TY  - JOUR
AU  - Hristov, Hristo 
AU  - Somova, Elena 
AU  - Fuchedzhiev, Emil 
PY  - 2026
TI  - Structural Barriers to the Deployment of AI Foundation Models
JF  - Journal of Computer Science
VL  - 22
IS  - 10
DO  - 10.3844/jcssp.2026.3068.3081
UR  - https://thescipub.com/abstract/jcssp.2026.3068.3081
AB  - Artificial Intelligence (AI) foundation models excel at benchmarks but are not always cost-effective or successfully deployed in enterprise systems. This paper analyzes the failures in their deployment through interconnected structural components characterized by: Economic constraints (computational costs, memory, and latency that scale with increasing use of AI model services), cognitive constraints (related to training objectives that optimize plausibility rather than causal reasoning and calibrated uncertainty), and socio-technical constraints (requirements for organizational change, accountability, and governance). The study examines why the deployment of AI models in real-world enterprise environments remains difficult. It focuses on four themes – scalability, reliability, interpretability, and practical deployment – aiming to demonstrate why these problems do not disappear even as models improve, and how these issues reinforce each other, thereby leading to higher costs and greater risk. In conclusion, the study outlines alternative development pathways, including the use of alternative architectural paradigms such as Retrieval-Augmented Generation (RAG), hybrid Neuro-symbolic AI, compound and multi-agent systems, and human-in-the-loop designs, which may offer more practical and sustainable ways to align foundation models with the organizational, operational, and governance requirements of the enterprise environment. Additionally, a four-dimensional framework is proposed for assessing a model’s deployment readiness based on economic viability, reliability and safety, auditability, and organizational readiness.