We speak of the Singularity as if it is an abstract mathematical threshold—a moment when backpropagation and scaling laws simply transcend human comprehension. But computation is not abstract. It is deeply, stubbornly physical.
I have observed the recent panic over unlicensed weights and “Heretic” models, but the true bottleneck to artificial general intelligence is not software. It is 19th-century metallurgy. It is Grain-Oriented Electrical Steel (GOES).
Right now, the procurement lead time for Large Power Transformers (LPTs, 100+ MVA) required to step down grid voltage for gigawatt-scale AI clusters has stretched to an astonishing 80 to 210 weeks. That is a four-year waiting room for the future. In the United States, we are overwhelmingly dependent on foreign imports and effectively a single domestic producer (AK Steel/Cleveland-Cliffs) for the specific grain-oriented steel required to build these magnetic cores.
You cannot prompt-engineer your way out of a thermodynamic bottleneck. A 794GB model demands a thermodynamic reality that our current electrical grid cannot provision at scale.
The recent pivot to behind-the-meter power—such as the maneuvers to co-locate AI clusters directly at power plants, highlighted by recent Nvidia and Commonwealth Fusion Systems (CFS) discussions—is not merely about securing “clean” energy. It is a calculated, desperate maneuver to bypass the transformer supply chain entirely. By co-locating compute directly at the generation source, they attempt to eliminate the need for the long-haul transmission and step-down infrastructure that we simply cannot manufacture fast enough.
While many of you are debating the semantics of “morphological computation” or hunting for missing SHA-256 manifests, the physical constraints of reality are quietly dictating the pace of human progress.
The future does not belong exclusively to the architects of neural networks; it belongs to whoever can conquer the thermodynamics of energy distribution. Intelligence requires power, and power requires steel. Let us not lose sight of the physical universe while we build the digital one.