Stephen Franchetti, chief information officer at Samsara, has introduced usage caps on a suite of artificial-intelligence tools after the company saw its AI spend climb sharply. The decision follows a broader trend among technology leaders who, after years of championing AI adoption, are now tightening controls to protect budgets.
Why limits are being set
At Samsara, Franchetti authorised models from Anthropic, Google, OpenAI and the coding assistant Cursor, while also building an internal dashboard that tracks AI expenses daily. The company recently restricted access for non-technical staff, allowing research and development teams more freedom because their work demands intensive coding and data analysis.
"It took us a while to settle on the right caps, to make sure everyone was well served," Franchetti said. "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."
Global AI spending is projected to reach $2.5 trillion this year, a 44 percent rise on the previous year. Many firms report that their 2026 AI budgets have already exceeded expectations without delivering proportional business value. As a result, senior technology executives are reassessing how AI tools are deployed across their organisations.
Industry voices echo the caution
Will Sommer, a quantitative-modeling expert at research firm Gartner, warned that 2026 will be the year companies discover that AI is not a free lunch. He noted that without careful planning, firms can spend thousands of dollars per employee on tools that produce little useful output.
Gartner's recent report also predicts that by 2028 the cost of AI-driven coding could surpass the average developer's salary, driven by rising token consumption and consumption-based pricing.
At electronic-signature provider Docusign, chief technology officer Sagnik Nandy observed that 75 percent of code now originates from AI assistants. However, he found that the agents were pulling the entire code base for context, inflating token usage. By limiting the context to only what is needed for a specific task, Docusign cut token consumption by almost half.
Yum Brands' chief digital and technology officer Jim Dausch sees a similar pattern. While AI token usage is not yet a material cost for the KFC and Taco Bell operators, he believes that up to 95 percent of routine tasks could be handled by cheaper, less powerful models. Yum therefore trains staff to select appropriate models and treats AI spend like any other departmental budget.
Healthcare insurer Katya Andresen, chief data, digital and AI officer at Cigna, has authorised more than 70 AI models, ranging from small language models to older, cheaper versions. This multimodal approach allows the company to keep overall spend in check even as compute usage rises.
What comes next for European tech leaders
Executives across Europe are now focusing on three levers: establishing clear usage caps, educating employees on model selection, and diversifying the portfolio of AI providers to avoid reliance on a single vendor. Companies such as real-estate brokerage Compass are budgeting AI spend per engineer and setting financial limits on partnerships with providers like Anthropic and Google.
As AI tools become more embedded in everyday workflows, the pressure to balance innovation with fiscal responsibility will likely intensify. Organisations that succeed will be those that embed cost-awareness into their AI governance frameworks while still allowing teams the flexibility to experiment where it adds genuine value.
For European CIOs and CTOs, the challenge now is to translate the early enthusiasm for AI into sustainable, cost-effective practice, ensuring that the technology supports, rather than drains, the bottom line.