After the dust has begun settling on the arrival of Chinese artificial intelligence (AI) startup DeepSeek’s new AI chatbot – which claimed training its latest model cost only a small fraction of the multibillion-dollar AI budgets enjoyed by US tech giants such as Microsoft OpenAI for ChatGPT and US-owned Google for Gemini – data centre operators were left with an interesting sliding door moment in their AI strategies.
Despite a huge uptake, Western companies, institutions and governments are already banning the use of the DeepSeek chatbot service so it’s future is uncertain outside China. However, AI companies globally are trying to reveal the methods that it used to achieve leading-edge performance at a fraction of the cost. If DeepSeek’s methods can be reproduced outside China, smaller countries like Australia and New Zealand may have a chance to enter the foundation model space with a sovereign offer – previously impossible to contemplate given the high cost of training models.
For data centre operators in ANZ, this potential shift in the cost of AI technology addresses the threat that using foreign models poses security and reliability risks. One of the biggest operators in sovereign data centres is Macquarie Data Centres and according to group executive David Hirst, the AI industry is evolving faster than any technological trend we’ve seen before and it’s still in its infancy.
“This is an exciting space to be in,” he told W.Media. “AI continues to prove it is, and will continue to be, one of the fastest-evolving technologies the world has ever seen. And we’ve only just scratched the surface of what can be achieved. It will fundamentally change how all industries operate and what humans can achieve.”
Hirst said that in a short amount of time, AI has already made significant progress. “First with the rapid development of LLMs which saw the building, training and inferences of multiple market-ready applications with thousands of use cases. Then we had the introduction of agentic AI tools, which can autonomously plan, execute, and refine tasks with minimal human intervention leading to increased industry efficiencies,” he said.
“Now most recently, DeepSeek’s model claims to reduce the effort and compute required to train AI models by simplifying the training process and more efficiently use hardware, ultimately reducing cost,” said Hirst. “This innovation marks a significant step forward in reducing the barriers to entry to AI training and inference, opening the door for more companies to consume the technology.”
Jevons’ Paradox
“Microsoft CEO Satya Nadella said it best when he explained how DeepSeek could lead to the Jevons’ Paradox [1]—the idea that greater efficiency doesn’t always reduce resource consumption,” said Hirst. Jevons’ Paradox was proposed by 19th century economist William Stanley Jevons, who observed that as coal use became more efficient, it was also paradoxically leading to an increase, and not a decrease, in the consumption of coal across different industries.
Hirst concurs this applies to AI. “Instead, more affordable, accessible AI will lead to increased demand through faster adoption and wider proliferation in the market. As current leaders are forced to innovate faster and new players enter the market, demand for data centres will only continue to skyrocket,” he said.
While that is a double-edged sword from an environmental viewpoint, for sovereignty arguments, the development has implications for ANZ DC operators. “As this innovation curve continues to break down barriers to AI development, it is possible and even inevitable that this could spark innovation in countries currently not in the AI arms race,” he said. “We will see a variety of new training workloads hosted outside of the United States. This will further accelerate capacity demands alongside the existing AI and cloud workloads already onshore.”
Another key Australian data centre player NextDC’s CEO Craig Scroggie described AI as “the arms race of our time”. He described Deepseek’s innovations as fascinating. “Because the model is open-source, the entire ecosystem will benefit – further accelerating both hardware and software evolution. This milestone reinforces what we’ve been seeing for some time: AI breakthroughs are happening faster and their impact is deeper,” he said.
Far from reducing data centre demand, Scroggie agrees that advancements like this will fuel it even further. “It’s not just about training – inference at scale is becoming a bigger driver of compute demand. Jevons’ Paradox is playing out in real time: as AI becomes more efficient and accessible, adoption skyrockets, driving unprecedented demand for compute, data, and next-generation applications,” he posted recently on LinkedIn.
Jevons’ ugly side for emissions
While the economics of DeepSeek, if true, stands to boost further AI uptake opportunities in ANZ, researchers are only just now moving beyond the direct environmental impact of data centres, which may be accelerated by Jevons’ Paradox, to understanding and mapping the ways AI innovations reshape economic structures and societal practices that, in turn, drive increased resource usage.
In a new paper, “From Efficiency Gains to Rebound Effects: The Problem of Jevons’ Paradox in AI’s Polarized Environmental Debate” by Alexandra Sasha Luccioni, Emma Strubell and Australian AI expert Kate Crawford [1], the authors warn that if AI is deployed without adequate consideration of its direct and indirect effects, it has the potential to deepen inequalities, accelerate resource depletion, and exacerbate the very climate problems it hopes to address.
The authors say that while much attention has been given to AI improving productivity and resource efficiency, these gains can result in higher overall consumption. They give an example of the paradox. “For instance, an AI-driven logistics system might reduce delivery times and fuel usage per vehicle, yet simultaneously encourage more frequent online orders, thus elevating total miles driven; AI-driven targeted advertising can help us more easily find the products that we need based on our clicks and searches, but also push us to make more superfluous purchases,” they write.
Such systemic shifts in behaviour challenge linear expectations that efficiency improvements alone will drive down emissions. Instead, argue the authors, they underscore the need for a detailed, interdisciplinary approach that links AI deployment to broader assessments of environmental, social, and economic feedback loops.
[1] https://arxiv.org/abs/2501.16548
The Melbourne Convention 2025: Transforming Enterprise, Datacenters & Cloud in Victoria – colocated with Interconnect World – returns on 3 April 2025 at CENTREPIECE at Melbourne Park. The return takes place as the city achieves recognition as a growing leader in digital infrastructure. Equinix recently described Melbourne as the fastest growing ‘edge metro’ in the world and analysis from DCByte indicates a substantial forward programme of build which could see capacity across the metro area move over the next few years towards 1 GW.
Our Convention for Melbourne in 2025 will focus on the key factors and requirements that are driving enterprise demand and which will drive growth and direction in future demand for digital infrastructure. The emergence of AI and digital technologies will be considered particularly in terms of the profile of different user profiles and requirements. The decisions made on the basis of ‘sustainability’ will also be discussed and broadened to include the issues of how far corporate brand objectives drive technology decisions and the value of achieving an overall industry identity, locally and beyond. Other impact factors from both inside and outside the industry each command their own discussion – hyperscale, legislation, risk and security, connectivity needs.
To view the agenda or plan your visit, please visit:
https://clouddatacenter.events/events/melbourne-convention-2025/
[Author: Simon Dux]
