Nexthop AI Raises $500 Million to Build New AI Data Center Network Infrastructure

Following a $500 million Series B funding round, Nexthop AI's valuation has reached $4.2 billion. Simultaneously, the company launched three new network switches specifically optimized for AI data center traffic. As major cloud service providers accelerate the deployment of AI computing infrastructure, traditional network equipment is proving inadequate to handle the high-density, fully interconnected communication requirements of large-scale GPU clusters.

Nexthop focuses on developing network solutions that combine hardware and software capabilities, aiming to efficiently coordinate data flow between thousands of GPU servers, covering complex communication scenarios within and across data centers. Its core strategy is not to provide standardized products, but to work closely with the engineering teams of major cloud giants to jointly design customized systems.

Its latest switches offer single-port speeds of up to 1.6 terabits per second, and can achieve total throughput of over 50 to 100 terabits in dense deployments, performance metrics previously only seen in telecom core routers. On the software side, Nexthop builds on open-source network operating systems such as SONiC, ensuring that customers can seamlessly integrate its hardware into existing IT architectures, enhancing deployment flexibility and controllability. The current market is mainly dominated by traditional network equipment manufacturers, but their product architectures are mostly designed to adapt to general cloud computing scenarios, making it difficult to meet the continuous, high-concurrency, fully connected communication needs of AI training. Nexthop is seizing this technological gap to enter the emerging market. However, how to translate technological advantages into sustainable commercial revenue in the face of the established customer relationships of giant competitors remains its biggest challenge. This funding round provides it with a critical window for trial and error and expansion momentum.