The Future of AI: Agentic AI and Its Economic Implications
In the rapidly evolving landscape of artificial intelligence (AI), the distinction between consumer chatbots and agentic AI is crucial. While chatbots serve basic conversational needs, it is agentic AI—an advanced form of AI capable of independent reasoning and action—that holds the potential to transform industries and secure sustainable profits. According to GlobalData, a leading intelligence and productivity platform, this distinction is critical for determining whether the substantial investments in large language models (LLMs) will lead to profitable outcomes or merely sink into the abyss of unfulfilled potential.
The Shift Toward AI-Native Operations
The financial commitment from companies into AI infrastructure signals a collective vision: we are on the brink of an AI-native reality. Generative AI, agentic AI, and machine learning are anticipated to become integral components of future enterprise operations and workflows. This ambitious shift is underpinned by the belief that advanced AI systems will automate complex tasks, streamlining operations and boosting overall productivity.
Understanding Revenue through GlobalData’s Financial Model
GlobalData has crafted a financial model aimed at demystifying how consumer and enterprise adoption of generative AI can translate into substantial revenue. The key takeaway from their latest Strategic Intelligence report, “The AI Journey – From Generative to Agentic,” is clear: while consumer adoption can generate recurring subscription income, it is the usage fees tied to enterprise applications—that is, the tokens sold for API calls—where the serious profitability lies.
The Rise of API Calls and Token Consumption
As enterprises increasingly incorporate agentic AI software into their daily operations, a monumental rise in API calls is expected. Over the next two to four years, businesses will be making tens of thousands of calls to LLMs daily, resulting in the consumption of millions, if not trillions, of tokens each day. This high volume is essential for offsetting the significant capital expenditures involved in building and operating advanced AI data centers.
Energy Consumption: A Double-Edged Sword
William Rojas, Director of Tech Research at GlobalData, emphasizes that energy consumption plays a pivotal role in the generative AI economic model. The energy consumed, measured in watts per prompt, is intrinsically tied to computational demands, which include the floating point operations per second (FLOPs) necessary for processing LLMs. For instance, with advanced models like ChatGPT-5 and DeepSeek V1 boasting between one to two trillion parameters, the computational burden remains substantial. Estimates suggest that even after employing various optimization techniques, processing each token could still require between 100 to 200 billion FLOPs.
The Token Explosion and Its Implications
As the industry transitions toward reasoning models, the context window is expected to expand dramatically, potentially increasing the number of tokens generated per prompt tenfold or more. This phenomenon, termed a “token explosion,” reflects not only a surge in capabilities but also the increasing demand for computational power and energy—a factor that will significantly impact net margins for AI providers.
Identifying Winners and Losers
As the landscape evolves, the players that stand to gain the most are those providing the crucial hardware and facilities for AI data centers. They are well-positioned to benefit from the ongoing capital expenditure boom. However, the creators of LLMs themselves are facing challenges. Despite the immense potential, they are currently encumbered by rising token processing costs, which negate profit-making possibilities in the short term.
The Unique Business Model of Generative AI
The generative AI business model is distinct due to the central role energy consumption plays in its economic sustainability. As the number of tokens processed per prompt surges and the demand for resource-intensive computations continues to rise, the cost structure for AI providers becomes increasingly complex and precarious.
The Semiconductor Race: A Sisyphean Challenge
Rojas encapsulates the challenges faced by the industry with a poignant metaphor. The efforts within the semiconductor sector to enhance the cost performance of GPUs, high-bandwidth memory, and server networking feel akin to the plight of Sisyphus, the mythological figure doomed to endlessly roll a boulder uphill only for it to roll back down. The continuous advancements in generative AI capabilities mean that the quest for cost-effective performance is likely to persist indefinitely.
As AI integrates more deeply into everyday business practices, understanding the dynamics between consumer interaction and enterprise applications will be paramount. The journey toward a profitable AI future hinges not just on advanced technology but also on navigating the intricate interplay of costs, energy consumption, and market demands.