The Structural Path to AGI : Why Compute Scaling Cannot Produce Intelligence

For more than a decade, the dominant assumption in frontier AI research has been that artificial general intelligence (AGI) will emerge from continued scaling of compute, data, and model capacity. This belief — the Scaling Hypothesis — asserts that sufficiently large models will spontaneously acquire the structural properties required for general intelligence. This paper challenges that assumption at the level of first principles. We introduce the Structural Path Thesis: AGI is not a quantitative limit of computation but a qualitative emergence of structural capacities. Scaling increases magnitude; intelligence requires architecture. We formalize the distinction between functional intelligence (performance profiles produced by optimization) and structural intelligence (architectures supporting grounding, self-models, world-model closure, reflexive reasoning, and cross-layer coherence). We then prove the Structural Gap Theorem, demonstrating that no amount of scaling within a fixed architecture can produce the structural capacities required for AGI, because those capacities occupy a strictly different representational space. The result: Compute scaling can yield superhuman performance but cannot produce general intelligence. AGI requires structural conditions that cannot be approximated by magnitude alone. This reframes the future of AI development: the frontier challenge is not “bigger models,” but the design of the structural substrates in which intelligence can arise.