The U.S. Is Winning the AI Adoption Race. The Receipts Are Starting to Show.

American companies are out front on AI. They have the compute, the spending, and the workforce already reaching for the tools. What the last few weeks have made clear is that being out front carries its own price, and some of the country's most advanced adopters are now learning what that price looks like in a budget meeting.

The lead is real

If you have felt behind on AI, this is a useful moment to watch. The companies setting the pace are still working out the economics in real time. There is room to learn from what they are discovering before you spend a dollar of your own.

The U.S. genuinely sits ahead of other advanced economies on AI. The Federal Reserve, studying competition across advanced economies, found the country holds more than four times the compute performance capacity of its closest G7 peer. That edge narrows once you adjust for the size of the economy or the population, but the raw capacity advantage is there.

Adoption is climbing too. A separate Federal Reserve analysis found work-related generative AI use among U.S. workers reached about 41% by November 2025, up from 33% in August 2024. A large share of American workers are now using these tools in their jobs.

So the picture looks like a country accelerating. The wrinkle shows up when you put the worker number next to the company number.

More than four times the compute performance capacity of its closest G7 peer.

Source: Federal Reserve, 2025

The split between workers and companies

While about 41% of U.S. workers report using generative AI for work, about 18% of U.S. firms had adopted AI by year-end 2025, with the older Census question, focused more narrowly on producing goods and services, showing lower adoption before the survey wording changed. Worker uptake is still running ahead of company-level adoption.

One caveat belongs with those numbers. The exact figures depend heavily on the survey. Worker surveys, firm-weighted business surveys, and employment-weighted executive surveys are each measuring something different, so read the direction of the split more than the precise points.

That delta is worth sitting with. It means a large share of AI use is happening without a company policy behind it. People are bringing tools into their work on their own initiative, often faster than their organizations can set rules, track cost, or decide what good use even looks like. That is exactly the surface where uncontrolled spending and quiet risk tend to collect.

What the spending curve actually looks like

The money side is moving fast. Gartner forecasts worldwide AI spending will total $2.59 trillion in 2026, a 47% increase over the prior year, with AI infrastructure making up more than 45% of that total. Gartner calls 2026 the enterprise "inflection year."

The usage underneath that spending is climbing even faster. Goldman Sachs Research projects AI token consumption will multiply roughly 24 times to about 120 quadrillion tokens per month by 2030, with agentic AI doing most of the lifting. Enterprise agents, the kind that run multi-step work on their own, are the single largest multiplier of token use.

Tokens are one of the main billing units behind many AI tools and APIs. When consumption grows that quickly, the bill can grow with it unless falling token costs, vendor pricing, usage limits, or workflow design offset the increase. A flat per-seat license stops describing what a company actually owes the moment its people start running agents at full speed.

What happened at Uber

Uber gives the clearest example so far. The company exhausted its entire 2026 AI coding budget roughly four months into the year, driven largely by Claude Code. According to reporting on the rollout, engineer use of the tool climbed to about 84%, a sharp jump over roughly a month. Adoption raced ahead of the plan, and the spend went with it.

Even Uber is still working out whether the spend pays off. COO Andrew Macdonald said the company could not yet clearly connect rising Claude Code usage to more useful consumer-facing features, telling an audience that the link "is not there yet." The reported response was to limit employees to $1,500 in monthly token spending per AI coding tool.

The lesson here has little to do with the tool being too expensive. Plenty of teams would gladly pay that for an engineer who ships faster. The real story is that the cost stayed invisible until it had already blown through a year's plan. No usage limit was set going in, so the spending curve went unseen until it arrived as a surprise.

The cost stayed invisible until it had already blown through a year's plan.

The real lesson is governance

It would be easy to read all of this as a reason to slow down. That is the wrong takeaway. The companies in these stories are not failing at AI. They are running into the part of adoption that comes after the excitement: deciding how to pay for it, measure it, and keep it pointed at real outcomes.

This is where The 7 Levels of AI Proficiency is useful as a way to think about your own progress. Early on, proficiency looks like using a tool well. Further up, it looks like designing the system the tool lives inside, with usage limits, cost tracking, and a clear read on what each dollar is producing. The front-runners are learning that the higher levels are about the system more than the speed.

If you are building your own AI practice now, you get to start where Uber ended up. Set a usage limit before you need one. Track cost per outcome from day one rather than after a budget review. Write down a simple policy so the people already using these tools are working inside a plan rather than around one.

Related reading: Level 5: Captain, the Design Thinker.

Your next step

Pick one AI tool your team already uses and answer two questions about it this week: what is the spending limit, and what outcome are you tracking it against. If you cannot answer both, you have found the right place to start. Set the limit, name the outcome, and you are already ahead of where some of the biggest adopters were a month ago.

Sources

Frequently Asked Questions

Does the U.S. lead mean American companies are using AI well?

Not necessarily. The Federal Reserve data shows the U.S. leads on compute capacity and worker adoption, but firm-level AI adoption was about 18% at year-end 2025, and lower still under the older Census question focused on producing goods and services. Leading on adoption and using it well are two different things.

Why did Uber's costs run away from them?

Adoption climbed faster than anyone budgeted for. By reporting accounts, engineer use of Claude Code climbed to about 84% over roughly a month, and the token spend climbed with it until the 2026 budget was gone in four months. The company then limited employees to $1,500 in monthly token spending per AI coding tool.

What does "token consumption" have to do with my AI costs?

Tokens are one of the main billing units behind many AI tools and APIs. Goldman Sachs Research projects token use will grow about 24 times by 2030 as more work moves to autonomous agents. The more agentic the work, the faster the meter runs, which is why flat licenses can understate the real cost.

Harrison Painter

Executive AI Advisor. Founder, LaunchReady.ai and AI Law Tracker.

Harrison is an Indiana AI Advisor who helps business owners and executives get their time back by building AI systems that run the work for them. Nearly 20 years in business and author of You Have Already Been Replaced by AI. Creator of The 7 Levels of AI Proficiency.

Connect on LinkedIn

Find your AI Proficiency level

The free 7 Levels assessment places you across seven stages of AI capability. Under ten minutes. Research-backed scoring.

See where you stand Or book a discovery call