The rise of on-device AI threat could reshape the future of data centres, according to Perplexity CEO Aravind Srinivas. Speaking in a recent podcast, Srinivas warned that as artificial intelligence becomes capable of running directly on user devices, the need for massive centralized data centres may decline significantly. His remarks come at a time when tech giants are spending billions to expand global AI infrastructure.
Today, companies like Google, Meta, Microsoft, OpenAI, and Perplexity are investing heavily in data centres to support large-scale AI models. Industry estimates suggest total spending could approach $1 trillion by the end of the decade. These facilities are essential for powering cloud-based AI tools, but they also consume enormous energy and resources. Srinivas believes this model may not dominate forever.
According to him, the on-device AI threat lies in the ability to run intelligent models locally on consumer hardware. In the podcast with Prakhar Gupta, Srinivas explained that if AI intelligence can be compressed and executed efficiently on personal devices, the dependence on centralized servers could drop sharply. This shift would fundamentally challenge the role of traditional data centres in the AI ecosystem.
He noted that on-device AI offers several advantages. First, it reduces latency since processing happens directly on the device. Second, it improves privacy, as user data does not need to be sent to external servers.
Third, it lowers operational costs by reducing the need for large-scale cloud infrastructure and constant cooling systems. These factors make local AI execution increasingly attractive for both users and developers.
At present, most popular AI chatbots such as ChatGPT, Gemini, and Perplexity rely on powerful data centres to function. These facilities consume vast amounts of electricity and water, raising environmental concerns. As AI adoption grows, so does the pressure on energy grids and sustainability goals. This is one reason the on-device AI threat is gaining attention among industry leaders.
Srinivas also highlighted how on-device models could better adapt to individual users. Since the AI would “live on your computer,” it could learn personal preferences more effectively while keeping sensitive information private. This approach contrasts with cloud-based systems that process data remotely and must comply with complex data handling rules.
However, he acknowledged that current hardware still limits this vision. Today’s AI models are large and resource-intensive, making them difficult to run efficiently on smartphones or laptops. But he remains optimistic.
Advances in chip technology from companies like Apple and Qualcomm are steadily improving performance while reducing power consumption. Over time, these improvements could make on-device AI practical at scale.
Another key point raised was the issue of AI hallucinations. Srinivas admitted that today’s models still generate incorrect or misleading responses at times. However, he believes these issues can be largely resolved within the next five years as models become more refined and better optimized for local processing.
The implications for the tech industry are significant. If on-device AI becomes mainstream, it could reduce the dominance of cloud providers and change how AI services are monetized. Data centre investments may slow, while demand for advanced chips and edge computing solutions rises. For companies heavily invested in centralized infrastructure, this shift could be disruptive.
From a broader perspective, the on-device AI threat also raises strategic questions for governments and regulators. Reduced reliance on centralized systems could improve data sovereignty and reduce cross-border data transfers. At the same time, it could introduce new challenges around device security and software standardization.
Ultimately, Srinivas’ comments point to a future where AI is more personal, private, and efficient. While data centres will remain essential for training large models, inference may gradually move closer to users. If that happens, the balance of power in the AI ecosystem could shift dramatically over the next decade.
As AI continues to evolve, the race is no longer just about building bigger data centres. It is increasingly about making intelligence smaller, faster, and closer to the user. Stay ahead of the AI curve, visit ainewstoday.org for more breaking updates, insights, and expert analysis on the future of artificial intelligence.
