At Cisco Live EMEA in Amsterdam, the networking giant launched a number of artificial intelligence (AI) and datacentre infrastructure products and opened up over some of the tricky issues around supporting the intense energy demands that AI workloads make of current systems, that have been a subject of discussion and concern since ChatGPT went mainstream at the end of 2022.
Last year, National Grid CEO John Pettigrew said AI would be a significant factor in an anticipated sixfold increase in datacentre power requirements between now and the mid-2030s, and a more recently published RAND research report estimated that AI-driven compute would add 68GW (gigawatts) of power demand worldwide by 2027 – almost as much electricity as is consumed by California alone – and over 325GW by 2030.
Tom Gillis, senior vice-president and general manager of Cisco’s security, datacentre, internet and cloud infrastructure group, spoke of an approaching inflection point given the astonishing power requirements of building datacentres fit for the AI era.
And in conversation with Computer Weekly on the fringes of the show, Cisco executive vice-president and chief product officer Jeetu Patel said that while the firm rightly acknowledges the criticality of sustainability, it believes that it is helping keep things moving in the right direction on this issue.
“The larger sustainability issue we should keep in mind is that our products are getting more and more efficient from a power perspective,” said Patel.
However, he continued, if the use case that naturally arises from this improving efficiency is that Cisco users feel comfortable running more and more AI workloads across its infrastructure, the consequential rise in power consumption will quickly wipe out the gains made from the more efficient, probably liquid-cooled, electronics.
“The way I think about this is that you have to get AI to a point where intelligent outcomes come about from AI – because it is smart enough – and that’s the race right now. Once we get there, AI will be able to help us solve a lot of power problems and sustainability problems,” said Patel.
Trump’s impact on the environment
Asked whether or not the second Trump administration’s environmental policies – which include an increased commitment to exploiting the US’s fossil fuel reserves and the country’s withdrawal from the Paris Agreement that aims to limit global heating to 1.5˚C – would impact Cisco’s sustainability goals, Patel said it was too early to tell how things would pan out over the next four years.
However, he said, it is clear that the tech industry today finds itself in both a compute- and energy-constrained environment, particularly in regard to the appetite for power that AI workloads have.
“We will see some major breakthroughs on that front that reduce the cost of compute, and therefore reduce the cost of power consumption,” he said.
“For example, DeepSeek. DeepSeek might have actually been very good for the environment, because what it did was taught the models to train at a lower cost and identify techniques that can actually be more efficient.
“This then will allow you to get many, many more models built at a fraction of the cost of what you could do,” he suggested.