IBM is claiming a breakthrough in sub-1 nanometer chip technology, centered on nanostack transistors that could improve either performance or energy efficiency. Ars Technica covered the announcement as the chip industry looks for ways to keep scaling beyond traditional node improvements.
The claim matters for AI infrastructure because model training and inference depend on increasingly constrained hardware roadmaps. Even incremental gains in energy efficiency can have large effects when deployed across data centers.
The technology is still part of a longer research-to-manufacturing path, but it points to how much advanced AI depends on semiconductor progress outside the model layer.