Discussion about this post

User's avatar
Scenarica's avatar

NVLink Fusion is the move that reframes the whole Broadcom question. Opening the interconnect to third-party XPUs, Marvell and MediaTek and even Trainium on the diagram, means Nvidia stops fighting to be the only accelerator and starts collecting a toll on everyone else's. The fabric becomes the moat rather than the chip.

Which is precisely where it pressures a custom-silicon story. Broadcom's pitch is that hyperscalers route around Nvidia with their own ASICs, but if those ASICs plug into Nvidia's fabric to scale, the escape hatch leads back through the tollbooth.

No posts

Ready for more?