A new survey finds that 84 percent of large enterprises have cancelled at least one AI project because of ageing legacy systems. The findings underline how old infrastructure is slowing AI adoption.

The survey covered 945 chief information officers and chief technology officers. Many said their existing systems cannot support modern AI workloads, forcing projects to be shelved before they reach production.

The report also notes a gap between ambition and readiness. Companies are spending heavily on AI pilots, but the underlying data and technology foundations are not keeping pace.

Experts say modernising infrastructure is now the biggest bottleneck for enterprise AI. Without clean data pipelines and scalable systems, even well-funded projects fail. The findings suggest that 2026 will be the year companies are forced to fix the basics first.

Legacy systems often store data in formats that modern AI tools cannot easily use. Cleaning and migrating that data is expensive and slow, which is why many pilots stall. Consultants say companies that invest in data foundations first see far better returns on their AI spending.