The short version
The AI hardware fight has moved off the chip and onto the rack. Whoever owns the interconnect keeps earning whether or not the accelerator inside sells. That reframes every deal a custom silicon shop signs.
- 144 accelerators per fabric. d-Matrix Raptor chips will connect over a single NVLink fabric.
- Nvidia wins even without selling a GPU. It supplies the CPUs, switches, NICs, and Ethernet.
- It is paying partners to join. Nvidia invested $3.5B in MediaTek and $2B in Marvell.
- The race moved to the rack. Nvidia is defending the architecture under the chips, not just the chips.
Why this quietly deepens lock-in even as it adds choice. Read the full analysis.
Get the next one before it is old news
Independent analysis of cloud-native infrastructure, Kubernetes and data centre economics. No vendor spin.