Uber Technologies Inc. set aside a substantial sum for artificial intelligence initiatives this year. Then its engineers got to work. In just four months the company exhausted the full amount earmarked for 2026. One two-hour coding session ran up a bill of $1,200. The episode has forced executives to rethink hiring plans, impose spending caps and question whether the productivity gains justify the accelerating costs.
The details surfaced first in reports from Yahoo Finance. Engineers at the San Francisco company embraced tools such as Anthropic’s Claude Code after the firm rolled them out in December 2025. Adoption jumped from 32 percent in February to 84 percent in March. By spring roughly 95 percent of Uber’s engineers used the systems each month. About 70 percent of the code they committed originated from these AI assistants.
Costs mounted quickly. Average monthly spend per engineer ranged from $150 to $250. Heavy users racked up $500 to $2,000. Internal leaderboards ranked teams by AI-tool consumption. The incentive worked too well. Uber burned through its entire allocation by April. The revelation prompted immediate changes. The company introduced a $1,500 monthly cap per employee and per agentic coding tool. Exceptions require approval. A dashboard now tracks every user’s consumption in real time.
Chief Executive Officer Dara Khosrowshahi addressed the overrun directly. In an interview with investor Patrick O’Shaughnessy he said, “We blew through our AI budget in a quarter, for the whole year essentially.” The remark, reported by Yahoo Finance, came alongside news that the surge forced adjustments to hiring goals. Khosrowshahi noted that 11 percent of the company’s live backend code now comes from AI agents. On a separate earnings call he added that about 10 percent of committed code is built by autonomous agents.
But the return on that investment remains hazy. President and Chief Operating Officer Andrew Macdonald voiced skepticism. He told colleagues it was becoming harder to justify rising token costs without clear evidence the tools produced more useful features for customers. The comment, first published by Fortune, captured a growing tension inside the company. Engineers loved the speed. Finance teams worried about the bills. Even the finance department itself began to rely on AI tools, a shift noted in CFO Dive.
The pattern is not unique to Uber. Other technology firms have encountered similar shocks. Tesla reportedly limited AI-tool spending to $200 per employee per week after observing runaway consumption. Bloomberg detailed Uber’s new caps in a June report that TechCrunch later summarized. The article highlighted how quickly internal dashboards revealed the problem once leaderboards encouraged maximum usage.
Forbes examined the root causes in greater depth. Writer Janakiram MSV explained that token pricing broke the company’s financial assumptions. Uber had modeled conservative uptake. Instead Claude Code spread across 5,000 engineers faster than anyone projected. The piece, available at Forbes, noted that retries, tool calls and long agent loops drove much of the unexpected spend. Few organizations had instrumented those variables before scaling.
Yet the experiment has produced measurable output. Uber’s research-and-development expense climbed 9 percent to $3.4 billion in 2025 and continues to rise. Some of that money now buys AI infrastructure. Khosrowshahi has repeatedly signaled confidence in the long-term payoff. During a recent CNBC appearance he discussed continued investment in autonomous vehicles alongside AI. The full interview is hosted on YouTube. He described autonomy as “incredibly promising” and said the company would keep buying back stock while funding these areas.
Analysts remain divided. Some see the early overspend as proof that generative coding tools deliver real velocity. Others warn that without tighter measurement the expense could erode margins at a time when Uber is finally profitable. The company’s stock has declined 24 percent over the past year. Options markets show moderate bullish sentiment despite the AI headlines.
Recent conversations on X reflect the same debate. One post from @tldrmarket noted the budget exhaustion and new limits while highlighting the stock’s performance. Another from @BSE_Files contrasted Uber’s $1,500 monthly cap with Tesla’s tighter weekly ceiling. The common theme: companies once eager to push adoption now scramble to control costs. A thread by @its_mohamedomar asked why tokens underperform expectations and pointed to missing instrumentation around agent behavior.
Uber’s experience offers a cautionary data point for any enterprise racing to embed large language models. Rapid rollout yields code output. It also generates bills that arrive faster than productivity metrics. The firm responded with caps, dashboards and frank executive commentary. Whether those steps restore balance will determine if the 2026 overrun becomes a footnote or a template for the industry.
Praveen Neppalli Naga, Uber’s chief technology officer, first disclosed the four-month burn rate during an internal briefing covered by The Information. That disclosure set off the chain of public reports. Macdonald’s later remarks to staff, relayed through Fortune, added a layer of operational realism. The two perspectives—enthusiastic engineering adoption and cautious financial oversight—now coexist inside the same organization.
So far the company has avoided major layoffs tied to the AI push. Instead it has recalibrated hiring. Khosrowshahi described the budget exhaustion as forcing a reassessment rather than a retreat. The distinction matters. Uber still plans to expand its autonomous-vehicle partnerships. It still intends to grow AI-driven features across rides, delivery and freight. The question is how to do so without repeating the early-year surprise.
Industry observers will watch the next earnings release closely. If management can link incremental AI spend to concrete gains in feature velocity or driver earnings, the current unease may fade. If not, further restraints could follow. For now the message from Uber’s leadership is clear. The tools work. The economics require refinement. And the days of unconstrained experimentation have ended.
WebProNews is an iEntry Publication