The general idea 2-die Rubin Ultra at ~half performance means ~2x more racks to deliver the same compute.
Double the racks = double the rack-to-rack links. Those links are pluggable transceivers. The pluggable window just extended by 12 months while volume doubled.
WINNERS (pluggable window extends volume up)
$CRDO — copper NVLink just got an extended lease on life. More copper racks before CPO arrives = more AEC/connectivity demand through 2027-28.
$AAOI Their CW laser self-supply advantage gets a longer runway.
$JBL named
$SIVE production partner for CW laser ramp. Pluggable first timeline confirmed.
$LITE
$COHR
Potential WINNERS (competitive gap opens for non-NVIDIA)
$AMD Nvidia tumbling does not mean AMD winning. AMD has the potential to catch up, but execution risk is high. Even Nvidia is struggling
$2454.TW MediaTek is the TPU silicon supplier. NVIDIA's gap is Google's window.
$MRVL If hyperscalers accelerate custom silicon while NVIDIA stumbles on scale-up, Marvell's Celestial custom programs gain urgency.
The bottom line: NVIDIA's CPO delay is bad for NVIDIA's competitive moat. It is good for the supply chain that sells into racks. More racks, more pluggables, more memory, more power, longer runway for every laser and transceiver maker to print revenue before CPO even arrives.