Perplexity CEO Says On-Device AI Can Disrupt the Data Centre Industry
Perplexity CEO Aravind Srinivas says the next big shift in artificial intelligence may move computation out of massive server farms and onto the devices we carry. In a wide-ranging podcast with YouTuber Prakhar Gupta, he argued that powerful on-device AI could upend the economics of today’s centralised infrastructure while reshaping how people work with, and think about, intelligent systems.
On-Device Intelligence vs. Centralised Data Centres
“The biggest threat to a data centre is if the intelligence can be packed locally on a chip that’s running on the device, and then there’s no need to run inference on all of it on one centralised data centre.”
Srinivas made the comment when asked which hardware advance could trigger a step change in AI. He sketched a future where many tasks handled by large, specialised facilities move closer to the user. If models can run efficiently on phones, laptops, and edge devices, the need to route every request to a central server diminishes. That shift promises lower latency, potential privacy gains, and reduced dependence on costly, power-hungry cloud infrastructure.
Economics and Architecture in Flux
According to Srinivas, bringing inference to the edge could alter the return on billions of dollars invested in data-centre buildouts. A more decentralised landscape would redistribute compute across countless devices rather than concentrating it in a few hyperscale locations. The result: new traffic patterns, new cost models, and a different balance of hardware priorities—smaller, efficient accelerators in your pocket instead of ever-larger clusters behind the scenes.
While centralised systems won’t disappear—training frontier models and coordinating complex workloads will still require them—Srinivas believes the centre of gravity could shift. Developers might prioritise model compression, quantisation, and energy-aware design so that intelligence can travel with the user rather than live solely in the cloud.
Biology vs. Silicon: The Curiosity Gap
Beyond infrastructure, Srinivas contrasted biological and artificial intelligence. He noted that the human brain is extraordinarily energy-efficient compared to warehouse-scale compute measured per watt. More importantly, he said, people are propelled by intrinsic curiosity—the urge to challenge assumptions and reexamine the familiar. Current AI systems do not naturally possess that drive; they can respond impressively within given prompts and data, but they don’t independently set agendas or pose genuinely novel questions in the way humans do.
This distinction matters for how we deploy AI. Humans offer judgment, creativity, and the spark to pursue new directions. Machines offer scale, speed, and recall. The meaningful progress, in Srinivas’s view, lies in orchestrating the two.
From Chatbots to Agents
Srinivas also discussed the evolution from chatbots—tools that answer questions—to agents that can take actions on a user’s behalf. As models become more capable and context-aware, they can do more than converse: they can plan, execute multi-step tasks, and adapt to feedback. If those agents run locally, they could act faster and handle sensitive data more privately, further strengthening the case for on-device intelligence.
Leveling the Playing Field
Ubiquitous, personalised AI, Srinivas argued, could democratise capability much like smartphones did. Individuals, not just large institutions, would gain access to powerful tools for research, writing, analysis, and decision support. He emphasised that age is not a barrier to adoption; what matters is a mindset of curiosity and a willingness to experiment. People who approach AI as a partner—probing, iterating, and validating—will benefit most.
What to Watch Next
- Advances in on-device accelerators and memory that enable faster, more efficient inference.
- Breakthroughs in model compression and quantisation that preserve quality while shrinking footprints.
- Improved privacy and security techniques that allow sensitive data to stay on-device.
- Agentic workflows that move beyond chat to autonomous, user-directed action.
Srinivas’s message is clear: if intelligence truly becomes portable—packaged into chips inside everyday devices—the cloud will share the stage with the edge. That could redraw business models across the AI stack and expand who benefits from the technology. The result may be less about replacing humans and more about amplifying them—combining machine speed with human curiosity to explore problems in new ways.
