AI is usually described as a compute story: Larger models, faster GPUs, and data centers packed with accelerators. But as systems continue to scale, the real constraint is not computation but the communications of compute across hundreds, thousands, and even millions of processors.
