- Broadcom Unveils Tomahawk Ultra, a new high-speed networking chip for AI workloads in data centers.
- The chip targets Nvidia’s NVLink Switch, boasting the ability to connect 4x more chips using a faster Ethernet-based protocol.
- Built on TSMC’s 5nm process, the chip is now in production and shipping.
- Originally designed for high-performance computing, it was adapted for AI as demand for generative models surged.
- Acts as a data traffic controller between chips packed in server racks, helping scale AI compute efficiently.
Broadcom already partners with Google on AI chips, positioning itself as a key Nvidia alternative in AI infrastructure.