Ramp's heaviest AI spenders cut per-employee spend 9.7% to $7,205 in August

Work backwards from the $7,205 median. A 9.7 percent decline implies July sat near $7,979 per employee, so the heaviest buyers took roughly $774 a head out of one month's bill ^[15]. Spend per employee is price times volume, and the price side moved first.

Ramp puts the effective cost of a million tokens at $0.68, which is 41 percent below the March 2026 peak ^[5]. That implies a peak near $1.15, so about 47 cents came off ^[16]. The report does not break out how much of that decline landed in August itself, and both OpenAI and Anthropic have announced further cuts recently ^[20].

For a seller, a price at 59 percent of its March level needs volume up about 69 percent to bring in the same revenue ^[14]. Ramp says usage volume is growing, and Kharazian says it may not grow fast enough to offset the price decline ^[7]. Most of that growth is coming from the cheaper standard tiers, GPT-5.6 Terra and Claude's Sonnet series ^[21].

The mix moved with it. The 8 points frontier models gave up between the start of August and early September is about 15 percent of where that tier began the month ^[17]. Adoption itself kept climbing: 43.8 percent of US companies paid for Anthropic services in August, up 0.34 percentage points ^[3], while OpenAI reached 39.8 percent on a 0.09-point gain ^[4]. Anthropic added penetration at roughly 3.8 times OpenAI's rate ^[24].

The reasons to hold off on calling this retrenchment are in Ramp's own note on method. The top 1 percent is a small pool of firms, so the estimate is more volatile than other segments and may be revised later ^[2]. Ramp also says its data covers only a slice of the overall market ^[19]. Because it measures usage through routing platforms that also sell closed models, Ramp says real open-weight adoption is probably lower than the count ^[12]. Its own denominators imply AI users are about 56 percent of companies on the platform ^[18].

I'd expect September to recover part of the August fall while the blended price keeps dropping. That would make the per-employee line a measure of discounting by two vendors competing on tier, not a cut in what buyers are willing to spend. The case against that is Kharazian's own from last month. He flagged what he called "Cracks in the AI Thesis" in his report and pointed to weak Fable 5 adoption and the shift toward cheaper models as warning signs for providers ^[10]. If the frontier share keeps sliding and volume growth does not accelerate, September prints lower again.