While headlines fixate on AI’s surging energy consumption, they’re missing an important story. The breathtaking pace of efficiency gains has the potential to fundamentally reshape global power dynamics and investment opportunities. For responsible investors, understanding this trajectory is essential to seeking to deliver sustainable returns while supporting a transition to a more resource-efficient future.
The evidence is already clear in the numbers, though curiously overlooked by most analyses. In just two years, AI models have become more efficient at an incredible speed – Microsoft’s Phi-3-Mini Ai model achieves the same benchmark performance with 142 times fewer parameters than Google’s Pathway Language Model 2. Meanwhile, the costs needed to power ChatGPT-like AI systems have dropped 280-fold in 18 months.
While total energy use continues to climb, the energy required per unit of AI processing is improving very dramatically — with newer chips from Nvidia claimed to deliver 30 times better performance while using 25 times less energy. The early years of any transformative technology sees large efficiency gains once economic incentives align.
Why efficiency changes everything
The efficiency revolution extends far beyond environmental considerations. Nations and companies aren’t competing in a static game with fixed rules – they’re racing to capture the steep portion of AI’s efficiency curve.
Consider China’s rapid advancement in AI development quality. The gap between leading Chinese and US models on key benchmarks narrowed from double digits in 2023, to near parity in 2024. Models such as DeepSeek, whose developers claim training costs of just $6m, signal a fundamental shift in competitive dynamics.
See also: Embodied AI: The race to build robots that think, move – and earn
As important as AI efficiency becomes for climate responsibility, it is also about geopolitical advantage. Companies and countries which are able to master efficient AI deployment will gain profound strategic and economic advantages. This is the primary driver behind the frantic race for AI sovereignty across major powers.
The market correction
Potentially, the most disruptive part of this AI revolution is how it will rewrite energy economics. Markets haven’t priced AI’s efficiency trajectory because they’re operating on overly simplistic assumptions about how energy demand grows. We’ve seen this movie before with technology — when solar panel costs began their rapid decline in the early 2000s, energy markets continued valuing fossil fuel assets based on historical trends. The result? A devastating repricing as markets recognised the changing landscape.
A similar repricing is on the cards for electricity markets, assuming AI’s energy demand will follow linear growth patterns. Hardware costs for enterprise AI are declining 30% annually, while energy efficiency improves 40% each year. For fiduciary investors, recognizing this trajectory early creates opportunities to both protect capital and capture value.
AI democratisation and beyond energy determinism
The idea that AI’s future depends on electricity supply misunderstands the economic forces at work. As energy efficiency gains are realised, the geographic constraints on AI development will likely dissolve. Today’s pattern where data centres consume 1.5% of global electricity, but 20% in Ireland and 25% in Virginia, will likely give way to a more distributed landscape.
See also: Two-thirds of advice firms plan to invest more in AI
As the energy needed per unit of AI work falls, regions that were previously considered unsuitable for AI development due to limited power infrastructure will become viable. So, capital allocated based on today’s energy consumption models, risks becoming stranded as efficiencies improve.
Bottom line
When investors discuss AI and energy impact with companies across the utilities and technology spaces, the focus should not only be on total energy consumption but also efficiency improvements. It is vital to ask companies across the value chain for more details on what efficiency metrics they are tracking beyond total energy consumption, or how might current infrastructure investments change with rapid improvements in processing efficiency?
For long-term responsible investors, AI efficiency represents a fundamental shift in how we evaluate technology investments. By engaging with companies, we can help align corporate strategies with both financial returns and environmental responsibility, precisely the systemic change that benefits our institutional clients.
