Half of US Businesses Now Pay for AI. If You Sell 'We'll Add AI to Your Company,' the Easy Half Is Already Gone.
The Ramp AI Index (built from real card and expense data across tens of thousands of US businesses) shows AI adoption reaching 50.6% in April, after it first crossed the 50% line in March. More than half of the businesses Ramp tracks are now paying for AI of some kind. If your living comes from selling "we'll help you add AI to your company," your instinct is to read that as a rising tide. Read it the other way. The half that was easy to sell has already bought. What's left is the half that couldn't, or wouldn't, do it on their own, and that's a different, harder, more durable business than the one a lot of solo AI consultants are running right now.
What the number actually says
Be careful about what Ramp is measuring, because it changes the conclusion. Ramp tracks paid adoption (money actually leaving a business's account for an AI product), not vague "we're exploring AI" survey sentiment. The same dataset showed Anthropic crossing OpenAI in business adoption, 34.4% to 32.3% in April, the first time more tracked businesses paid for Claude than for ChatGPT. Useful gossip for the model wars, but the number that should reorganize your week is the headline one: 50.6% are already paying for something.
Here's why "paying" matters more than "exploring." A business that's already paying for an AI tool has crossed the hardest threshold there is: the first purchase. They've decided AI is worth money, picked a vendor, and gotten budget approved. The classic entry-level consulting pitch, "let me show you how AI could help your business," lands very differently on a company that already swiped a card for ChatGPT Enterprise last quarter. You're not introducing them to the category anymore. You're competing with the thing they already bought and the in-house champion who bought it.
The easy work is the commoditized work
Look at what that first purchase usually is. For most of the businesses in that 50.6%, "adopting AI" meant buying a seat to a chatbot, turning on a summarizer, or letting their existing SaaS vendor switch on an AI feature they didn't even shop for. That's the work a lot of solo AI consultants have been selling: set up a chatbot, wire a summarizer to their docs, build a basic retrieval thing over a knowledge base. It was good money for a while because the category was new and the buyer was confused.
The buyer is less confused now, and the obvious version of that work is a button inside software they already pay for. When the "just add a chatbot" engagement competes with a feature the client can self-serve, the engagement loses on price every time. The early-adopter half of the market bought the obvious thing, and the obvious thing is exactly what got commoditized. Selling generic "AI enablement" into that environment is selling a thing the market has mostly solved for itself.
Where the remaining demand actually lives
So where's the money in the other half, the businesses that haven't bought? Mostly, they haven't bought because the easy version doesn't work for them. They have messy, contradictory data that a generic tool chokes on. They're in a regulated industry where "just pipe it to an LLM" is a compliance problem, not a feature. They have no internal technical person to own the thing, so a self-serve tool sits unused after the trial. These aren't laggards waiting for a nudge. They're companies where adoption requires actual integration work: sitting inside one business's specific operations and making AI survive contact with their reality.
That work doesn't commoditize, because it isn't a product, it's judgment. A generalist model and a self-serve tool can't reconcile a client's three disagreeing systems, can't absorb their regulatory constraints, can't sit in their operations and figure out what they actually need versus what they asked for. That's the work the remaining half will pay real money for, and it's the work a thoughtful solo operator is positioned to do that a button in ChatGPT can't.
What I'd actually do
Pick a vertical and learn its operations cold. The durable solo AI business in 2026 isn't "I do AI for anyone." It's "I do AI for dental practices / law firms / freight brokers, and I understand their workflow, their data, and their compliance better than a generalist ever will." Depth in one niche is the moat that a 50.6% adoption number can't erode, because the value is in the specificity, not the AI.
Sell integration and outcomes, not enablement. Stop pitching "let me help you adopt AI" to a market that mostly already has. Pitch the specific, gnarly thing the self-serve tools can't do for that vertical (the data cleanup, the legacy integration, the workflow that crosses three systems) and price it as the operational project it is, not as a tech demo.
The honest counter-take
Let me argue against the doom read, because "AI consulting is dead" would be lazy and wrong. The 50.6% number is paid adoption, not depth: plenty of those businesses bought a couple of seats, never rolled it out, and are getting almost nothing from it. There's a real, large second wave of work in helping companies actually use the AI they already pay for, and that wave is barely starting. So the opportunity isn't shrinking so much as moving: from "get them to buy" to "get them to succeed."
But that move is the whole point. The land-grab phase (easy clients, obvious work, low skill floor) is closing, and the consultants who keep running the land-grab playbook into a half-saturated market are going to feel it. It's not that AI consulting is dead. It's that generic AI consulting is dead. Go deep in one vertical, sell the integration the other half can't self-serve, and you're not fighting the 50.6%: you're selling to the part of the market the number can't reach.
Author
Lukas
@lukcombinator