Top AI spenders cut per-employee costs by nearly 10 percent in August

What happened

New data from Ramp's AI Index shows that the largest AI spenders among US companies reduced their per-employee AI costs by nearly 10% in August, even as their overall AI usage continued to grow. Ramp, the corporate spend management platform, attributes the shift to falling token prices, which have dropped 41% since March, according to reporting from The Decoder.

The top 1% of US firms by AI spend — those investing most heavily in generative AI tools and infrastructure — are the cohort driving this efficiency gain. Rather than spending less on AI outright, these organisations appear to be extracting more capability per dollar as the underlying cost of model inference falls.

Why it matters

The story signals an inflection point in enterprise AI economics: unit costs for AI usage are now falling meaningfully even as adoption deepens. For leaders weighing AI investment, this reframes the calculus from "can we afford to scale AI" to "how do we capture the efficiency dividend as prices keep declining." Organisations that were early, heavy adopters are the first to see this benefit, suggesting that the cost curve rewards depth of usage and integration rather than caution.

For technology and operations leaders, falling token prices also lower the barrier to embedding AI more deeply into workflows — customer service, back-office processing, decision support — without a proportional rise in run-rate cost. That changes the business case for expanding AI-driven service and automation initiatives that might previously have been shelved on cost grounds.

By the numbers

• Nearly 10% reduction in per-employee AI costs among the top 1% of US firms by AI spend in August
• 41% decline in token prices since March, per Ramp's AI Index

The Renascence take

The headline number here is cost, but the more interesting signal is behavioural: the firms already spending the most on AI are the ones benefiting most from falling prices, not the cautious latecomers waiting for costs to drop before they commit. That is the opposite of how many organisations are currently sequencing their AI investment decisions.

Most leaders treat AI cost as a reason to wait. The data suggests the opposite logic applies: it's the heavy, committed spenders who are positioned to ride the price curve down, because they've already built the integrations, workflows and internal fluency that let falling token costs translate directly into margin. A customer-obsessed operator shouldn't be asking whether AI has got cheap enough to justify a pilot — they should already be running at scale, so that every drop in inference cost falls straight to the bottom line rather than being absorbed by the friction of starting from zero.