IBM CEO: rationing AI is a slow way to lose

In a recent interview, IBM CEO <name> warned that rationing artificial intelligence (AI) systems is an ineffective strategy for managing costs and could ultimately lead to losing out on the technology's potential benefits. The CEO emphasized that the real cost of AI lies in engineering efforts, as AI agents require a significant amount of data to function effectively.

Companies are currently paying to re-read the same documents repeatedly, similar to how the early days of the commercial internet incurred costs due to repetitive network traffic.

To address this issue, the CEO suggested implementing caching mechanisms and other optimization techniques to minimize redundant data processing. By caching frequently accessed data and utilizing parallel processing, the hardware can handle significantly more work with the same hardware resources. The CEO also highlighted the importance of choosing AI models that align with the specific needs of the workload and the company's data infrastructure.

By running open-weight models on internal infrastructure, companies can achieve substantial cost savings compared to proprietary models.

The CEO further explained that managing AI costs requires a comprehensive approach that includes owning the AI operating layer. This layer connects various systems, governs user access, determines workload locations based on value and economics, and maintains a clear record for regulatory purposes. The CEO emphasized that IBM is uniquely positioned to help clients own and manage this layer, ensuring that AI costs are driven by their data and models rather than relying on external suppliers.

In conclusion, the CEO argued that rationing AI systems is a suboptimal strategy and that businesses should focus on optimizing their AI operations to maximize value while minimizing costs. By implementing caching, choosing appropriate models, and owning the AI operating layer, companies can bend the AI cost curve and unlock the full potential of this transformative technology.