As we watch the tremendous development of Generative AI, it is clear that this technology would not exist without the advancements of GPUs. Although organisations are eager to dive headfirst into the opportunities that Generative AI can present, some challenges need to be addressed.
Many are competing in the race to build powerful AI models, but the infrastructure required to support them is a double-edged sword. This revolution offers a variety of possibilities but also raises concerns about its ethical use and sustainability.
The Power of GPUs
Let’s start with some facts. When people use applications such as ChatGPT, they do not realise the power required to produce a single output. For example, running three NVIDIA H100 GPUs for one hour uses the same amount of electricity as running a full load in a washing machine. This is approximately two litres of water per hour to cool.
If we really put this into perspective, xAI has deployed 100,000 H100 GPU clusters in a single facility. Can you imagine the size of this launderette?
Naturally, this raises questions about the sustainability of this technology and although these applications are improving our day-to-day lives, are we prepared to handle the environmental costs associated with it?
Deciding on the Right AI Infrastructure
Now we have raised the question of technology and sustainability, you’re probably sitting there wondering which is the right infrastructure, that is both cost-effective and isn’t going to cause devastating consequences to this planet.
Typically, organisations have a variety of options to choose from when it comes to obtaining access to GenAI infrastructure, each with its own set of advantages and disadvantages.
- Public Cloud: The default option for most enterprises
- Private GPU Clouds: Dominated by niche players which offer tailored solutions.
- On-Premise Solutions: Companies that have the resources to invest in their own hardware and the physical space to house them.
You’re now thinking about the few options that one can take to build an amazing GenAI model. However, there’s a catch: access to money equals access to this technology. As we continue to watch Generative AI mature, there is a continuous gap that widens between organisations that have the resources and capabilities to invest in AI infrastructure, and those without. Naturally, this leaves the smaller organisations behind.
The Ethical Use of AI
When ChatGPT was released a few years ago, it triggered a fierce debate. One side was more than enthusiastic about its ability to improve “everything”, whilst the other more risk-averse side had major concerns around GenAI applications being developed with less ethical objectives; which resulted in a push for the EU AI Act and other such legislation.
Regardless of where you stand in this debate, it is clear that the ‘Pandora’s Box’ of Generative AI has been well and truly opened and we can no longer turn back. The question around the ethical use of GenAI isn’t how we use it today, but the associated decisions we make today that will shape our future.
Will we be part of a society where GenAI is a force for good and enhances human creativity sustainably and ethically? Or will we watch it fall into the hands of a very small, but also very wealthy minority who seek to use GenAI for their own quest for power?
The Need for Balance
Generative AI is here to stay. Therefore, it’s on us to ensure we encourage all the various architects, providers and users of this technology to strike the right balance. A balance between overcoming the continuous thirst for speed and winning the GenAI arms race, with a more considered approach that ensures sustainability and ethical controls are at the heart of all advancements.
AI manufacturers, cloud service providers and AI developers are the ones that need to be held responsible for building Generative AI infrastructure in the right way. They need to be:
- Locating infrastructure at renewable energy sources.
- Utilising liquid cooling to improve energy efficiency.
- Continuously maintaining transparency around PUE (Power Usage Effectiveness) and WUE (Water Usage Effectiveness) metrics.
- Ensuring energy usage metrics are transparent with customers.
Those developing the foundational models should not just be looking at the cost of GPU access as the primary determining factor. The challenge is clear, but the solution is even clearer:
- Model markers need to develop AI on platforms that are built for sustainability and in which the technology can be used ethically.
- Sellers need to do their due diligence and screen their buyers regardless of how attractive the revenue is.
- Buyers need to demand transparency and clarity on aspects such as sustainability, compliance, sovereignty, etc.
The Road Ahead
Our choices today and how we build and use GenAI infrastructure ultimately determine the future. We need to ensure that the rapid growth of Generative AI does not come at the expense of destroying our planet but also protects society.
Designing AI systems that have a synchronised balance of efficiency and output, whilst placing sustainability and ethics at the core of the development will allow us to unlock the true potential of AI.
