A Lobste.rs discussion argues that small language models will replace large-scale LLMs. Proponents claim efficiency and local deployment outweigh the raw power of massive clusters. This shift threatens the dominance of hyperscalers relying on compute moats. Practitioners should monitor SLM performance benchmarks to determine when specialized, tiny models outpace general giants.