A provocative essay published by CleanTechnica revisits computer scientist Niklaus Wirth's 1995 article "A Plea for Lean Software" to question the long-term viability of neural-network-based artificial intelligence. The piece argues that the core threat to AI progress is not a looming shortage of electricity but the growing inefficiency of the software underpinning these systems.
Wirth's original thesis highlighted the trade-off between hardware advances and bloated code—a dynamic that has only accelerated with the rise of massive AI models. The essay contends that as AI models balloon in size and complexity, their software stacks become increasingly wasteful, consuming vast computational resources without proportional gains in performance.
This software inefficiency manifests in training runs that require clusters of specialized processors for weeks, and inference that demands ever more energy per query. The analysis suggests that if trends continue, the marginal returns on adding more hardware will diminish, potentially stalling AI's exponential progress curve.
Geopolitically, the critique carries implications for nations racing to dominate AI hardware supply chains. If the bottleneck is software rather than chips or energy, then policy focus on semiconductor fabrication may miss a more fundamental vulnerability.
The article does not offer new data or specific benchmarks, making the argument more a philosophical caution than a empirical forecast. It serves as a reminder that technological trajectories are rarely linear, and that overcoming hardware limits often requires rethinking software foundations.