Signal 75-85% of generative AI proofs of concept are said to fail because of infrastructure gaps

Summary

NTT DATA cites Everest Group research indicating that most generative AI pilots never make it past the proof-of-concept stage. According to the figure, between 75 and 85 percent of generative AI proofs of concept fail. The cause named is shortcomings in infrastructure, rather than problems with the models themselves. The report uses the statistic among its enterprise adoption signals for informed infrastructure. It supports the argument that computing, connectivity and data foundations must be modernised before AI can scale. For organisations, the finding suggests that infrastructure readiness should be assessed early in any AI programme rather than after pilots stall.