# AI Costs Drop but Enterprise Bills Soar Due to Increased Usage McKinsey highlights the paradox of cheaper AI models leading to higher overall costs. - **Source:** [Fortune | FORTUNE](https://brifing.si/source/769282ef-327a-4454-909a-3c7936216004) - **Published:** 2026-09-23 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/c72846b4-d6ba-4d24-b1c1-a29e25b405ea ## AI model costs plummet GPT-4 launched in early 2023 at $60 per million output tokens. Today, similar performance models cost hundreds of times less, with prices falling from tens of dollars to below $1 per million tokens. This dramatic price reduction reflects rapid advancements in AI efficiency and market competition. However, enterprises are consuming exponentially more AI capabilities, offsetting these unit cost savings. ## Agentic AI consumption surge Businesses are deploying AI agents for complex tasks like code inspection and rewriting entire codebases. These agents perform vastly more reasoning work than human developers typically handle. While individual operations become cheaper, the sheer volume of AI usage is driving total costs upward. AI vendors are capturing some efficiency gains through higher margins, further complicating cost dynamics. ## Unpredictable cost structures Agentic AI introduces variable cost structures unlike traditional software. The same task can cost up to 30 times more between runs due to different reasoning paths. System design choices significantly impact costs - single versus multiple agents can dramatically change task expenses. Much cost comes from the reasoning and refinement processes behind final outputs. ## New evaluation metrics needed McKinsey suggests evaluating AI agents at task level rather than cost per token. Key metrics include task execution cost, agent success rate, and human verification time. Agents create value when verification takes significantly less time than human completion. For example, if a task takes 1 hour manually but 6 minutes to verify, agents with >10% success rate may be worthwhile. ## Three cost management areas McKinsey identifies three levers for AI cost control: visibility into spending drivers, workflow optimization, and sourcing discipline. Companies must track which use cases and models drive costs. They should match model complexity to tasks, cache reusable context, and limit unnecessary tool calls. Vendor management includes removing unused licenses and avoiding single-vendor dependence. ## Workflow redesign challenge The biggest challenge isn't cost reduction but workflow redesign to leverage freed capacity. Companies must determine where AI delivers highest returns rather than indiscriminate cuts. Successful implementation requires understanding how to best utilize the additional capacity that efficient AI agents create within existing business processes. ## Strategic spending approach McKinsey cautions against blanket cost-cutting, advocating instead for strategic optimization toward high-return AI applications. Companies need to identify which AI investments yield maximum business value. This requires balancing cost management with performance optimization, focusing on areas where AI can transform rather than just automate existing processes. ## Future of AI economics As AI capabilities advance, enterprises face complex economic decisions. The paradox of cheaper units but higher total costs reflects AI's transformative potential. Companies must develop sophisticated cost management frameworks that account for variable agentic behaviors while maximizing the strategic value of their AI investments in an increasingly competitive landscape. ## Sources covering this story - https://fortune.com/2026/09/23/mckinsey-cheaper-ai-models-bigger-ai-bills-cfo/ — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._