Companies that initially accelerated the deployment of artificial intelligence (AI) tools across their workforce are now scaling back usage amid rising costs associated with advanced AI models. Major corporations including Amazon, Walmart, Cisco, Uber, and Meta have introduced measures such as usage caps, discouragement of unnecessary AI interactions, and shifts toward more cost-effective models to manage escalating expenses linked to AI deployment.

This development represents a new phase in corporate AI adoption, as organizations move from using chatbots to implementing AI agents—automated systems capable of performing complex tasks independently but requiring significantly more computing resources. The transition has increased scrutiny on the cost-benefit balance of each AI query or task.

Compounding budget concerns, leading AI providers such as Anthropic and OpenAI have transitioned from flat subscription fees to token-based billing systems, where charges correspond to the amount of data processed by AI models. This pricing structure has made corporate customers more acutely aware of expenses incurred by individual prompts and automated workflows.

Cost considerations have become a priority at the executive level. Costi Perricos, global generative AI leader at Deloitte, noted that CFOs and boards are increasingly focused on compute costs, emphasizing the misconception that AI is inexpensive or free. OpenAI CEO Sam Altman acknowledged that while cost was not a significant concern last year, it has emerged as a “huge issue” for users in 2024.

Several companies have responded with internal controls. Uber, after exhausting its AI budget by April, now limits employees to $1,500 in monthly token spending on each AI tool. Walmart similarly imposes token caps on its internal AI agent users. Suresh Kumar, Walmart’s global chief technology officer, described a rapid surge in usage of the company’s coding assistant platform, now prompting efforts to optimize tool selection.

Cisco’s president and chief product officer, Jeetu Patel, highlighted the substantial infrastructure demands of AI agents compared to chatbots, noting that a single employee might be supported by hundreds or even thousands of agents continuously consuming resources. Analysts at Goldman Sachs project that AI agent token usage could increase 24-fold by 2030, potentially aggravating global semiconductor shortages over the next year to year and a half.

While AI-related spending continues to rise, these cost pressures may impact growth prospects for major AI labs such as Anthropic and OpenAI, both planning initial public offerings at near-trillion-dollar valuations. Notably, since early 2024, Chinese AI models have surpassed U.S. counterparts in token consumption, leveraging lower energy costs and more efficient technology to offer cheaper pricing, strengthening China’s competitive position in the AI sector.

Smaller companies have also encountered abrupt cost increases. Workato’s chief information officer, Carter Busse, reported a sevenfold surge in spending following Anthropic’s switch to token-based pricing, which ended prior usage subsidies. Instead of restricting access, Busse has focused on educating employees about cost management and encourages using older, less expensive models when suitable.

Large corporations have adjusted internal messaging to curb overuse. Amazon cautioned staff against using AI merely to boost internal metrics, and the company revised adoption measurements to reduce costs tied to tool misuse. Meta undertook comparable initiatives earlier in the year. Both firms have developed proprietary AI models and also rely on third-party services like Anthropic’s Claude Code.

AI platform providers, including Microsoft, Amazon, and Google, have implemented tools to route user requests to cost-effective models aligned with the task’s requirements. Microsoft’s GitHub chief operating officer, Kyle Daigle, emphasized discussions around “fit and purpose” in AI model selection, discouraging indiscriminate use of the most advanced models.

Some organizations are directing employees to open-source AI models that can operate locally, thereby minimizing expenses related to external AI service usage and cloud infrastructure.

Despite heightened costs, corporate leaders continue to balance AI investment against promises of improved productivity and innovation. Cisco’s Patel acknowledged the challenge: “Our engineers want more tokens… We have to figure out a way to fund it,” reflecting the ongoing tension between AI’s potential benefits and its financial implications.