In recent years, technology giants such as Google, Meta, Microsoft, OpenAI or Perplexity have aggressively invested billions of USD in building and expanding data centers.
According to analysts' forecasts, total spending on this infrastructure could approach the $1 trillion mark by the end of the decade. However, not everyone in the industry believes that data centers will forever be the "backbone" of artificial intelligence.
Perplexity CEO Aravind Srinivas has given a noteworthy view that the biggest threat to data centers is the ability of artificial intelligence to be integrated and operate directly on users' devices. At that time, AI no longer needs to process all data at huge centralized facilities.
Currently, most popular AI chatbots such as ChatGPT, Gemini or Perplexity rely on server systems located in data centers to process user requests.
These facilities consume a very large amount of electricity, requiring high maintenance costs and heavily dependent on water for cooling. This is also the reason why AI infrastructure is increasingly questioned about sustainability.
According to Srinivas, if AI models can run locally on devices, the role of data centers will be significantly reduced.
Processing right on a personal computer or phone not only helps reduce electricity and operating costs but also brings clear benefits in terms of privacy.
User data will not need to be sent to the server remotely, reducing the risk of leakage or illegal exploitation.
Another advantage of AI running on devices, according to CEO Perplexity, is the ability to personalize. When the model "lies right on the machine", AI can easily adapt to the habits and preferences of each user without relying on complex cloud infrastructure.
However, Srinivas also acknowledges that the biggest barrier today is the scale and resource consumption of AI models.
Most current models are still too heavy to operate effectively on smartphones or laptops.
However, he believes that rapid advances in chip technology, especially from companies like Apple or Qualcomm, may soon change the situation.
In addition, CEO Perplexity also mentioned the issue of "vision", an AI phenomenon that creates false information but sounds convincing.
He admitted that current systems still encounter this error, but predicted that within the next 5 years, the phenomenon of hallucinations could be completely overcome.
Srinivas's assessment shows a different perspective for the AI industry, from heavily dependent on giant data centers, artificial intelligence may gradually shift to personal devices.
If this becomes a reality, the AI race in the future may not only lie in infrastructure scale, but also in the power of compact chips in users' pockets.
