Uber Burned Its AI Budget in Four Months. Marketing Is Next.

Elad Gil, CEO of the investment firm Gil & Co, told the No Priors podcast on Thursday that the open-ended phase of corporate AI spending is closing, and he named the question that replaces it:

“if you have a certain token budget who do you give it to and why”

That’s at 23:04, in a conversation about why AI labs have slowed researcher hiring. His explanation there was that “the cost isn’t the researcher, it’s the compute associated with the person.” A few seconds later he described what comes next inside companies generally: “what are the projects and people that should actually get outsized pieces of a token budget and what is that return on investment?”

He was talking about labs and engineering teams. Marketing should read it anyway.

Most AI tools are now billed by the token, meaning you pay for how much text the software reads and writes rather than a flat monthly license. That turns AI from a fixed cost into a meter, and a meter invites the question of whose work is worth feeding. Every function will get asked. Not every function has an answer ready.

Uber found out how fast the meter moves. The company burned through its entire 2026 AI coding tools budget in four months, after encouraging adoption through an internal leaderboard that ranked teams by total AI tool usage. President and COO Andrew Macdonald, in remarks reported in May, said the spending had outrun the evidence:

“If you’re not actually able to draw a direct line to how [many] useful features and functionality you’re shipping to your users, that trade becomes harder to justify.”

Two things about that sentence should worry a marketing team more than an engineering one. Uber’s CEO, Dara Khosrowshahi, had already told analysts on an earnings call that “we’re seeing uptake of these tools, whether it’s our legal team or marketing team or developers,” and that it was “creating employees with superpowers.” Marketing got named as a beneficiary before anyone asked it for a number.

Then look at what happened at the next company down the line. When Accenture moved to control its spend in June, the restriction didn’t land on engineers. Staff were told to stop using AI for basic tasks, and the firm’s internal data pointed at non-engineers running things like PDF-to-slide conversions through large language models. Justice Kwak, Accenture’s agentic AI strategy lead, put it plainly in remarks captured on leaked audio from an internal meeting:

“We’re hitting this inflection point where AI is becoming material to the cost structure.”

Kwak also said leadership at the CFO, COO and CIO level was “still asking the question of whether they’re getting value from what we’re spending on in the context of AI.” Amazon deleted a developer leaderboard that had gamified consumption, a practice the coverage has taken to calling tokenmaxxing. The incentive to burn tokens is being removed at exactly the moment someone starts auditing who burned them.

Marketing’s own numbers make it an easy target. We covered the readiness gap in Gartner’s spend survey when it landed in July: 15.3% of a flat marketing budget already pointed at AI, against 30% of CMOs who said their organization could scale it. That was a story about capability. Under a meter it becomes a story about allocation, because the spend now arrives monthly and variable instead of annually and fixed.

Some marketing leaders already carry the number Gil is describing. Patrick Shea, SVP of Global Marketing at BlueVoyant, a $175M cybersecurity company, described his method on The Dave Gerhardt Show on Thursday:

“I took all of the marketing money that we spent in the last year, and I divided it by the number of new business opportunities that the company created.”

Shea uses that figure to negotiate forward rather than to report backward. His pitch to finance runs: if the cost per opportunity is $5,000 today and next year’s budget brings it to $2,500, the contribution target gets hit. His other rule is to break the budget into pieces small enough to defend separately, because “each one of these buckets behaves a little bit differently.”

That’s the shape of a defensible AI line. Not a tools list, and not an adoption rate. A denominator the CFO already recognizes, with AI spend sitting inside it as an input rather than beside it as an initiative.

The mechanism working against marketing is billing, not politics. Metered pricing makes monthly spend hard to forecast, and that unpredictability is what pulled finance into the conversation at Accenture. Under a seat license, a marketing team’s AI cost was a rounding error approved once a year. Under a meter, it’s a variable in a monthly review, sitting next to media spend and compared against functions that publish output metrics. Engineering can point at shipped features and argue about whether the link holds. That’s what Macdonald was doing. A team that can only point at content volume is arguing that it used the tool a lot.

The pricing has now moved against marketing twice this summer. When OpenAI cut per-token prices, the cheaper unit didn’t produce a cheaper bill, because the tools started doing more work per request. Now the same metering is producing an allocation review, which is where our reporting on rising compute costs inside marketing budgets was heading.

Gil isn’t predicting cuts. He’s predicting a ranking, and rankings reward whoever shows up with a number. Accenture’s went after the function whose AI use looked like converting PDFs into slides, which is an unflattering but recognizable description of a lot of marketing AI work.

So the work is to have Shea’s number before the meeting, not after it. Take last year’s marketing spend, divide it by opportunities created, and then run the same division with the AI line included and excluded. If those two numbers are indistinguishable, the AI budget is not yet an investment, and someone in finance will notice before the next renewal.