
What happened
Ramp's September 2026 AI Index shows median per-employee AI spending at the top 1 percent of US companies fell 9.7 percent in August to $7,205, as overall adoption kept rising.
Why it matters
The effective price per million tokens has dropped 41 percent from its March 2026 peak to $0.68, and frontier models fell from 53 percent of tokens in early August to 45 percent by early September.
What to watch
Ramp's Ara Kharazian says volume may not grow fast enough to offset falling prices, so much hinges on whether Anthropic's reportedly planned October IPO filing clarifies the picture.
WHO IT HITSEnterprise AI budget owners at large US companies are already redirecting spending toward cheaper standard models and writing internal policies to restrict expensive frontier models, while model providers like OpenAI and Anthropic face pressure on per-token revenue.
Ask the AI about this article →
Summaries like this, in your inbox every morning.
Ramp has tracked monthly US business spending on AI services through its AI Index, and the September 2026 edition lands at a moment when the top spenders — the firms that drive most enterprise revenue for model providers — are pulling back. Adoption is still rising but at a slower pace: in August, 43.8 percent of US companies paid for Anthropic services, up 0.34 percentage points, while OpenAI gained 0.09 points to 39.8 percent. IT and finance sectors continue to lead.
The cutback among the largest spenders coincides with two other forces Ramp identifies: falling token prices and a steady migration to cheaper models. The effective price per million tokens has dropped 41 percent since its March 2026 peak to $0.68, and frontier models like Opus, Fable, and Sol saw their share of consumed tokens fall from 53 percent in early August to 45 percent by early September, as cheaper standard models such as GPT-5.6 Terra and Claude's Sonnet series picked up volume. Ramp chief economist Ara Kharazian attributes part of the August drop to seasonality, but also points to internal company policies restricting expensive frontier models. Open-weight or Chinese alternatives are not driving the shift: only 6.4 percent of AI-using companies on Ramp's platform run open-weight models, dropping to 3.6 percent across all companies.
Kharazian last month flagged what he called "Cracks in the AI Thesis," citing weak Fable 5 adoption and the shift to cheaper models as warning signs for providers. The key question is whether usage volume grows fast enough to offset the price decline — Kharazian suggests it may not. Ramp's data covers only a slice of the market, and things may get clearer when Anthropic files for its IPO, reportedly planned for October.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Michael Burry, famous for The Big Short, is short Nvidia, Palantir and Tesla, and in his Substack newsletter s…

Investors have three creative routes to Anthropic exposure before its expected IPO: buying Alphabet, Amazon, o…

DeepSeek launched V4.1-Flash, a 763B-parameter open-weight model with a causal encoder-decoder architecture

Much of the attention on AI infrastructure buildouts is now tied to sheer compute power, with dominance define…

Barron's reported September 10 that Kepler Computing emerged from stealth with a memory architecture using fer…

Dynatrace acquired Arize AI, adding AI observability, evaluation and agent monitoring to its application obser…