Unsustainable demands for AI are on the horizon. People will soon have to stop fake-caring about cloud sustainability and give a damn for real.
Cloud sustainability is a funny topic. It’s like dental care; everyone agrees it’s important, but who wants to discuss the particulars? It’s much more fun to talk about the new “shiny keys” of IT: AI and its endless possibilities. Cloud sustainability is so five minutes ago. However, like dental care, ignoring the issues will not make them go away.
AI growth was limited this year by the AI chip shortage. With the chip issue resolved, AI will accelerate the sustainability crisis due to its demands on our power grids that will outpace supply. Fortunately, we’re not at that tipping point—yet.
Looking in the wrong places
In my podcast discussions and videos on cloud and sustainability, which are becoming more frequent due to the rise of generative AI in the cloud, people tend to focus primarily on data center topics such as power and server efficiency. We need to understand that the power production and consumption side of things is not where the most improvements can be made.
Show me two data centers: one powered by dirty coal and another that relies solely on wind and solar energy, and I can show you how to make the coal-powered data center produce less carbon. How is this possible?
Simple. It’s not how you power the technology; it’s about how and why you use it in the first place. In other words, reevaluating the technology configurations to find areas for better resource optimization gives you better and more sustainable power consumption.
I’ll walk you through it
Don’t get me wrong, data center efficiency is essential. Technological advances have enabled data centers to reduce power usage through improved cooling systems, energy-efficient hardware, and renewable energy sources. Despite these efforts, focusing solely on decreasing the energy footprint of data centers does not fully address the sustainability challenges cloud computing poses. Indeed, it often distracts from the core problem.
A more holistic approach is to examine how cloud architectures are structured and optimized. The architecture of a cloud system determines its operational efficiency and resource utilization. By creating cloud systems that prioritize sustainability, businesses can optimize resource use and reduce waste. How much waste? My personal observation is as much as 500% more resources are used than needed, which roughly comes from power consumption.
Efficient cloud architecture involves several strategies, including server virtualization, workload optimization, and dynamic resource allocation. These strategies ensure that computing resources are used only as needed, reducing idle capacities and thus energy waste.
Artificial intelligence and machine learning technologies can also improve the efficiency of cloud computing. These technologies can predict usage patterns and automate resource scaling to optimize workload distribution. AI-driven analytics help identify inefficiencies and potential areas for energy savings, all without compromising performance.
Incorporating modular and flexible design principles into cloud architectures can significantly enhance sustainability. Modular architectures enable components to be scaled and updated independently, which helps minimize overprovisioning—an issue that often results in unnecessary energy consumption. Flexible architectures can dynamically adjust to varying workloads by reallocating resources to match demand, thereby effectively reducing excess capacity.
The cost of doing nothing
Why is cloud computing sustainability not better understood when it holds so much potential? In simple terms, many inefficient architectures were originally on premises and relocated to a public cloud provider. Now they are inefficient architectures on the cloud that burn too many cycles and need far too much storage.
Architectural problems are challenging and require many people like me to spend hours finding a better path forward for a specific technology stack. It’s easier for enterprises to kick the can down the road to avoid the cost and risk of fixing lift-and-shift problems. They would rather toss money at these systems as they operate on the cloud and not worry about wasting vast quantities of power. Ironically, I often hear system owners congratulate themselves on their “cloud sustainability” because a few points of presence for their cloud providers use renewables. Yay, you.
Here’s something to think about: If you must pay to fix something, that means there was a mistake. Many people struggle with the risk of potential backlash for admitting that. The benefits of fixing a system should drive its adoption, yet selling this idea remains a challenge.
I suspect that improving sustainability through cloud architecture will be overlooked or avoided until a more compelling fix comes along. Finops tools now monitor sustainability opportunities, and they’re getting better at spotting problems. They might be what finally rats out the bad lift-and-shift decisions that led to higher-than-needed carbon footprints.
Unsustainable demands for AI are on the horizon. People will soon have to stop fake-caring about cloud sustainability and give a damn for real.
