Topic · L2 · Chips & Hardware
Seven generations, one trajectory: more memory, bigger racks.
H100 to Rubin Ultra in five years. The number to watch isn't teraflops — it's the coherent rack domain, the GPUs that behave as one. It's what makes a trillion-parameter MoE trainable at all.
Signal readings
- ⬤Custom-ASIC share shiftTPU, Trainium, AMD all chasing a 'better GPU'; Trainium currently doing best; needs switch scale-up for MoE inference2026-05-20 source →
Companies in this layer
Browse all companies →