To no one’s surprise, the global energy consumption of computers is on the upswing. In 2018, electricity consumption hovered around 1%-2%1, and fast forward two years to 2020, it more than doubled, clocking in at 4%-6%1<. By 2030, experts predict it’ll spike up to 8%-21%<1. (Burning fossil fuels is widely recognized as the primary source of greenhouse gas emissions. Consequently, any industry heavily dependent on electricity is deemed to have an impact on global warming.) If those stats weren’t enough to give you pause, here’s a stunner: the carbon footprint of cloud computing now surpasses that of the airline industry.
The widespread adoption of Artificial Intelligence (AI) and Machine Learning (ML) further exacerbates this energy binge. The need to ingest, deploy, and apply more data globally in real-time demands increased cloud and edge availability, translating to a surge in computational power. The price tag for AI/ML is hefty—not just in terms of erecting more data centers but also in the toll it takes on our environment.
Nowhere was this discussion more prevalent than at the recent COP28, the United Nations climate summit in Dubai, United Arab Emirates. AI took center stage on the agenda. While attendees were wowed by its potential to help fight climate change, they were equally worried about AI’s potential to guzzle energy and harm the planet.
AI-related carbon emissions
Researchers at the University of Massachusetts, Amherst, discovered that training several AI models can unleash a whopping 626,000 pounds of carbon dioxide. To put it in perspective, that’s nearly five times the emissions of your average American car over its entire lifecycle, from birth to retirement.
But wait, there’s more to the tale than just the upfront cost of training an AI model. There’s a whole lifecycle to consider. AI models demand a constant stream of feeding and tuning, and the hardware required to sustain these intricate algorithms throughout their lifespan only adds to the carbon load.
Technologists have a role in driving sustainable AI
The cry for more efficient and sustainable data centers was heard loud and clear at COP28. While activists seemed to embrace AI, they also called for more urgent climate change action. One particular group called We Don’t Have Time took a novel approach and issued a series of videos. Ironically, with the help of AI, the activists appeared as their future (middle-aged) selves imploring action now.
As technologists, we bear a significant responsibility for ensuring the sustainability of the software and systems we employ. The silver lining? There are methods to achieve this without sacrificing performance—ranging from coding efficient software to maximizing the efficiency of our data centers. Furthermore, today’s database technology allows us to substantially shrink server footprints, simultaneously reducing latency and boosting throughput.
Sustainable AI is good for business, too
If you’re still in the camp thinking that sustainability is merely a “nice-to-have,” here’s some good news: you can have your cake and eat it too. In other words, you can amp up operational efficiency while giving sustainability the spotlight. Enterprises around the globe are committing to net-zero emission targets – in fact, 66% of Fortune 500 businesses have done so3 – forcing them to reduce resources and energy consumption while profiting significantly more. In point of fact, Climate Impact Partners, a leading global carbon finance organization, conducted a recent study and found that organizations that reduced carbon emissions earned about $1 billion more in profit than those that didn’t.
Until recently, carbon accounting has mostly overlooked emissions from operating software platforms. However, with the substantial emissions stemming from the expanding IT infrastructure supporting AI, IT decision-makers must now factor in the impact of their software platforms and the emissions efficiency of their IT systems. Efficient software, which accomplishes tasks with fewer CPU cycles, demands less hardware and energy and translates to a significantly lower total cost of ownership (TCO).
At the core of IT infrastructure, the database plays a pivotal role, and Aerospike stands out for its ground-up design focused on efficiency. In many scenarios, our clients use up to 80% fewer servers compared to competitive solutions, achieving increased scalability, higher throughput, and reduced latency, all with a 20% lower total cost of ownership.
For IT decision-makers, it’s essential to factor in the environmental impact when selecting a technology stack. A comprehensive comparison by my solutions architect colleague, Behrad Babaee, scrutinizes similar database technologies, specifically Apache Cassandra and Aerospike. The estimated differences in CO2 emissions and costs reveal a staggering contrast: Apache Cassandra at 546 kgCO2eq versus Aerospike at 109 kgCO2eq, and a fraction of the annual cost—$10,848,384 versus $2,102,400.
Sustainable AI should not be an afterthought
The escalating carbon footprint driven by the relentless growth of AI is unmistakable, and the environmental impact cannot be overlooked. With our increasing reliance on technology, there’s a growing urgency to address the costs contributing to global warming. The responsibility to shape a sustainable future lies with us. Fortunately, viable solutions abound, encompassing efficient coding practices, optimized data centers, and groundbreaking database technology. The need for immediate action at the IT infrastructure level, especially within the database realm, is not merely a choice but an imperative. It’s high time to revolutionize our approach to technology, embracing efficiency and sustainability in equal measure for a brighter, eco-friendly tomorrow.
