This article is sponsored by Weka.
Artificial Intelligence (AI) is transforming our world by dramatically accelerating the pace of modern research, discovery and scientific progress, fueling an unprecedented wave of innovation.
In January, the World Economic Forum hailed AI as a key pillar of its “Global Growth Story for the 21st Century”, promising not only to contribute to global GDP but also to the world’s fight against climate change.
There is only one problem. AI contributes to the exponential annual increase in global power consumption and carbon emissions.
The ethics of AI have been a hot public debate for some time, but in general, it’s all about privacy issues, unintended biases, or the potential for malicious actors to use AI to cause chaos. It focuses on possible negative social consequences, such as sex. Little, if any, is mentioned about the impact of AI on the environment.
The inconvenient truth is that AI, one of the most powerful tools in the fight against climate change, is also one of the worst perpetrators. The climate crisis will only accelerate without AI intervening if we do not quickly curb AI’s insatiable energy demand and carbon footprint.
But it’s never too late. Reducing the environmental impact of AI can be done by rethinking how to manage the massive amounts of data and energy required to power AI in more climate-friendly solutions that can be implemented today.
AI’s Vast Energy Needs
AI and its siblings Machine Learning (ML) and High Performance Computing (HPC) are very energy and performance hungry. To maximize productivity and potential, these digital transformation engines must run on a nearly infinite supply of data and massive amounts of power.
Worse, traditional data architectures weren’t designed to deliver data smoothly and continuously, which only exacerbated the problem and caused delays and bottlenecks in the data pipeline. A recent study found that the graphics processing units (GPUs) that power AI and ML workloads are typically underutilized up to 70% of the time, sitting idle while waiting for data to be processed. It is As a result, it can take days or even weeks to complete the training of an AI model.
From a sustainability point of view, this is a big problem. This is because an underutilized GPU consumes an enormous amount of energy while idle and emits unnecessary carbon dioxide. Industry estimates vary, but around 3% of the world’s energy consumption today is thought to come from the world’s data centers, double what he did just a decade ago. The explosion of generative AI, ML, and HPC in modern enterprises and research organizations is accelerating faster than anyone expected.
Independent research firm Gartner said in October that “without sustainable AI practices, AI will consume more energy than human labor by 2025, significantly offsetting carbon-zero gains.” Let’s go,” he predicted.
Reducing AI’s energy consumption and carbon footprint are issues that we urgently need to address. As the adoption of AI and HPC accelerates at breakneck speed, it is no longer possible to ignore their impact on the environment.
Rethinking the modern data stack
Traditional data infrastructure and data management approaches are the main drivers of AI inefficiency. They are not equipped to support AI workloads simply because they were not built to support next-generation technologies such as GPUs that move efficiently at impossible speeds. .
The era of cloud and AI requires a complete rethink of the enterprise data stack. Leverage next-generation workloads such as AI, ML, and HPC seamlessly wherever data is created, resides, or needs to move, including on-premises, cloud, edge, and hybrid should be able to run. multi-cloud environment. To do that, it must be designed for hybrid cloud and software-defined.
Rethinking the data stack requires rethinking and reevaluating the data lake. Data lakes have proven useful over the last decade, providing a central location from which data can be accessed more efficiently without creating multiple copies, but the large data processing requirements of generative AI, etc. To power workloads, GPU data demands often exceeded what was available in the average data lake. .
It’s time to start redesigning the stack to support datasets that are orders of magnitude larger than what today’s data lakes can offer. On the other hand, we need to abandon data storage silos and opt for more dynamic systems that can pipeline data in continuous, steady streams to meet the insatiable data requirements of AI engines. . This is more than just a bigger and better data lake. Processes need to be implemented to better manage the massive amounts of data that serve GPUs, which are in constant demand. This prevents the GPU from being idle again, improving efficiency and sustainability.
Draw your way forward with the cloud
Another solution is to integrate the cloud into modern enterprise data architectures. In an increasingly decentralized world, incorporating a hybrid cloud approach makes a lot of sense. Moving even some applications and workloads to the cloud can have an immediate and significant impact on an organization’s energy and carbon footprint in the short term. Especially as more and more public cloud providers build hyperscale data centers that deliver ultra-efficiency, parts and power. All renewable energy sources.
According to a recent study by McKinsey & Company, “By carefully migrating to the cloud and optimizing usage, businesses can reduce their data center carbon footprint by more than 55%, or about 40 megatons of CO2. .”2Worldwide, this is equivalent to Switzerland’s total carbon emissions. “
Now it’s making a visible impact.
Take the first step towards making a positive impact
Reversing climate change requires global action on many fronts. Reducing energy use and greenhouse gas emissions related to AI and enterprise technology stacks is a challenge for CEOs, CIOs, CDOs, and other business and research leaders to reduce their companies’ carbon footprint and benefit their organizations and the world. is one of the ways we can support our sustainability goals. But this is only the first step.
Now is the time for the scientific, business, political and technological communities to come together to harness AI more efficiently and sustainably, balancing the obvious potential of AI with increased awareness of its environmental impact. It’s time to find a solution for
