Enterprise AI adoption may be accelerating, but most organizations have yet to deploy it broadly. An IDC study conducted with AMD reports that 83% of companies use AI across fewer than 10 simultaneous use cases, pointing to a persistent gap between pilots and large-scale operational deployments.
The research also identifies shortcomings in the business planning behind those projects. Only 21.7% of organizations closely assess AI return on investment, while 22.2% examine whether AI initiatives align with their wider strategic goals. Infrastructure constraints, costs and the lack of a clear deployment strategy remain central obstacles.
AI agents present another area of uncertainty. More than 75% of organizations do not yet have a clear understanding of how systems that can perform tasks and interact with applications could fit into their operations.
Inference shifts the infrastructure equation
AMD expects inference, or the running of already-trained models, to account for more than 60% of global AI compute capacity in 2026. That shift increases the importance of enterprise infrastructure built not only for model training but also for sustained production workloads.
GPUs remain essential for demanding AI operations, but the report emphasizes the role of CPUs in supporting agents that need to access databases, run applications and coordinate multiple processes. Effective deployments therefore need a combined CPU and GPU approach rather than relying on a single type of processor.
- 77% of companies prioritize unified hardware and software platforms to simplify integration and system management.
- About two-thirds assess the full cost of generative AI projects from the outset, including infrastructure, software licenses and staffing.
For European organizations, expansion also brings requirements around data protection and sovereignty, traceability and sustainability. The study says open platforms can be appealing in that environment because they allow companies to combine technologies from multiple suppliers instead of becoming dependent on one ecosystem.
IDC's October 2025 report argues that adoption alone is no longer enough: companies seeking measurable outcomes need suitable infrastructure, financial tracking, and security and governance measures integrated from the start.
