The ubiquitious Central Processing Units (CPUs) are far from becoming obsolete in the age of artificial intelligence, despite the hype surrounding Graphic Processing Units (GPUs), as CPUs remain the affordable choice and they will continue to stay as an integral part of processing, according to experts and those working on AI models.

CPUs, sometimes called the “brain” of a computer, are primary processors in any given computer whereas GPUs are the chipsets that became popular to support AI models. Earlier in March, the Ministry of Electronics and IT had talked about subsidising GPU access to a rate of £1 per hour, significantly lower than the global cost of £2 to £2.40 per hour, when launching a new common computing facility. This had raised the question of whether CPUs will be completely replaced by GPUs in the future.

Ramesh Loganathan, Head of the Centre for Innovation and Entrepreneurship at IIIT Hyderabad pointed out that the rate will only be subsidised for specific providers.

“There’s a rate that is better than what different providers would get in the market. This is for research and specific purposes and the subsidy is only to the extent of 40 per cent. You have to fight hard to get it, put in a proposal, go through a review process. The subsidy is not available for everybody. You have to make a case and defend to get it,” said Loganathan.

Similarly, Vikas Tripathi, Vice President Tech and Product at CoRover.AI that owns BharatGPT, pointed out that broadly CPUs can cost between ₹50,000 to ₹1 lakh for a good machine whereas GPUs can cost ₹25-50 lakh depending on the use-case: a 50-100 times increase in cost.

Tripathi was also firm on CPU’s permanence in the tech space stating that all devices in even today’s time need CPU.

Use-case specific

“CPUs will not be replaced because they are always the integral part of the processing. For personal computing machine or a laptop, you cannot replace the CPU with a GPU because it has a high consumption of the power. GPUs are also use-case specific like for graphic design, developing, video editing,” said Tripathi, adding that high-end CPUs are important to run edge implementations or small language models (SLMs).

Tripathi said that CoRover extensively use GPUs for BharatGPT but also uses CPUs for in-house SLMs, which are very domain specific.

NTT Data, a global IT and business services company that talked about launching GPU as a service around October 2025said that there is a lot of workloads that still requires CPUs. The company anticipated anticipated future workloads that would be a hybrid of the two chipsets, current usage of CPUs being around networking equipment facilities. GPUs will have an important part in future but CPUs are likely to stick it out for the long term, said the company at a recent media conference in Mumbai.

Published on March 19, 2025