AI spending growth stalls despite rising token volumes, Ramp data reveals
Enterprise AI spending flatlined in late summer as price competition erodes vendor revenue, raising questions about growth sustainability ahead of Anthropic's IPO.
Last verified:
Token Growth Decoupling from Revenue
Spending growth among enterprise customers has stalled even as AI token consumption accelerates—a divergence that signals price competition is eroding margins faster than volume gains can offset. According to TechCrunch AI, citing Ramp’s spending data, token spend across customer bases grew only 1.5% in August and September combined, down sharply from July’s 42% month-over-month increase.
This deceleration reflects a structural shift in the AI vendor market. Ramp’s lead economist Ara Kharazian tells TechCrunch that the slowdown is not driven by flagging demand; token volumes are rising consistently. Instead, aggressive pricing moves by OpenAI and Anthropic have decoupled revenue growth from usage growth. In September, OpenAI slashed API pricing in half for GPT-6 Sol and GPT-6 Luna. Anthropic responded with “significant reductions” for Fable 5.1 and Opus 5.5—a decision Anthropic disclosed in a recent filing reduced projected revenue growth by approximately $600 million.
Enterprise Discounts Are Masking Revenue Pressure
Both vendors have deployed enterprise discounts ostensibly to lock in customer commitments, but the practical effect has been to compress per-unit economics without securing corresponding volume uplift. Kharazian describes the market dynamic as now “very much a buyer’s market,” where customer leverage has shifted decisively away from vendors.
The top 1% of spending firms—predominantly venture-backed, high-growth tech companies—have been the hardest hit. This cohort, which dominated AI adoption in 2026, peaked at approximately $10,000 per employee per month in mid-summer but fell just below that threshold in August and September. Because this segment accounts for nearly all enterprise AI spend, a contraction here cascades across the entire vendor revenue base. Median-spend firms allocate $12–$15 per employee monthly; the top 10% spend $600–$1,000. The top 1%’s stumble outweighs gains from these lower tiers.
The Scalability Question
Kharazian suggests the slowdown signals a ceiling on AI spending penetration. While software engineering and finance—high-value use cases for AI coding agents—continue to scale, generalist adoption among service-sector businesses (medical offices, car dealerships) may not reach the $10,000-per-employee monthly benchmark, even with declining unit costs. This implies a narrower addressable market for vendors than consensus assumptions had priced in.
October data showed a partial recovery—token spend rose 8.9% from late September to late October—but Kharazian characterizes the headwinds on spending growth as persistent. He has published a series titled “Cracks in the AI Thesis” documenting systemic challenges to the vendor growth narrative.
Why This Matters
The timing of this slowdown is significant for Anthropic’s impending IPO. Ramp’s data suggests that public-market investors may be pricing in continued hypergrowth without accounting for price-war dynamics and a contracting customer base (venture-backed tech firms). If the top 1% of spenders cannot sustain $10,000-per-employee-per-month benchmarks indefinitely, and lower-tier customers show no path to that threshold, vendor revenue growth will decelerate structurally—independent of token-volume increases. This shifts the narrative from “AI spending will grow forever” to “AI adoption may plateau sooner than expected, with margins under permanent pressure.” Finance and software firms will remain high-value cohorts, but the vision of ubiquitous, high-revenue AI consumption across all industries may require a new paradigm to materialize.
Frequently Asked Questions
Why is token usage growing if spending is flat?
Price competition has lowered the per-token cost. Customers are consuming more AI inference but paying less in aggregate because OpenAI and Anthropic have both cut API pricing and deployed enterprise discounts.
Which customer segment is driving the slowdown?
The top 1% of AI-consuming firms—primarily venture-backed tech companies—reduced spending in August and September below their summer peak of ~$10,000 per employee per month.
Is this a demand problem?
No. Ramp's economist Ara Kharazian attributes the slowdown to supply-side price compression, not lack of customer interest. Token volumes are up; revenue per token is down.
Could this affect Anthropic's IPO valuation?
The data suggests investor assumptions about sustained hypergrowth may be too optimistic, potentially influencing public-market expectations and comparable-company multiples.