LIND RESEARCH
Data Post | Analyst: Kristoffer Lindström
Everyone's Adopting; Almost No One's Using
Adoption is near a plateau at 54%. The median company still spends like it's running a chatbot, not a workforce. This post looks at the gap between AI adoption and AI intensity in Ramp's data and what it says about where demand goes next.
This is a new post format, short and actionable, on interesting technology-related data we find in our research process. As this is the first edition, we share it with everyone. In the future, this will be for premium members, while free readers will have to settle for the summary.
Summary
54% of US businesses now pay for AI (up from 7% in Jan 2023), so on penetration the S-curve is entering its upper reach. Adoption is nearly a solved problem.
But the median firm spends only ~$11 per employee per month, chat-seat money, versus ~$611 for the top 10% and ~$7,449 for the top 1%. That gap is ~53x and ~655x, and it is still widening.
A gap that widens while costs fall is the tell: the binding constraint was never price or access, it was organizational willingness to rewire work. That is a moat forming, not a diffusion lag.
Token prices are falling while spend per employee climbs, so real usage is growing faster than the dollars show. The likely driver of the 2026 pickup is agentic tooling going mainstream (consumption-priced, not per-seat).
AI is still just 2.59% of total business spend. Base case: the median moves toward today's top-10% intensity, pushing AI well past 10% of budget.
Read-through: consensus prices in the compute and memory bottlenecks based on leaders' demand, but the median has barely begun to be consumed. That keeps infrastructure (compute, memory, power) in play.
Adoption is a solved problem. Intensity hasn't started.
We have looked into the fresh data presented by Ramp in their Ramp AI Index. 54% of US businesses on Ramp now pay for AI, up from 7% in January 2023 and ~47% at the start of this year. That is a classic S-curve entering its upper reach. On penetration alone, the story looks close to won, and the marginal new adopter gets less interesting from here.
Then you look at what they actually spend. The median company pays ~$11 per employee per month for AI (from ~$2 in mid-2023). That is chat-seat money. One subscription, a few power users, nothing that resembles AI doing work. The adoption number counts whether a company has a credit card on file. It does not matter whether the technology is load-bearing. For the median firm, it is not.
The gap isn't a lag. It's a moat forming.
The leaders are on another planet. Top-10% firms spend ~$611 per employee per month, and the top 1% ~$7,449, versus the median's $11, per the same Ramp series. That is roughly 53x and ~680x the median. More important than the level is the direction: the ratio is still widening, not compressing.
That is the tell. If this were a simple diffusion lag, the gap should narrow as the technology gets cheaper and easier, with laggards converging on leaders. It is doing the opposite. Costs are falling, tooling is getting simpler, and the leaders are still pulling away. When something gets cheaper and easier to adopt, and the frontier still extends, the binding constraint is never price or access. It was organizational: the willingness to rewire how work gets done. That is a capability the median cannot buy overnight, which is exactly why "adapt before it's too late" is a real warning here and not a slogan. The top 1% are plausibly AI-native firms and development shops that consume the most; the rest of the economy is watching them compound an operating advantage.
Falling unit prices, rising bills. That's the whole point.
Token prices have fallen hard across this window, yet spending per employee keeps climbing. Rising dollars against falling unit costs mean real usage is growing faster than the spend line shows. The $11 median understates the consumption ramp, not overstates it.
The plausible driver of the pickup since the start of the year is agentic tooling going mainstream (Cowork, coding agents, and the shift from a chatbot you talk to toward software that does multi-step work), which is consumption-priced rather than per-seat. We hold that as the likely explanation, not a proven one.
AI is still only 2.59% of total business spend as of April 2026, per Ramp's Spend Share Index. A rounding error next to Software & IT, with almost all the climb still ahead.
Our view: Penetration is nearly a lie. It says the opportunity is maturing when the opposite is true. Adoption is on a plateau; intensity has not begun. The median company will not stay at chat-stage spend, because standing still against competitors whose unit economics have reset is not a survivable position. We monitor spend here, not value. But the T1 group should be seeing value; otherwise, they would not be increasing the spend levels.
We find it likely that over the next several years, the median company will come to look like today's top 10%, which implies many multiples of spend per employee and AI moving well past 10% of the total business budget. The thing stopping this is most people's natural resistance to change, but eventually they have to adopt.
The investable read-through relates to the perception of adoption versus intensity and what it implies for the future. The current build-out is fueled by leaders’ demand, while the entire median of the economy is still at chat-stage intensity and has barely begun to consume. If the median is forced up the intensity curve to survive, the underlying demand base is barely tapped.
That keeps the bottleneck, compute, memory, power, and infrastructure as the play for the foreseeable future. Timing is the risk: Infrastructure bets can be right on the decade and wrong on the quarter. One reason for compression is speed: AI spreads through software updates, not physical investment or retraining. So, its spread might happen in years rather than over a decade like the internet did.
Get Premium Insights
This post was shared with all readers, but the next one will be for premium members. Try our premium offering now for only $1 for 30 days.
Lind Research