Signal Report forecasts 30 percent of generative workloads running on premises or on device by 2028

Summary

Gartner predicts that by 2028, 30% of generative AI workloads will run domain-specific language models on an organisation's own premises or directly on devices. This means shifting not only to specialised models but also away from relying solely on a provider's cloud. Gartner's development map shows options for self-hosting, running on premises or on devices, alongside cloud deployment. The motivation connects to the need for measurable value, accuracy and compliance in critical workflows. The report does not explain the data or reasoning behind the 30% figure.