Next-gen AI – the answer to cost and energy concerns?

The future energy demands of developing AI systems that surpass current human understanding remain difficult to predict. However, it is expected that algorithmic advances will primarily accelerate progress toward artificial superintelligence (ASI), rather than drastically cut energy usage. ASI refers to a hypothetical AI system whose intellectual abilities surpass those of any human –featuring advanced cognitive functions and reasoning capabilities far beyond our own.

The counterargument: An opportunity to democratize AI access

The rise of efficient open-weight models such as DeepSeek, Mixtral and Llama brings significant opportunities. These systems make sophisticated language models available to a broader range of companies and developers, thereby democratizing access to AI technology. Open-weight models are not only more affordable, but they can also be deployed without having to transfer data to a third-party provider. Additionally, they can be further fine-tuned using proprietary data.

There is growing evidence that highly intelligent models like GPT-4+ can be used to train smaller, more efficient models that come close to their trainers in terms of intelligence and capability. Much like applying relativity theory in practice doesn’t require Einstein himself, but rather a capable student who can work with the distilled insights, smaller models can effectively build on complex knowledge.

Not every application requires the most advanced AI capabilities. Many practical use cases can be served efficiently by lightweight models, addressing a current challenge: while powerful AI systems integrated into businesses remain expensive, smaller models often lack the intelligence needed for broader application.

The success of open-source models like DeepSeek challenges the assumption that “bigger is always better”, and encourages innovation in model architecture and training methods. It’s becoming increasingly clear that cutting-edge infrastructure isn’t always a prerequisite for innovation. By focusing on the right strategies, developers can rival proprietary advancements from industry giants.

Competitive pressure of this kind is healthy for the AI landscape. It drives improvements in both efficiency and performance, fostering a more dynamic and inclusive environment for technological progress.

The outlook: Dual-track development

The future of AI development is likely to unfold along two parallel paths. On the one hand, we can expect the rise of efficient models designed to make AI more accessible for everyday applications and increase its practical utility. Such models excel at understanding and applying existing knowledge – for example, in content creation, data analysis and solving routine problems.

On the other hand, advanced models will continue to push the boundaries of what artificial intelligence can achieve. Although these systems demand significant computing power, they focus on tasks that require genuine innovation and novel problem-solving approaches. In this way, they serve a role similar to research labs in the AI ecosystem – driving breakthroughs that may eventually lead to more efficient implementations.

While lower-cost models will make AI more widely available, the pursuit of new insights and capabilities will continue to rely on substantial computational resources. The future of AI won’t be about choosing between efficiency and performance, but about understanding when each approach is best suited to the task at hand.