* **Infrastructure Architects:** Evaluate inference platform utilization; with only 6.4% of firms adopting model-serving platforms, focus on operational efficiency for current deployments rather than scaling infrastructure for projected demand that has yet to manifest.
## Contrarian View
The August data may prove to be a statistical outlier rather than a trend. Ramp’s user base skews heavily toward technology-forward companies, which contrasts sharply with the broader U.S. Census Bureau finding that only 22% of all businesses utilize AI. It is possible that the "slowdown" reflects a market maturation phase where companies are finally optimizing their spend—a healthy sign for long-term sustainability—rather than a rejection of the technology itself.
## Bottom Line
Lower token prices are effectively subsidizing corporate AI budgets, but they are not yet driving the massive volume growth needed to justify the current scale of hyperscaler capital expenditure.
*It is the classic hardware-software trap: the labs are winning the price war, but they are losing the revenue race. If the "killer app" doesn't arrive soon to drive up volume, the accountants are going to start asking very uncomfortable questions about those massive chip orders.*
Ramp’s latest spending data indicates a plateau in business AI adoption, with paid usage among its 70,000 clients growing by a negligible 0.4% in August. While seasonal vacations likely account for some of this lull, the underlying metrics reveal a structural shift: spending per employee at top-tier firms dropped nearly 10% to $7,205. This decline aligns with a significant reduction in average token costs—down to $0.68 per million from a March peak of $1.15. The data suggests that price competition between frontier labs is currently outpacing volume growth, forcing enterprises to shift toward cheaper, older models rather than premium-priced frontier releases. For hyperscalers and model labs banking on a "build it and they will pay" revenue model, the lack of a corresponding volume surge represents a genuine financial friction point.
## Impact and Advice
* **Enterprise Procurement Officers:** Audit your current AI spend; with token prices dropping, you may be overpaying for "frontier" tiers when older, cheaper models now handle most routine tasks with sufficient capability.
* **AI Lab Financial Planners:** Reassess revenue projections; the data suggests that lower unit costs are currently cannibalizing top-line growth rather than stimulating enough volume to offset the price drops.
* **Infrastructure Architects:** Evaluate inference platform utilization; with only 6.4% of firms adopting model-serving platforms, focus on operational efficiency for current deployments rather than scaling infrastructure for projected demand that has yet to manifest.
## Contrarian View
The August data may prove to be a statistical outlier rather than a trend. Ramp’s user base skews heavily toward technology-forward companies, which contrasts sharply with the broader U.S. Census Bureau finding that only 22% of all businesses utilize AI. It is possible that the "slowdown" reflects a market maturation phase where companies are finally optimizing their spend—a healthy sign for long-term sustainability—rather than a rejection of the technology itself.
## Bottom Line
Lower token prices are effectively subsidizing corporate AI budgets, but they are not yet driving the massive volume growth needed to justify the current scale of hyperscaler capital expenditure.
*It is the classic hardware-software trap: the labs are winning the price war, but they are losing the revenue race. If the "killer app" doesn't arrive soon to drive up volume, the accountants are going to start asking very uncomfortable questions about those massive chip orders.*