What we know

AI’s total cost extends beyond the well-known expenses of GPUs and compute power to include significant storage demands. When AI models perform inference—making predictions or decisions on new data—they generate and retain large amounts of data. This data accumulation requires substantial storage capacity, which can lead to ongoing costs that rival or even surpass the initial expenses of training the models. Storage costs linked to AI inference can be as high as or higher than training costs. AI inference produces extensive data that must be stored for operational needs.

Why it matters

While AI training is recognized for its heavy use of GPUs and energy, the costs do not stop once models are deployed. Each inference generates logs, intermediate outputs, and other data that organizations often need to keep for compliance, auditing, or future model improvements. These storage requirements create a persistent and sometimes overlooked expense that can significantly impact the total cost of AI projects.

What is still unknown

Details about the accuracy of storage management approaches, supported languages, and timelines for broader adoption or release are not available. No independent tests of cost efficiency or data handling effectiveness have been reported.