Behrad Babaee (pictured), Technology Evangelist at Aerospike outlines the challenge and the action organisations can start to take to minimise the impact of their AI on sustainability goals and reporting.

AI is going to continue impacting the working world in lots of ways. We are starting to see the first tentative steps of a concerted effort by vendors and governments to ensure we all reap the benefits. Efforts are also being made to manage the more extreme possibilities of frontier AI, as the recent UK Government AI Summit in November has shown. Issues such as bias, fraud, deep fakes and misinformation are all being tackled.

Add to this the more recent advances in areas such as Explainable AI (XAI) – which seeks to ensure companies understand how their AI solutions reach decisions and that they are complying with relevant industry and AI regulations  then it is easy to start feeling positive not only about the technical advances in AI, but how it will ultimately fit into society.

Whilst this is all a great start, there is one aspect of AI that we have not seen mentioned much, yet which poses a significant threat for the organisations that are investing heavily in this technology to support their goals, customers, and to improve the efficiency of their operations.

The hot potato

Sustainable data centres are nothing new to the tech world, with recycling and advances in cooling, for example, having a significant impact on reducing the effect of the industry on the world. But, in recent years, greater attention has been placed on its impact as we rely more heavily on cloud computing and the services provided by hyperscaler cloud providers, whether that be for gaming, to serve content, or provide raw storage and compute power. I am sure we can all also remember the startling facts about the environmental impact of all the energy and data centre power that was being dedicated to crypto coin mining.

While the tech industry is doing a great deal, new legislation is going to force businesses to have a better understanding of the sustainability impact of their IT operations, which will include their use of high-performance computing tasks such as machine learning and AI. Starting in 2025, new EU Corporate Sustainability Reporting Directive legislation (CSRD) will require organisations to calculate and publish their emissions in a much tighter framework than has been required to date. 

The biggest companies – those that are exchange listed with more than 500 employees – will be required to report on their 2024 emissions from 2025. That is nearly 12,000 organisations in Europe. In 2025, companies with two or more of other criteria will be required to comply with the framework too. Those criteria are revenues of over €40m, €20m in assets or 250+ employees.

It has become common for many companies to to use cloud compute services for the heavy lifting of storing data and training their AI. The hot potato was with the cloud providers, making them a cheap and easy way to offload the emissions problem of their IT operations. As things stand, in 2026, the potato shifts to the end user of the services, forcing them to report the emissions.

Cooling things down

For much of the IT industry’s history, efficiency has been the holy grail of computing. From its earliest days and up to about 2005, every line of code would be poured over to squeeze cycles and bytes out. During this time frame, hardware was scarce and expensive, making running inefficient software uneconomical.  

The internet started to change that in the late 90s.  Efficiency in terms of speed became less important because, on the one hand, to handle the scale of the internet, we needed scalability, not speed/efficiency, and on the other hand, many of our platforms were fast enough. Therefore, scalability over speed/efficiency became the order of the day.

Many software platforms today claim to be elastic, meaning that if the workload doubles in the future, the platform will handle it, if the amount of resources dedicated to it doubles. Although that works, it shouldn’t be forgotten that a 50% more efficient software platform can handle the doubled workload on the same amount of hardware!

As the size of workloads doubles every now and then, doubling the amount of resources forever is not a sustainable solution. The new CSRD legislation will hopefully make it more apparent that the industry has to focus on efficiency once again, to reduce the hardware footprint.

Changing the mindset

Companies need to start thinking about the efficiency of the software or cloud services they buy and the code they write. In the same way we consider the range of an electric car, or the emissions of a vehicle or the energy rating of a domestic appliance when we buy them, we need to do the same with software and assess its efficiency.

Research has shown that moving a database from a first-generation NoSQL database to a contemporary database like Aerospike can reduce cost and emissions by 80%. While energy ratings for software do not exist in a standardised format today, some companies, like Aerospike, are producing independent reports to demonstrate the efficiency of their software. Development teams can even get energy efficiency modelling add-ons for their development platforms that will help them assess and improve their code.

Shifting the mindset now will give companies a chance to start asking the right questions and make changes before the CRSD takes effect. The new reporting framework is much more stringent and designed to make it hard for companies to hide their emissions impact, as well as hold them to account.  In the same way that sustainability has been a focus for areas such as the supply chain and manufacturing, the CRSD will ensure that AI’s impact is accurately accounted for. Now, while there is still time, speak with teams about the efficiency of applications wherever they sit in the infrastructure.