On May 26, The Verge ran a one-line headline: “Uber's president says AI spending is ‘getting harder to justify.'” It's a story whose implications are too heavy to wave off as a routine executive remark.
Uber spent $3.4 billion on research and development in 2025—a 9 percent increase over the prior year. Then, just four months into 2026, reports surfaced that the company had already burned through its entire annual AI budget. Inside Uber, a question began to grow louder: “Is the money we're spending actually creating value?”
The answer came from Andrew Macdonald, Uber's president and chief operating officer, in an interview—and it cuts to the heart of the matter. “That linkage doesn't exist yet. Implicitly, more may be shipping, but it's very hard to draw a line from a single metric, like token usage, to ‘we're now actually building 25 percent more useful consumer features.'”
There's a reason this remark matters. It signals that the second round of AI adoption—the “prove it works” phase—has begun.
Does More Token Consumption Mean More Value?
To restate Macdonald's diagnosis: token usage for AI tools like Claude Code is climbing astronomically. Yet no direct link is visible between that surging usage and any rise in the value users actually receive.
This isn't just one company's worry. It's a stage nearly every company that has seriously adopted AI tools eventually reaches. In the early days, the very fact of “using AI” looked like an achievement. Lines of code generated automatically, reports compiled in minutes, several design drafts produced in a single pass—that alone felt like enough of a win.
Now it's different. A year in, the invoices have come into focus. Looking at those invoices, executives start asking: “How much did this spending grow our revenue? How much happier are our users? How much better is our product?”
Surprisingly few companies can give a clear answer—even a data-savvy giant like Uber.
The Structural Dilemma Uber Faces
Another striking part of Macdonald's comments concerns something Uber CEO Dara Khosrowshahi said earlier this month: “We're offsetting our increased AI investment with reduced hiring.”
The implication is clear. Uber is cutting labor costs and spending that money on AI tools. It's hiring fewer people and buying more tokens instead.
For that trade to make sense, one condition has to hold: AI has to genuinely do the work people used to do, and that has to show up in a measurable way—in revenue or in user experience.
Macdonald's own words show the weight of that condition: “You start to have to talk about token consumption and the associated cost versus headcount. If you can't directly draw a line to how much useful feature and functionality you're shipping to users, that trade becomes hard to justify.”
A decision to hire fewer people is hard to reverse. Once you've shrunk your staffing and then conclude that AI didn't deliver as hoped, operations start to wobble. Some big tech firms have already lived through this—laying people off in the name of AI, only to have to restart hiring.
The Measurement Trap
Why is it so hard to measure the ROI of AI investment? There are several structural reasons.
First, efficiency gains at the front line don't carry through to the final outcome. Say a single developer codes 30 percent faster. That doesn't translate into the company shipping new products 30 percent faster. In between lie planning, decision-making, review, deployment, marketing, and sales—and each is a bottleneck. Even if coding speeds up, if the other stages stay the same, the final output barely changes.
Second, acceleration casts a shadow. When AI generates code faster, there's that much more code to review. When AI produces more content, there's that much more content to vet. People's work doesn't shrink; it changes shape. And that new work tends to go unmeasured.
Third, the baseline for comparison disappears. It's hard to answer “How long would this have taken without AI?” because you can't compare before and after under identical conditions. The market changes, the product changes, the team's makeup changes.
Fourth, productivity gains aren't permanent. An AI tool's usefulness is greatest right after adoption, and its marginal utility shrinks over time. Many teams see striking changes in the first month, only to find the cost-benefit picture growing murkier six months later.
Together, these factors have made AI investment one of the trickiest areas of all to measure for ROI.
What “Harder to Justify” Means
Macdonald's phrasing is telling. He said the spending was “getting harder to justify”—not that it “can't be justified.” That nuance matters.
This isn't a decision to halt AI investment. It means the accountability around that investment is tightening. Last year, a single pressure—“fall behind if you don't invest in AI”—was enough to get budgets approved. Starting this year, you have to answer the question “So what are the results?” Fail to answer, and budgets get cut, usage gets restricted, or someone has to take the blame.
What this signals is that the sales round of AI adoption is over. We've moved from the stage of deciding whether to adopt to the stage of justifying what was adopted. In this stage, what's needed isn't a flashy demo but clear numbers.
Where Small Companies Hold the Edge
This shift is a burden for large companies, but for small ones and solo operators it can actually create an opening.
Large companies decide on AI adoption at the enterprise level. The budgets are big, the use cases sprawling, the ROI measurement complex. At Uber's scale, tracing the causal link between token usage and user value is itself a massive project.
Small companies are different. A single operator pays for a Claude Code subscription, builds some feature with it, and can track what revenue that feature generates—all inside one person's head. Because the unit of measurement is small, the causality is clear.
“I used this tool for a month—did revenue go up or not?” A small company can answer that within a month. A large company often can't answer the same question even on a quarterly or annual basis.
This asymmetry is an unexpected strength for small companies. The harder ROI measurement becomes in the AI era, the more the advantage tilts toward organizations that can experiment in small units and quickly verify the results.
A Practical Way to Measure ROI
For anyone trying to measure the ROI of AI investment in their own business or team, here's a practical approach.
1. Keep the unit of measurement small. Measure not “the effect of all AI investment” but “the effect of adopting AI in this one workflow.” One customer-service automation, one type of content generation, one step in code review. The narrower the unit, the more clearly the effect shows.
2. Don't count cost as direct cost alone. Beyond the subscription fee, include learning time, system-integration costs, the extra time spent on review, and recovery costs when something goes wrong. Add all of that up to get the true cost.
3. Measure value as changes in user behavior. A “30 percent gain in internal efficiency” has no value on its own. You have to see whether it led users to get answers faster, return more often, or buy more.
4. Reassess every three months. An AI tool's usefulness shifts over time. A tool that looked great at first may deliver less value for the cost six months later. Rather than adopting once and using it forever, build the habit of periodically re-examining its worth.
5. Deliberately ask, “What if we did this without AI?” Throwing AI at every task isn't the answer. Some tasks are faster and more accurate done by hand. Consciously distinguishing where to use AI from where not to is the most direct way to raise ROI.
The Second Round Begins
Here's the current landscape of the AI market.
Big tech is pouring enormous capital into infrastructure. Nvidia alone spent $40 billion on AI equity investments this year, and OpenAI and Anthropic have teamed up with private equity to build consulting organizations worth billions of dollars. The side building the tools is still aggressive.
The side using the tools is another story. Even a giant like Uber has now begun asking, “Does this tool really give us value?” More precisely, the adoption sales pitch is over and the justification phase has begun.
The most interesting place is where these two currents collide. The tool companies push “use more,” while the customer companies doubt whether they're getting value for what they've used. This tension will shape the pricing, packaging, and differentiation strategies of the AI market over the next several quarters.
The first round of AI adoption ran on excitement. The fear of “fall behind if you don't use it” was the most powerful marketing there was. That round is now winding down.
The second round is verification. “How did it actually go?” becomes the central question. The tools that survive this round are the ones that can prove clear value, and the companies that survive are the ones that can measure that value inside their own business.
Uber's AI budget, gone in four months, is a product of the first round. And Macdonald's “hard to justify” remark is the opening signal of the second.
It's a moment to check which of these two rounds your own business and team are standing in. Are you still swept up in the excitement, or have you moved calmly into verification? Whether you can answer that honestly will determine your competitiveness in the next phase of the AI era.
The more powerful the tools become, the more it matters to have the judgment to decide where to point them. The question Uber is asking now is, in the end, the question every business will soon confront.
