Artificial intelligence is making organizations more productive, but those productivity gains come at a cost. Organizations are saving costs in some places, but increasing costs in others. Over time, these cost increases have become more visible and ‘out there’. Addressing these costs and securing the net benefits that AI brings will require several innovative solutions.
Hidden costs in the AI era
There is a growing consensus that artificial intelligence has “changed the calculus” when it comes to data volumes, data center capacity, and power grid demands. We’ve been hearing for years that the amount of data being generated around the world continues to grow. This continues, but AI’s data demands are blowing previous assumptions and estimates about future growth rates out of bounds. A frequently cited statistic is that 90% of the world’s data was generated in the past two years. This number is likely to be conservative, as AI requires large amounts of data. It is no longer possible to predict data growth with any degree of certainty, as new AI services and models completely change the landscape.
There’s also been a lot of discussion lately about the data center capacity needed to support AI. Since the advent of the current AI era, there has been a significant influx of investment into data center campus projects around the world, but the emergence of more efficient AI models has caused some investment to retreat in recent months. Having a good understanding of data center capacity is critical to meeting the accelerating growth opportunities presented by AI, but the rapid pace of change in this field makes it difficult to predict.
AI is almost always more energy intensive than standard web-based workloads. Although numbers vary, it has been suggested that ChatGPT queries consume up to 25 times more power than a standard Google search. Obviously, this is not an “apples to apples” comparison. ChatGPT queries generate more specific and customized responses to the questions asked. However, given that ChatGPT is estimated to receive tens or hundreds of millions of queries per day, this signals a gradual shift in energy requirements in the AI era.
AI energy intensity has a flow-on effect. According to the International Energy Agency (IEA), “power demand from AI-optimized data centers is expected to increase more than fourfold by 2030,” and power consumption in U.S. data centers alone “is expected to account for nearly half of the increase in power demand between now and 2030.” Over the past few years, grid capacity has been able to withstand increased demand due to improvements in management efficiency and the addition of capacity, including from renewables, but there are signs that the demand for power-intensive AI workloads is outpacing the grid’s ability to handle it.
It is important for AI adopters to recognize these challenges in AI implementation and incorporate them into the overall business case for AI. It is no longer enough to prepare a one-dimensional business case for AI based on delivering productivity gains. The full environmental, social, and governance (ESG) costs of an AI-driven future need to be understood and factored into strategy and decision-making.
Sustainable AI era
AI may be part of the problem, but importantly, it’s also part of the solution. The core of AI is to bring intelligence to different areas that can be used to increase efficiency. Where the real value can be created is in the ability to optimize numerous variables across IT, data center, and energy infrastructure and work in synchronization to efficiently run other AI workloads.
Several AI-based innovations are being introduced to data storage infrastructure. One of them, Dynamic Carbon Reduction, provides an algorithmic way to reduce energy consumption by switching the CPU into eco mode during times of low activity. The other type is “always-on compression”, where the system can switch from inline data reduction to post-processing, further reducing energy consumption and CO2 emissions by as much as 30-40%. Optimizing how data is stored can positively impact data center cooling and power consumption requirements.
Improving efficiency with data storage is just the beginning. Similar AI applications are needed for the servers that support AI workloads and the power systems that power data centers. Additionally, all of these different AIs must work together to increase overall efficiency.
Optimizing a single system is valuable, but when optimization extends to all the systems that enable the current AI era, in the form of an integrated layer of software and services for IT equipment, data centers, and power grids, it changes the current landscape and ensures that the AI era can continue sustainably for years to come.
This is still a work in progress, but we expect it to be an area of considerable innovation over the next three to five years. Partnering with a service partner with expertise and experience across these areas can significantly improve your organization’s ability to operate sustainably and cost-effectively in the AI era.
