A recent study by Alex De Vries, PhD candidate at the VU Amsterdam School of Business and Economics, called the Growing energy footprint of AI, suggested that the AI industry could consume as much energy as a country the size of the Netherlands by 2027. While this is incredibly illustrative, it echoes the concerns of academics, such as Professor Kate Crawford who has long warned of the environmental risks of AI. While the data center industry has been one of the most progressive, when it comes to addressing energy consumption and carbon emissions, clearly there is a lot more that needs to be done, especially given the rapidly growing interest in AI.
Senior Cloud Economist at Nutanix.
The impact of generative AI
The impact of generative AI tools, such as ChatGPT has been the story of the year. Boardrooms across the country have been busy setting up working groups to determine how AI can help propel their businesses forward. In fact, in our State of Enterprise AI report, we found that organizations of all shapes and sizes want to embrace AI technologies, as soon as possible.
While 90% of respondents say AI is a priority for their organization, the report also reveals a widespread lack of readiness. From measuring energy consumption to managing and securing data effectively, many organizations are still determining which IT environments are best to run different parts of their AI processes and workloads, or even which type of AI applications are most useful.
However, two major concerns are its environmental impact, especially at a time of increased ESG reporting responsibilities and data privacy. Over 90% of respondents to the State of AI report say that security and reliability are important considerations in their AI strategy - data security and governance, including data quality and data protection, are of paramount importance to support AI technologies and services.
Environmental impact and privacy concerns
What is interesting in the survey is that so many respondents recognize the need to align their ESG reporting with AI needs but few actually know how or have the relevant skills. ESG is actually ranked as a key area requiring AI skills development over the next 12 months.
As we know, ESG reporting is now mandatory in some regions. There is certainly growing pressure from governments, investors, and customers and in 2024 companies in Europe and US-based companies with European links will have to report their carbon emissions within the ‘three scopes’, which are essentially the ‘direct’ ‘indirect’ and ‘all’ carbon emissions.
As it stands, AI use has the potential to blow this apart, such is the demand of the technology on IT infrastructure. Enterprises understand the massive amounts of energy required to run compute- and GPU-hungry AI algorithms and workloads but knowing whether to go it alone and build bespoke AI systems on prem, use public cloud services, or go for a standalone platform on the edge, requires some thought and planning.
Another big concern is data privacy. AI engines require data for training purposes, so how can organizations maintain control of their own data if they are using public cloud-based services for AI? This lack of control of data is almost certainly a governance concern, as no businesses can be entirely sure where their data will end up. As well as potential latency issues and the costs associated with the flex nature of public cloud, data privacy has to be a primary objective.
We would argue the costs of a DIY-approach, as well as the global shortage of GPUs, and the internal AI skills required, make going it alone a non-starter. A standalone, AI platform on the edge, offers data privacy and speed, as well as a small footprint with reduced energy consumption and carbon impact. Clearly, understanding how the organisation wants to deploy AI is key but the nascent nature of this technology segment means there is a dearth of strategic best practices, established guardrails, or even reference architectures.
Organizing infrastructure
The research shows that despite the complications and challenges of a DIY approach, 59% of respondents are expecting to run AI solutions on-prem or in a private cloud, with 51% going for managed data centers and 44% at the edge. The challenge will be how to manage these services within the realms of ESG and data governance and that may yet to determine how these figures shape-up in future.
The bottom line is infrastructure. If businesses are going to pursue an AI future, we all have to look at infrastructure differently. We cannot just assume that AI is just another application that can be thrown onto the cloud computing pile and everything will be ok. There is more at stake here, not just the security of data and the future of the planet (if that’s not enough) but also the viability of businesses deploying AI systems. Recognizing and confronting the challenges now will go a long way to ensuring organizations don’t make expensive mistakes that could result in costly environmental or reputational damage.
