High Chip Costs and Cheap AI Models Threaten Infrastructure
Infosys CEO Salil Parekh warns that high chip costs and cheaper AI models could threaten data centre capacity and infrastructure investments.
The massive global rush to build Artificial Intelligence infrastructure is facing a reality check. Infosys CEO Salil Parekh has raised significant concerns regarding the long-term viability of current AI investments, citing skyrocketing costs and rapid technological shifts that could leave expensive data centres underutilised.
The High Price of AI Power
Speaking at the Bloomberg New Economy 2026 event, Parekh highlighted that the financial barrier to entry for AI is immense. The primary drivers of this risk are the astronomical costs associated with constructing data centres and the "massively expensive" nature of the chips required to power them.
Beyond the initial capital expenditure, there is a looming question of longevity. Parekh noted that because the technology is evolving so rapidly, it remains unclear how long the current capacity being built today will remain relevant or technologically competitive in the near future.
The Rise of Cheaper AI Alternatives
A major disruption is coming from the software side. While massive companies have dominated the field with expensive, closed-source models, new foundation models are emerging using entirely different architectural approaches. These new methods are significantly reducing the cost of "inference"—the process of a model providing an answer.
This shift is changing how enterprise clients approach AI procurement. Parekh observed a growing trend where businesses are questioning the necessity of using the most advanced, high-cost models for every task. Instead, clients are increasingly asking:
- Can we achieve the same results using lower-cost tokens?
- Do we actually need the most recent, most expensive models for this specific workload?
This move toward cost-efficiency could potentially impact the overall capacity usage of the massive data centres currently being constructed.
Data: The New Fast-Growing Commodity
While hardware and model costs pose risks, the demand for data is reaching unprecedented levels. Parekh described data as becoming the "fastest growing commodity," particularly for large-scale enterprises and government bodies.
However, the way data is handled is becoming more complex. He pointed out that requirements for sovereign clouds—where data must reside within specific national borders—are adding new layers of restriction to how capacity is built out. Moving forward, the industry will see an intense focus on how data is scaled for both massive corporations and sovereign nation-states.
The Human Element: Training and Skilling
As Infosys moves to support this transition, the company is focusing on two key areas: building AI agents for enterprise clients and helping companies migrate legacy technology to modern systems. Yet, Parekh emphasized that the technology is only as good as the people using it.
"What is critical there is really training and skilling our employees and making sure that they are available and ready to use all these techniques for our clients," Parekh stated, underlining that human expertise remains the vital link in the AI revolution.