CIOs rein in AI spending as costs soar, shift to cheaper models

Stephen Franchetti, CIO at tech firm Samsara, has built a broad AI ecosystem for the company's 4,100 employees, authorizing tools from Anthropic (Claude), Google (Gemini), OpenAI (ChatGPT), and the AI coding agent Cursor. Yet rather than opening these tools to everyone equally, Samsara has capped usage for non-technical employees while allowing more intensive use in groups like research and development that rely on AI for coding and data analysis. Franchetti also developed an internal system to monitor AI expenses on a daily basis. "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."

Samsara is not alone. With global AI spending projected to total $2.5 trillion this year—a 44% increase from prior-year levels—CIOs and C-suite technology leaders are finding themselves navigating a delicate moment. After years of championing AI adoption through training courses, hackathons, and widespread access to AI coding tools and chat assistants, some are now tightening usage and retraining staff on smaller, cheaper AI models that can handle many workplace tasks. Some companies have reported that their 2026 AI budgets overran expectations without producing corresponding business value.

At Docusign, CTO Sagnik Nandy praised his engineers' embrace of AI tools, noting that 75% of the code they develop is initiated by AI. However, he quickly discovered a hidden cost: the AI coding agents were designed to pull Docusign's entire code base for context before executing any task, which created massive token consumption. "That's a lot of tokens, because you're trying to read everything," Nandy explained. After adjusting the agents so they pull only the relevant context for a narrow task by default, token usage dropped by almost 50%. At Yum Brands, Chief Digital and Technology Officer Jim Dausch noted that token usage and expenses were rising earlier this year at the KFC and Taco Bell operator, though he said it is not yet "a material number." Crucially, Dausch found that perhaps as high as 95% of tasks AI tools are asked to perform can be handled by more basic, less expensive models—suggesting much early deployment was overspecified. To address this, Yum is promoting training on AI model usage and advising business leaders to manage digital spending the way they budget for a department's headcount. "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 said.

Healthcare giant Cigna has taken a multimodal approach, authorizing more than 70 different AI models for internal use and encouraging the workforce to use small language models or cheaper earlier versions for tasks that don't require advanced reasoning. Katya Andresen, Cigna's chief data, digital, and AI officer, observed that "the way you really run up costs is you use the most expensive models with no guardrails around them." At Cigna, compute and AI token usage has increased, but total spending is not rising at the same pace because of this deliberate model mix. Real-estate brokerage Compass took a similar disciplined approach, settling on partnerships with Anthropic and Google for AI coding tools but imposing financial limitations and setting individual budgets for every engineer so they remain aware of spending.

Gartner's Will Sommer, a quantitative modeling and economic forecasting expert, offered a blunt assessment: "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." Sommer warned that companies can easily spend thousands of dollars per head on AI tools whose output is essentially junk and does not improve productivity. In June, Gartner issued a bearish report projecting that AI coding costs would overtake the average developer's salary by 2028, driven by rising token consumption and a shift to consumption-based fees. These findings underscore the financial pressure now forcing CIOs to abandon the open-access model and embrace cost discipline.