As generative AI tools become increasingly integrated into business operations, the associated costs are rising rapidly. This has created a surge in demand for tokens—the standard units used to measure and price AI usage. According to Google CEO Sundar Pichai, the company alone processes 3.2 quadrillion tokens monthly.
The escalating token expenses have prompted businesses to seek ways to reduce spending without sacrificing productivity. One approach is to utilize more cost-effective models, such as Google’s Gemini 3.5 Flash, which offers “frontier capabilities at less than half the price” of comparable models. Experts suggest that companies could realize significant savings by strategically combining different AI models based on specific needs.
Beyond model selection, optimizing infrastructure and hardware can also yield substantial cost reductions. Solutions like caching mechanisms, data virtualization layers, and efficient prompt engineering are gaining traction. For example, ManpowerGroup reduced token consumption by 60% through more refined prompts for their talent acquisition platform.
The shift towards local AI processing is another emerging trend. Nvidia and Microsoft recently introduced RTX Spark, a desktop PC capable of running large language models locally—eliminating usage-based pricing entirely. This aligns with broader efforts to enhance data sovereignty and reduce reliance on cloud providers.
As the industry matures, measurement frameworks may evolve beyond token counts toward outcome-based pricing, where value is determined by results rather than input units. Several companies are already exploring this shift, potentially marking a fundamental change in how AI consumption is tracked and valued.