The current AI pricing was always going to go away (arnon.dk)

The AI pricing landscape is undergoing a significant shift as major companies like Microsoft and Uber grapple with escalating costs linked to AI capabilities. Microsoft has canceled internal AI licensing agreements, while Uber has depleted its entire AI budget for 2026 in just four months. Industry insiders are noting the end of the so-called "AI subsidy era," where the expectation was that inference costs would continually decrease. However, many have overestimated this trend, leading to unsustainable business models that cannot cope with the rising expenses associated with memory and GPU prices—up to 95% for GPUs and 4x for high-bandwidth memory (HBM) in just 18 months. As companies confront these financial pressures, the focus is shifting from broad AI integration to specifically identifying use cases that justify the incurred inference costs. This change will necessitate a reevaluation of pricing strategies, moving away from flat-rate models to more adaptive structures such as per-action billing, prepaid credits, or hybrid pricing systems. These methods allow companies to align revenue generation with actual usage and cost fluctuations, preventing margin erosion and maintaining growth potential despite rising operational expenses. The need for a more strategic approach reflects a broader understanding that simply adding AI features isn't enough; firms must critically assess the value these technologies provide against their operational costs.