Uber burned through its entire 2026 AI budget by April. The ride-hailing giant’s engineers had gone all in on tools like Anthropic’s Claude. Costs exploded. Productivity gains? Not so much.
That story, shared by Uber CTO Praveen Neppalli Naga, sent a jolt through corporate tech departments. Yahoo Finance reported Naga’s blunt assessment: the company quadrupled the number of employees using frontier AI tools. Yet expenses per token dropped. Prompt caching, smarter default models, rigorous evaluations, and dashboards showing usage per hour did the trick. “We’re coming to the end of the so-called tokenmaxxing era,” Naga wrote on X. Costs fell even as adoption grew. Efficiency became an engineering problem, not just a budget one.
But not every company caught on so fast. Global AI spending is projected to hit $2.5 trillion this year. That’s a 44% jump from the year before. Fortune laid out the shift. After years of hackathons, training sessions, and open access to ChatGPT, Claude, Gemini, and coding agents like Cursor, executives now draw lines. They cap usage. They push smaller models. They retrain staff. Budgets blew past forecasts in some cases. Value did not follow.
Stephen Franchetti, CIO at Samsara, authorized multiple AI tools but built a daily expense tracker. Non-technical staff face caps. Research and development teams get more leeway for heavy coding and analysis. “It took us a while to settle on the right caps, to make sure everyone was well served,” Franchetti told Fortune. “But it puts people in the position where they’re kind of in control and they can make choices as to which models they use.”
Short. Sharp. Necessary.
Sagnik Nandy, CTO at Docusign, saw 75% of code initiated by AI. Impressive on paper. The catch? Agents pulled the company’s entire codebase for context. Token counts soared. Nandy changed the default to pull only relevant context for narrow tasks. Token usage fell nearly 50%. “We’ve taken this very seriously, both in the internal use case and we are propagating those learnings externally,” he said.
At Yum Brands, which runs KFC and Taco Bell, Jim Dausch watched token usage and costs climb earlier this year. Not material yet. The trajectory worried him. He estimates 95% of tasks can run on cheaper, basic models. So Yum trains employees on model selection and treats AI spending like headcount budgets. “We’re trying to kind of democratize where the costs live and how they’re managed, so it isn’t just an IT line item,” Dausch explained in the Fortune piece.
Cigna takes a different tack. Its chief data, digital, and AI officer Katya Andresen authorized more than 70 models. The healthcare company mixes small language models and older versions for routine work. Premium models stay for complex reasoning. “The way you really run up costs is you use the most expensive models with no guardrails around them,” Andresen said. Token usage rose. Total spend grew more slowly thanks to this approach.
Shay Artzi, CTO at Compass, focused AI on engineers first, then an internal assistant for real estate agents, and finally the broader workforce. He avoided mandates that all code must use AI. Instead, pilots led to partnerships with Anthropic and Google, plus per-engineer budgets. “We also put budgets for every engineer, so they are aware of how they’re spending,” Artzi noted.
The Control Gap Widens
These moves come as visibility lags badly. An IBM study from June found AI spend expected to rise from under 15% of IT budgets in 2025 to nearly 25% by 2027. A 71% jump in two years. Yet 84% of tech leaders have not fully operationalized AI financial management. And 85% lack real-time visibility into spend. Two-thirds of CIOs and CTOs feel accountable for AI systems they do not fully control. Business teams deploy faster than IT can track. IBM Newsroom quoted IBM CIO Matt Lyteson: “For CIOs and CTOs, the challenge now is scaling AI systems that operate continuously and autonomously, often within governance models and architectures designed for a far slower, more predictable environment. It is no longer just about deploying AI faster. It’s redesigning how organizations control, govern and invest in it and embedding control and visibility from the start, so they can scale with confidence.”
Organizations that get discipline right deploy 2.4 times more AI agents without increasing budgets. They spend four times less of their AI allocation. They deliver 18% higher operating margins. They stand three times more likely to feel ready for scale. The gap between ambition and readiness shows elsewhere too. Gartner forecasts end-user spending on AI models and platforms will reach $64 billion in 2026, up 63%. Agentic AI drives much of that. CIO Dive highlighted analyst Arunasree Cheparthi’s advice: move toward outcome-based pricing, add contract protections like hard limits and token rollovers, track cost per token, and use multivendor routing for routine versus complex tasks.
Will Sommer, a Gartner expert, put it plainly in Fortune. “2026 is the year of everyone finding out that AI is actually really hard. It’s not a free lunch. It requires a lot of thought and effort to get right.” His team warned that AI coding costs could exceed the average developer’s salary by 2028. Token consumption and usage-based pricing accelerate the trend. Companies spend thousands per employee on tools that produce junk.
But. Efficiency gains appear when leaders treat costs as a technical challenge. Uber proved it. Docusign cut tokens in half. Cigna spread usage across dozens of models. Samsara gives teams choice within caps. The hype phase ends. Measurement begins.
Recent reports reinforce the pressure. A July analysis on PointFive noted worldwide AI spending near $2.59 trillion, up 47%, with real-time visibility at just 26%. Token consumption forecasts show 24-fold growth by 2030. Hyperscalers pour hundreds of billions into infrastructure. Boards demand returns. CIOs face the heat.
So companies ration. They train. They route queries intelligently. They build dashboards. They negotiate better contracts. The all-you-can-eat era fades. Precision takes its place. Not every experiment delivers. Many do not. Leaders who embed visibility early win bigger margins and faster deployment. Those who don’t watch budgets vanish and questions mount.
The message spreads. From Uber’s April reckoning to Samsara’s daily monitors to IBM’s call for redesigned governance. Tech executives once raced to adopt. Now they race to control. The bills arrived. Adjustments follow. Results will decide who thrives.