Google integrates TPU v7/v8, Ironwood racks, and Apollo OCS into a unified fabric, shifting the scaling unit from servers to racks. This drives 800G+ optical module penetration above 60% by 2026. Supply chain focus has shifted to laser chips and MEMS capacity. Strategic decisions must now track 800G/1.6T transceiver deployments, alongside GPU/TPU shipments, to accurately gauge the AI compute cycle.