The Evolution of AI: From Foundation Models to Intelligent Agents
Artificial Intelligence (AI) is reshaping economies, transforming industries, and driving innovation across a wide range of sectors. From productivity tools to national digital strategies, AI is becoming an essential component of modern life. The AI landscape is currently undergoing a significant shift—from an era focused on building the largest foundational models to one centered on developing practical, intelligent AI agents that can reason, adapt, and act autonomously. This transition presents new opportunities for planning and policy, particularly for nations like the United Arab Emirates (UAE), which have made substantial investments in AI infrastructure and initiatives such as the UAE’s 2031 AI Strategy.
The Foundation Model Era
In recent years, the focus of AI development has been on foundation models—large neural networks trained on vast datasets. These models are capable of performing a wide array of tasks, including content generation, language translation, code writing, and even creating music and images. According to the AI Index Report 2025, the scale of these models continues to grow rapidly, with training compute doubling every five months, datasets expanding every eight months, and power consumption increasing annually.
The Elo system, used to rank AI models, indicates that the skill gap between top AI models is narrowing. In just one year, the difference between the best and the 10th-best models has decreased by more than half, and the top two models are now nearly identical in performance.
Some experts argue that AI development has reached a peak, suggesting that smaller, more specialized models may be needed moving forward. Reasons cited include limited data availability, computing power, and energy constraints. For example, a single query to ChatGPT consumes about ten times as much energy as a single Google search.
The Infrastructure Behind Foundation Models
Building foundation models requires extensive infrastructure and resources. Major players in the AI space, such as leading tech companies, have made significant breakthroughs, but their achievements would not be possible without the companies that provide the underlying infrastructure for AI development at scale.
Training a foundation model involves massive investment. According to Epoch AI, the cost of training frontier AI models has increased two to three times per year for the past eight years. Researchers predict that training the largest models will cost over a billion dollars by 2027.
This process demands advanced computational systems, including clusters of high-end GPUs or TPUs, fast networking solutions like 100 Gbps Infiniband, and extensive SSD storage to handle large datasets. A robust data infrastructure is also crucial, ensuring efficient collection, cleaning, and secure handling of petabyte-scale data volumes. Distributed machine learning frameworks and orchestration tools are employed to manage complex workloads.
Beyond hardware, maintaining these operations requires sophisticated monitoring tools, reliable backup systems, and physical infrastructure such as uninterrupted power and advanced cooling. Expert teams spanning machine learning engineering and data engineering are essential to oversee the entire process.
The UAE has recognized the importance of this infrastructure and is taking strategic steps to build local capabilities. Entities like G42, the Advanced Technology Research Council, and the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) are investing in digital infrastructure, chip innovation, and AI research. These efforts are not only fostering innovation but also ensuring long-term technological self-reliance and national sovereignty.
From Smart Assistants to Smart Agents
As the momentum in the AI model race stabilizes and infrastructure investments continue, the next major leap is toward agentic AI—AI that can reason, plan, and act on behalf of users. AI is already being widely used across various domains, from entertainment to productivity. Examples include AI avatars and AI companions that assist with note-taking during conference calls, making AI an integral part of daily life.
Agentic AI represents a shift toward a more practical, product-driven approach, showcasing AI’s ability to perform real-life actions rather than just searching, analyzing, and generating data. It combines context awareness, memory, logic, and goal setting to understand and act effectively.
This transformation is especially relevant for the UAE, where digital services are becoming increasingly advanced and interconnected. Agentic systems could help residents with tasks such as document renewal, travel planning, health routine management, and personalized education—all through a single, proactive interface.
Leading the Next AI Frontier
As the focus of AI shifts from large-scale experimentation to localized, deployable solutions, success will depend on integrating smarter AI into everyday life. With AI expected to contribute $100 billion to the UAE’s GDP by 2030, the nation’s forward-thinking leadership, digitally native population, and strong public-private tech ecosystem will drive this growth. From government services and enterprise productivity to healthcare, media, and education, the real value of AI lies in its ability to be context-aware, locally relevant, and seamlessly embedded into everyday systems.