Powering the future: The energy shift for sustainable AI

A world where our energy needs are met without harming the planet, and where artificial intelligence (AI) not only boosts productivity but also bridges social gaps, isn’t a distant dream — it is a necessity, and it is within our grasp.

Climate change is reshaping our world in ways that cannot be ignored. To protect our planet and future generations, we must accelerate the energy transition at scale — shifting from fossil fuels to sustainable, renewable and low-carbon energy sources. The stakes couldn’t be higher, and the time to act is now.

At the same time, revolutionary AI technologies like generative artificial intelligence, agentic AI and artificial general intelligence have tremendous potential to drive the next wave of productivity improvements and social equity, ultimately improving the quality of life for people all over the world. AI is promising great change, driving a positive impact on labour productivity, access to healthcare and education, especially in disadvantaged communities and much more. AI is expected to add as much as $4.4 trillion annually to the global economy — the equivalent of the GDP of Japan today.

This potential comes at a cost, however. AI is a power-hungry giant poised to drive 160% increase in data centre power consumption. Supporting the immense computational requirements of both training and deployment of its models requires massive energy. Inference — the process of deriving output for multiple users from AI models — adds to the load, fueling data centre demand. Left unchecked, AI will add to the energy crisis our world is already facing.

How can we harness AI's promise of impact while maintaining or even accelerating the energy transition? We must keep net zero deployment at heart. While it’s not a low-cost transition, it is a feasible one. Ensuring sustainable AI along the journey will be as essential to the planet as it is proving to be for the people who call it home. To make AI a significant part of our future, we must look at how we reinvent the future responsibly.

To make AI initiatives more sustainable, we must address both sides of the equation. On the supply side, this means prioritizing the use of clean energy sources and building power efficient infrastructure. On the demand side, it involves optimizing AI systems to be more energy efficient.

First, we can prioritize using or training models run by efficient, net zero data centres that are powered by renewable and low-carbon energy sources, like nuclear. But why stop there? Leaders need to promote programmes that enable their organizations to capture carbon and store their emissions.

This also requires a completely new approach to our existing data centre networks. As-is, data centres require large footprints, have enormous energy consumption requirements and are concentrated in specific regions due to their networking efficiency needs.

By shifting to an all-photonics network (APN), we can connect data centres via photonics replacing electronics, which allow for direct data connections to improve efficiency. It’s like taking a direct flight for data instead of multiple stopovers, which is inefficient and time-consuming. The APN approach allows data centres to be located closer to renewable energy sources, reducing their impact on the electrical grid and high-density data centre locations.

The private sector is moving to enable and implement the move from electronics to photonics in data centres. For example, NTT's Innovative Optical and Wireless Network technology, which consists of an APN to enable data processing 125 times greater than today’s networks, enables near-instant transmission with latency reduced by 200 times and ultra-low power consumption to achieve 100 times more efficiency as compared to today.

As AI’s demand on our energy systems rises, especially with the next wave of inference-driven applications, infrastructure and resources will come under increased strain, reinforcing the need for sustainable approaches.

The demand side of AI brings its own set of challenges and opportunities. Efficient AI compute architecture and infrastructure, including the ones powered by photonics, will be key. And this can be supported by small and medium language industry- and domain-specific models that deliver high-quality outcomes but consume less energy. We also can and should be focused on innovative cooling technologies as well as leveraging the benefits of energy efficiency that can be gained particularly in countries with colder climates.

But we can’t manage what we don’t measure. This is why we need a standardized measurement strategy to measure our carbon footprint, as well as the impact of moving from fossil fuels to renewable energy sources to gauge improvements. It also means applying energy-efficient standards throughout the cycle – from AI's infrastructure, development, deployment and usage.

AI comes with tremendous potential for progress. It has the power to level the playing field, opening doors to opportunities that were once out of reach for many. At the same time, the ecological crisis we’re facing is more dire than ever before. We must harness the power of AI with vision and bold action — and responsibility must be at the core of everything we do.

From becoming a catalyst for the energy transition, to ethical energy usage, to AI governance, we’re just scratching the surface of what sustainable and responsible AI can and will mean for the world.