Uber Burned Through Its Whole-Year AI Budget in Four Months. Here Is What It Did Next.

Then the bills started arriving.

On June 2, 2026, Bloomberg reported that Uber had set a hard limit on how much employees can spend on its AI coding tools. The number is $1,500 per month, per person, per tool. The cap covers agentic coding tools like Cursor and Anthropic's Claude Code, and it applies to each tool separately. An engineer who wants to go past it needs manager approval.

The news followed an earlier report. Uber's CTO, Praveen Neppalli Naga, told The Information in April that the company had already used its full-year 2026 AI budget in roughly four months. Usage had climbed fast: Uber's CEO has said about 10% of the company's committed code is now built by AI agents.

This is the moment the bill came due. And it is a useful one for any leader who feels behind, because the question Uber is now asking is the same question you can ask this week, no matter where your team sits on the climb.

The pendulum is swinging from "adopt" to "prove it"

The spending itself is not slowing down. Gartner forecasts worldwide AI spending will reach $2.59 trillion in 2026, a 47% jump over the prior year. Much of that total is infrastructure bought by technology providers and hyperscalers, not direct spending by ordinary business teams. Even so, the firm's distinguished VP analyst, John-David Lovelock, described 2026 as the year enterprises, not just the big cloud providers, significantly expand what they put into AI.

So money is pouring in. The harder question is what comes back out.

Here the numbers turn sober. The MIT NANDA report "State of AI in Business 2025," also called The GenAI Divide, studied 300 publicly disclosed AI implementations, interviews with 52 organizations, and survey responses from 153 senior leaders. It found that roughly 95% of the organizations it examined had yet to see a measurable financial return on generative AI, while about 5% of integrated pilots were producing meaningful business outcomes. The report put the enterprise spending behind that finding at $30 billion to $40 billion.

Read those two findings together. Gartner expects worldwide AI spending to jump 47% this year, while the organizations MIT studied mostly cannot yet point to a return. The two measure different things, but they describe the same tension every leader is now holding: the money is moving faster than the proof.

Uber's response, a cap plus a dashboard, is one answer to the cost problem, and a good one. What it does not settle, on its own, is whether the spending is buying enough value to justify it. That question sits one level deeper.

Why a spending cap alone does not fix the problem

A cap answers how to stop the spend from running away. It does not answer whether the spend was buying enough to justify the cost. Uber's own COO, Andrew Macdonald, named that problem to Fortune: the link between rising AI usage and the features actually shipped to users, he said, "is not there yet."

MIT NANDA is direct about where the real barrier lives. In the report's words:

The core barrier to scaling is not infrastructure, regulation, or talent. It is learning.

Read that again, because it cuts against the usual story. The thing holding companies back is not the model, the budget, or the headcount. MIT means "learning" partly in a technical sense: most GenAI systems do not retain feedback, adapt to context, or improve over time. Our broader read is that organizations have to learn too, folding AI into how the work really gets done and then measuring the result.

That is our read on what separates the 5% from the 95%, and it is worth stating as a read rather than a finding: the teams seeing returns did not get there by spending the most tokens or by capping the hardest. They got there by putting AI inside a defined process and measuring it against a known result. When the workflow is clear, the spend has somewhere to go and something to be measured against. When the workflow is vague, more spend just buys more activity.

So the cap is a fine guardrail. The deeper work is designing the process the AI is supposed to serve.

The question to ask before you set a single dollar limit

Here is the practical part. Before you decide between an open AI budget and a per-person cap, ask one thing about every place your team is using AI:

What specific work is this AI doing, and how do we know it paid off?

If you cannot answer that, a cap will save money without creating value. If you can answer it, the spending decision gets easy, because you are now measuring a result, not a token count.

This is where The 7 Levels of AI Proficiency gives you a way to think about your team. At the early levels, people are experimenting with tools and learning to prompt well. That experimentation is good and necessary. Further up the climb, the work changes. A leader operating at the design level is not asking which tool everyone should standardize on. They are designing the workflow first, mapping the steps a real process takes, and only then deciding where an AI agent fits and what success looks like. The tool comes after the design, not before it.

That order is the difference between rapidly rising spend with an unclear return and a pilot that ends up in the 5% MIT describes.

What to do this week

You do not need a Gartner-sized budget to apply the lesson. Three steps fit any team.

1. Find out where the spend is going

Uber's engineers track their own usage through an internal dashboard that shows consumption across tools. You can do a simpler version. List every AI tool your team pays for and roughly what each person spends. Putting it in one place is the first decision, and it is often the first time anyone has seen the full picture.

2. Tie each use to a specific workflow

Pick the top two or three places AI is being used. For each one, write down the actual process it supports and what a good outcome looks like. If a use has no defined workflow behind it, that is your finding. Define the process before you spend more on the tool.

3. Set limits with an approval path, not a hard wall

The Uber model is worth copying in shape. Per-person and per-tool limits, a dashboard people can see, and a path to exceed the limit with approval when the work justifies it. As an Uber spokesperson put it, the goal is "to responsibly encourage agentic AI adoption and experimentation at scale across the company." A cap with an approval door encourages the right spending while stopping the runaway kind.

The takeaway

Uber has been one of the most aggressive adopters of AI coding tools, and it still blew through a year of budget in four months. This is what happens when adoption runs ahead of design.

The leaders who come out ahead in 2026 will not be the ones who spent the most or capped the hardest. They will be the ones who can answer, for every dollar, what work the AI is doing and how they know it paid off.

Start with one workflow this week. Map it, then decide where the AI belongs. The spending question tends to answer itself once the work is clear.

Related reading: Level 5: Captain.

Sources

1. Uber caps employee AI spending after blowing through budget in four months
2. Uber caps monthly employee AI spending
3. Uber burned through its entire 2026 AI budget in four months. Now its COO is questioning whether it's worth it (Fortune)
4. Gartner Forecasts Worldwide AI Spending to Grow 47 Percent in 2026
5. Global AI spend set to climb in 2026, Gartner says
6. State of AI in Business 2025 / The GenAI Divide (MIT NANDA)
7. MIT report: 95% of enterprises see no return on generative AI

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