The rapid advancement and uptake of artificial intelligence (AI) has resulted in a significant increase in projected demand and subsequent investment in data centres. Questions are now being asked about the sustainability of such investments, the accuracy of demand projections, and the true implications for future growth.
Concerns include whether the current level of data centre investment driven by AI growth is sustainable, the impact of growing scrutiny over the projected AI workload power demand from data centres, and whether this demand is manageable or even possible within existing and future energy frameworks.
On this basis, we thought it would be interesting to ask AI itself, in the form of ChatGPT, what it thought of this topic and its potential outcome.
To do this we asked our Head of Technical Due Diligence, Mark Acton, to come up with a couple of questions that would test ChatGPT’s take on the future of AI concerning the viability of data centre investments, based on AI demand projections.
Is the Current Amount of Investment in Data Centres Based on AI Demand Resulting in an Investment Bubble, and Will That Bubble Burst?
ChatGPT:
There are signs that the current investment in modern data centres driven by AI demand could form a bubble, such as rapid growth and high expectations. However, sustained AI demand, technological advancements, and diverse applications could stabilise the market.
Risks to a data centre include overcapacity, economic downturns, and regulatory changes, but the market might stabilise rather than burst if these factors are managed well. Therefore, while a bubble is possible, these stabilising factors mitigate the likelihood of it bursting.
Mark Acton:
Although brief, overall I agree with the response from Chat GPT. I would however add the risk factor that if the way we do AI changes significantly, either on the hardware or software side, this could result in reduced demand for both space and cooling at data centre facilities.
It is well known that very intensive research is being conducted to achieve improvements in this area.
Is the Projected Power Demand by Data Centres Based on Increasing Power Usage by AI Workloads Sustainable?
ChatGPT
The sustainability of the projected power demand by data centres, driven by increasing AI workloads, depends on several key factors:
- Energy Efficiency Innovations: Advances in energy-efficient hardware and cooling technologies can help manage power consumption.
- Data Centre Renewable Energy Adoption: Increasing the use of renewable energy sources can offset the environmental impact of higher power usage.
- Optimised AI Workloads: Improvements in AI algorithms and more efficient data centre management can reduce energy requirements.
- Regulatory and Policy Support: Government policies promoting sustainable practices and renewable energy use can help ensure long-term sustainability.
While the projected power demand is significant, its sustainability depends on technological advancements, renewable energy adoption, optimised AI workloads, and supportive regulatory frameworks.
Mark Acton:
A key point about the response above is that the word ‘sustainable’ was interpreted in the environmental sense. The intent of the question was really directed in the growth sense, i.e. can the current and predicted rates of growth continue? The response touched on this to a degree but largely focussed on the environmental aspects.
I clearly need some AI prompt training!
However, there are a few points where I disagree with ChatGPT:
- The use of renewable power does not reduce power demand from network infrastructure, it merely reduces the environmental impact as the answer suggests. What is not considered is the availability of this power to support data centre growth (other industries need it too!), plus the challenges in overcoming the intermittency of many commonly used renewable power sources.
- Energy efficiency improvements will certainly help in this area but focussing on cooling alone will result in diminishing returns, whatever technology is employed. Energy-efficient IT hardware, software, improved algorithms and optimised workloads as suggested above will have a far greater impact.
Power demand is driven by the IT hardware and data centre services that are hosted within the building, regardless of the workloads being supported. The building does not create that demand, it merely satisfies it.
If we reduce the demand from the IT infrastructure we reduce the demand by the data centre, ultimately reducing environmental impact and allowing more services to be delivered for the same power usage.
- The huge potential of moving away from current chip limitations, based on a silicon substrate and the advent of new processing techniques for all workloads, not just AI, is not really included in the answer.
These could vastly reduce power demands for all types of data centre facilities, and significantly reduce both power and space demands.
How can Future-tech Help?
The rapid growth of AI has spurred significant investment in data centres, raising questions about sustainability.
Using Future-tech’s expert advisory and feasibility services, our data centre consultants can help your organisation ensure your data centre projects are viable, addressing concerns about potential investment bubbles, and managing power demands within energy frameworks.
Our expertise in optimising AI workloads and sustainable practices is crucial for your data centres’ long-term success. For more information, contact our team today.
Epilogue
Since this article was written, the stock market adjustments worldwide on the 1st/2nd of August off the back of announcements by Intel and others suggest that the question being asked was a little more prescient than expected, and ChatGPT’s response was rather more bullish than accurate!
The path to real financial benefit from AI is a great deal more complex than the initial hype would suggest. In this context, it is worth remembering the wisdom of the late Roy Amara, whose ‘law’ applies to AI at present and states that ‘We tend to overestimate the effect of a technology in the short term and underestimate the effect in the long term.’