PRZYSZŁOŚĆ AI - Pięć dróg rozwoju sprzętu i algorytmów sztucznej inteligencji: od Snapdragona do technologii post-krzemowych

This conceptual essay starts from the observation that Snapdragon-class systems on chip (SoCs), designed for minimal power consumption in battery-powered devices, can already run an entire voice pipeline with a small language model, whereas higher tiers of AI acceleration (e.g., Dragonwing IQ9 and IQ10) require external power and heat dissipation. This raises the question of whether SoCs remain the right choice or whether a mid-range graphics card becomes preferable. Building on this, the paper offers a structured overview of five paths for the further development of AI technology: (1) better power delivery and cooling, (2) continued improvement of manufacturing technology, (3) a shift away from silicon, (4) evolution of the algorithm, and (5) a complete change of algorithm. It argues that the main bottleneck for large language models is memory bandwidth and capacity rather than raw compute, and that algorithms and hardware co-evolve through mutual feedback. The conclusions are qualitative forecasts, not empirical measurements.