Published · 5h agoInvest2 min read
The 679x gap: enterprise AI spend is concentrating, not spreading
The number is a ratio of increments, not of levels, and what it turns on is which firms can price a unit of work rather than which model they buy.
Context for builders, not their beat.See today for builders
What happened
- Exponential View reports that since October 2023 the top 1% of firms raised AI spend per employee by $6,542, while the median rose only $9.63.
- The top 1% of firms' increase in AI spend per employee since October 2023 is about 679 times the median firm's increase.
- Ramp's August 2026 AI Index reports that in July the top 1% of businesses spent a median $7,400 per employee on AI, the top 10% spent $650, and the median firm spent $11.95.
- On Ramp's July levels, top 1% spend per employee is about 619 times the median firm's.
- If Exponential View's increments end at Ramp's July 2026 levels, the implied October 2023 base is about $858 per employee for the top 1% and about $2.32 for the median firm, a ratio of roughly 370x, which would mean the gap between the two widened about 1.7 times.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
Put the two series side by side and you can back out a starting point. If Exponential View's $6,542 increment ends at Ramp's July level of $7,400 per employee [1][3], the frontier firm was spending roughly $858 in October 2023 and the median roughly $2.32 [5]. On those figures the frontier is about 8.6 times its former self and the median about 5.2 times its own [6], and the multiple between the two widened from roughly 370x to 619x [4][5]. Neither publisher states that the increments end in July, and Exponential View does not name its data provider, so treat the base as an estimate rather than a reported figure.
What makes the increments diverge is the denominator nobody can compute. Exponential View's State of AI work found that a 10% price cut lifts token use by 12 to 18%, enough to raise total spend but not by much [7]. Patrick Saner's comment, quoted in the same letter, is that cost per token is irrelevant next to the cost of completing a useful unit of work, and that firms buy knowledge work in bundles precisely to avoid pricing it; the newsletter's own guess is that the top percentile's spend is mostly software, where commits, pull requests and releases give you something countable [8].
That also caps how much of the 679 reaches the model vendors. Fable 5 costs about twice GPT-5.6 Sol per token, yet took 11.4% of Anthropic dollars on 6% of tokens in its first month [9], so the incremental frontier dollar is not going to the most expensive tokens. Ramp's June letter adds that the advanced cohort has not left the American labs either: 96.4% of model-serving users still paid for OpenAI or Anthropic [15].
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Exponential View reports that since October 2023 the top 1% of firms raised AI spend per employee by $6,542, while the median rose only $9.63.
- [3]
Ramp's August 2026 AI Index reports that in July the top 1% of businesses spent a median $7,400 per employee on AI, the top 10% spent $650, and the median firm spent $11.95.
- [7]
Exponential View's State of AI report found a positive but underwhelming token elasticity: a 10% price cut lifts token use by 12 to 18%, enough to raise total spend but not by much.
- [8]
Exponential View quotes Patrick Saner saying the cost per token is irrelevant and what matters is the cost of completing a useful unit of work, notes that firms buy knowledge work in bundles such as a salary or retainer to avoid pricing it, and guesses the top 1%'s AI spend is mostly software, where AI is most proven and most measurable via commits, pull requests and releases.
- [9]
Ramp reports that one month after release, Anthropic's Fable 5 made up only 6% of tokens businesses purchased from Anthropic and 11.4% of dollars spent on Anthropic models, and that at roughly $10 per 1M tokens it is twice as expensive as GPT-5.6 Sol.
- [10]
Ramp's author writes that with Fable 5 they have found a new upper bound for how much businesses are willing to spend on AI, and that here more performance is not worth the price tag.
Sources & coverage · 2 publishers
The reporting this story was synthesized from, earliest first. Every link goes to the original.
- exponentialview.co5h ago🔮 Why one AI is better than four #598
- ramp.com5h agoAugust 2026 Ramp AI Index: Cracks in the AI thesis
- ramp.com5h agoRamp's latest data on China vs. the American AI Labs
