Broadcom Launches Tomahawk Ultra to Challenge Nvidia

On July 15, Broadcom announced the release of its latest network processor, Tomahawk Ultra, designed to accelerate AI data processing by interconnecting hundreds of chips working in unison.

This new processor is Broadcom's strategic response to the growing AI hardware demands currently dominated by Nvidia. Broadcom also plays a crucial role in manufacturing AI chips for Google, a subsidiary of Alphabet, and industry experts see Tomahawk Ultra as one of the few credible alternatives to Nvidia's powerful GPUs.

Tomahawk Ultra serves as a high-performance traffic controller, enabling rapid data transmission between dozens or even hundreds of chips within data centers, often across a single server rack. According to Ram Velaga, Broadcom's Senior Vice President, the chip directly competes with Nvidia's NVLink Switch but boasts quadruple the connectivity capacity. Unlike proprietary data protocols, Tomahawk Ultra leverages an advanced version of Ethernet to enhance data speeds while maintaining open standards.

Both Broadcom and Nvidia's networking solutions help data center architects build “expanded” computing systems by tightly linking chips within close proximity—measured in mere feet. This architecture allows software developers to efficiently harness the computational power necessary for cutting-edge AI applications.

Velaga confirmed that the Tomahawk Ultra is manufactured using TSMC's 5nm process technology and shipments have already begun. Broadcom’s engineering team spent around three years developing this chip, originally targeting high-performance computing markets. However, with the explosive growth of generative AI, Broadcom refined the processor to better support AI scalability, positioning it as a vital asset for next-generation AI infrastructure.