VDE Outlines Germany’s Path to Becoming the Global Leader in AI Energy Efficiency

Large language models like ChatGPT offer immense potential – but they also demand enormous computing power and energy. According to the VDE, Germany could rise to become the world’s leader in AI energy efficiency by advancing energy-optimized hardware and neuromorphic computing.

Training the groundbreaking language model GPT-3 took a staggering 2,187 megawatt-hours of energy in 2020, according to researchers at the University of California, Berkeley – roughly equivalent to the annual electricity consumption of 380 German households, as this report shows.

The report outlines six approaches for sustainable generative AI (GenAI) development:

• Right-sizing AI models
• Energy-efficient systems
• Optimal resource utilization
• Intelligent data management
• Software-driven energy efficiency
• Powering operations with renewable electricity

A recent position paper published by the Information Technology Society (ITG) within the German Electrotechnical and Information Technology Association (VDE) points in a similar direction. It also proposes concrete strategies for how Germany can achieve global leadership in energy-efficient and sustainable AI and GenAI.

Maximizing AI Benefits Sustainably

“Large language models such as GPT-4 offer a wide range of opportunities currently being further exploited for economic benefit. However, it is crucial to address the associated challenges,” states the position paper. Only then, heise online quotes, “can the benefits of today’s AI drivers be maximized sustainably.”

For generative AI – including assistants like OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude – the development of algorithms and software products enabling breakthrough applications is only part of the equation. Equally critical are advances in energy-efficient hardware – and their tight integration with optimized software.

“There’s a global discussion about building more nuclear power plants to satisfy AI’s growing energy appetite,” explains Damian Dudek, Managing Director of the ITG and co-author of the paper. “We’re talking about making data processing more efficient and reducing energy consumption – so AI can be used sustainably.”

Learning from Nature – and Surpassing Moore’s Law

If Germany succeeds in establishing itself as the global leader in AI energy efficiency through such approaches, it could “unlock entirely new market potentials for societal benefit while preserving technological sovereignty,” heise quotes Dudek as saying.

The challenge lies in the fact that training large language models requires – as noted in the ITG paper – “enormous computational resources, currently delivered by dedicated multi-core systems.” These refer to high-performance graphics processing units (GPUs) operating within specialized software environments “that enable highly parallel processing with high throughput and high memory bandwidth.”

The traditional approach – shrinking transistor structures and packing more transistors onto smaller chips – is increasingly hitting the limits defined by Moore’s Law. Yet, according to the ITG paper, several promising new paradigms exist to overcome the constraints of conventional chip design.

Next-Generation Chips for Greater AI Energy Efficiency

Quantum computing and neuromorphic computing are identified as particularly promising avenues. Neuromorphic computing draws direct inspiration from nature – mimicking the brain’s architecture – to push data and signal processing into ever-smaller physical dimensions.

Germany should “stay engaged” in these fields – and assume global leadership. Recently, the ITG has also advocated leveraging photonics, especially to reduce data center energy consumption. As reported by heise, the German startup Q.ant unveiled an energy-efficient photonic AI chip in November 2024. Such solution-oriented innovations fully align with the vision of the VDE – the German Electrotechnical and Information Technology Association – which calls on the research community to tackle these challenges “for societal benefit, with foresight toward national and European economic development – and our technological sovereignty.”

Header Image Source: Pexels / Google DeepMind