Greg Ryder walked onto the stage at TEDxWollongong with a deceptively simple provocation: AI is too hot right now, and not in a good way. Every image generated, every data set analyzed, every workout plan spat out by a chatbot carries a real physical cost measured in watts and heat, and the bill is staggering. In 2024, the International Energy Agency estimated that data centers consumed roughly 1.5% of global electricity, the equivalent of powering two entire Australias, at a cost of just under a hundred billion dollars. The urgency is not abstract. It is thermodynamic.
Where all that electricity actually goes
Most people picture AI as something airy and intangible, algorithms drifting through the cloud. Ryder, who describes himself as a material scientist working in energy storage and thermal management, wants to correct that picture fast. Inside every data center, tiny chips grind through computations at ferocious speed, and every single computation generates heat. As a chip gets hotter it becomes less efficient, which means each subsequent computation generates slightly more heat than the last, a compounding spiral that the industry cannot ignore.
The industry’s answer has been to spend electricity fighting heat with more electricity. In advanced data centers, 30 to 40% of all power consumed goes purely to cooling, which means that in 2024 alone, close to half a percent of global electricity use went toward keeping chips from frying themselves. Some operators have gone as far as building facilities in the Arctic to cut those costs. The direction the industry is moving, Ryder explained, is direct-to-chip liquid cooling: a metal block sits on top of the hot chip, a coolant mixture of water and antifreeze flows through it, carries the heat away to a chiller, and recirculates. Clean, targeted, and increasingly standard.
But there is a bottleneck nobody talks about in the boardroom. The metal block conducts heat well. The coolant does not. Engineers have added tiny fins and pinprick turbulence channels to speed up the exchange, but the fundamental mismatch between metal and fluid remains. ‘The limits on AI may not be software,’ Ryder said. ‘It may be thermodynamics.’
A fluid engineered to behave like metal
This is where Ryder’s own research enters. Independent modeling showed that tripling the thermal conductivity of the coolant fluid could deliver a more than 50% reduction in the overall energy use of these cooling systems. Fluids with that level of conductivity already exist in nature, the problem is that most of them are hazardous, prohibitively expensive, environmentally unfriendly, or as thick as honey. Pumping something honey-thick through hundreds of thousands of liters of pipe across a data center kills any efficiency gain before it starts.
Using a combination of chemistry and nanotechnology, Ryder’s team built a coolant that achieves the required thermal conductivity without the viscosity problem. The solution moved from gram-scale lab experiments to kilogram batches, and is now being tested in a pilot facility running inside a real server rack. The projected payoff extends well beyond data centers: the same principle applies to space-heating systems such as air conditioners, which account for roughly 7% of global electricity use on their own.
The stakes keep climbing. The CSIRO projects that by 2028, a trillion dollars per year will flow into AI investment, and in several communities where large data centers have already landed, local residents have seen their own power bills rise in step with the facilities next door.
The pilot rack running the numbers
A server rack at the pilot facility is currently circulating Ryder’s nanofluid through its cooling loop while engineers watch the heat exchange data come in, waiting to see whether the lab promise holds at real-world scale.
At the TEDxWollongong stage, Ryder framed the whole endeavor as a lesson in paying attention to the hidden costs inside any booming technology. The chip is visible. The electricity bill is visible. The thermal conductivity gap between a metal block and a coolant is not, until someone with the right background decides to look.
