AI: Sustainability friend or foe?

While AI can play a key role in supporting telcos to drive sustainable practices, we must acknowledge the negative implications AI will have for sustainability – particularly when it comes to the energy consumption required to build and run AI models. We explore how telcos can face the challenge of leveraging AI in a way that supports sustainable growth.

AI will play a role in driving sustainable transformation, but we must be mindful of the negative implications for sustainability

In a recent article we discussed how telcos can leverage AI to drive sustainable transformation. We covered the various ways AI can support more sustainable networks, from enabling self-optimised networks, advanced sleep modes and ML-optimised QoS scores, to supporting more granular data capture and emissions reporting.

While it is clear that AI has the potential to play a substantial role in supporting telcos to drive sustainable practices and accelerate the move to net zero, it is important to acknowledge that AI can have negative implications for sustainability – particularly when it comes to energy consumption and the embedded carbon in the vast compute infrastructure needed to build and run AI models.

Open AI published research showing that there has been an exponential increase in the amount of compute power used in the largest AI training runs since 2012. They found that the compute has a 3.4-month doubling time – massively outstripping the historical trajectory of Moore’s law, which had a 2-year doubling period. Since 2012, the metric has increased by more than 300,000x, far exceeding the 7x increase that a 2-year doubling period would yield.

Another study conducted by researchers at the University of Massachusetts showed that training a large AI model can produce 280,000kg of CO2eq emissions, which is the equivalent of around 300 round-trip flights between London and New York.  When we factor-in the number of unsuccessful AI models, this places a hefty carbon toll on the few that prove out.

Carbon emission benchmarks

Source: College of Information and Copmuter Sciences at University of Massachusetts Amherst, Oct 2019

If sustainability is a strategic focus, our support and insights can help

The telecoms industry is key to reducing carbon emissions. Discover how with STL Partners’ sustainability insights. Schedule your demo today.

Book a demo

While this surge in compute power has been a driving force behind significant AI capabilities, it raises critical considerations. The huge amount of Greenhouse Gas (GHG) emissions associated with training and running AI models, combined with the potential for rapid increases in the years to come, will have significant environmental implications. And this doesn’t even take into account the non-environmental sustainabiltiy risks, such as potential biases in algorithms, security concerns, bad actor empowerment, and other societal impacts.

Fortunately, telcos are not blind to the sustainability challenges that AI poses. STL Partners conducted a poll at Digital Transformation Week to understand if operators feel the power consumption of AI (and associated scope 2 & 3 emissions) justifies the benefits. We found that there are a wide range of perspectives on the purported benefits of AI for sustainability, despite the enthusiasm surrounding this technology. Surprisingly, more than 40% of respondents expressed a cautious to negative view, highlighting the nuanced attitudes towards AI in the context of sustainability.

In networks does the power consumption burden of AI justify the benefits?

Source: STL Partners poll at DTW, Sept 2023

It is clear that there is a need to find the delicate balance between pushing the boundaries of AI capabilities and ensuring responsible, sustainable development. For some telcos this means seriously considering if pursuing AI is a strategy that will enable them to grow in a way that aligns with their core values of sustainability.

Another ‘quirk’ of emissions reporting, is that the GHG emissions reported by telcos (indeed any company), do not include the emissions associated with employees’ existence (as consumers heating homes, eating food, going on holiday, personal use of cars etc). Since many AI initiatives are focused on improving employee productivity, even successful AI projects do not necessarily result in any GHG emissions reduction. Indeed, in many (potentially most) cases the ‘net’ impact on GHG emissions is likely to be negative – unless emissions reduction is also an objective.

Sustainability gAIn not pAIn

Despite the drawbacks, at STL we remain proponents of AI and don’t believe that integrating AI into the technology roadmap is inconsistent with keeping sustainability at the heart of the broader strategic roadmap. Rather, we argue that it’s imperative to place sustainability at the centre of any AI strategy. This means ensuring it is systematically included in decision-making, and setting up the mechanisms to capture the sustainable benefits of AI while also including the sustainability impacts in all evaluations.

In our upcoming report we will dive deeper into the relationship between network automation / AI and sustainability. We will explore how telcos can (and must) incorporate sustainability into their decision-making, from the initial stages of project conception through to the decommissioning phase, to ensure that sustainability remains a core consideration throughout the entire lifecycle of network automation and AI implementation.

Miran Gilmore

Author

Miran Gilmore

Senior Consultant

Miran Gilmore is a Consultant at STL Partners, specialising in edge computing and sustainability.

Download the sustainability insights market overview

This 31-page document will provide you with a summary of our insights from our sustainability research and consulting work:

Sharing is caring for the planet: Learn from Singtel’s sustainability approach

Sustainability is an area where collaboration is more important than competition. Recently, Singtel has shared its approach to embedding sustainability in the fabric of its organisation, ensuring it is taken into account in decision-making at all levels and operations. This article gives a glimpse of the findings of the case study developed by STL Partners.

How to successfully measure and reward sustainability benefits in a network’s lifecycle?

This article summarises the key findings from a recent CSP-exclusive virtual roundtable, focusing on how service providers can drive sustainable growth by embedding sustainability into their decision making and network innovation processes.

How to build and run edge data centres sustainably?

Sustainability is becoming increasingly crucial for data centres. This article explores how operators can build, and run, edge data centres sustainably.

Edge network APIs: the key to a stickier edge proposition?

This article evaluate the technical and commercial considerations operators must address to capture value from edge APIs.

Sharing is caring for the planet: Learn from Singtel’s sustainability approach

Sustainability is an area where collaboration is more important than competition. Recently, Singtel has shared its approach to embedding sustainability in the fabric of its organisation, ensuring it is taken into account in decision-making at all levels and operations. This article gives a glimpse of the findings of the case study developed by STL Partners.

How to successfully measure and reward sustainability benefits in a network’s lifecycle?

This article summarises the key findings from a recent CSP-exclusive virtual roundtable, focusing on how service providers can drive sustainable growth by embedding sustainability into their decision making and network innovation processes.