Cisco(CSCO.US) launches new AI networking chip! Targeting the large data center market, directly competing with Broadcom and NVIDIA

Cisco (CSCO.US) has announced a new type of network chip designed to accelerate data transmission within large data centers. The product may directly compete with related offerings from Broadcom (AVGO.US) and NVIDIA (NVDA.US).

Cisco states that its new Silicon One G300—a switch chip silicon capable of 102.4 Terabits per second—can power gigawatt-scale AI clusters to support training, inference, and real-time agent workloads, while maximizing GPU utilization and increasing task completion times by 28%.

The Cisco Silicon One G300 chip will be used to drive the new Cisco N9000 and Cisco 8000 systems. These systems are designed for hyperscale cloud providers, emerging cloud service providers, sovereign clouds, service providers, and enterprise customers.

According to the company, the Silicon One G300 chip, systems based on this chip, and supporting optical components will begin shipping this year.

The company specifically highlights that the new systems feature a 100% liquid cooling design, combined with new optical technology, which can help customers improve energy efficiency by nearly 70%.

Additionally, Cisco has enhanced its Nexus One data center networking architecture to make it easier for enterprises to operate their AI networks locally or in the cloud.

“As AI training and inference continue to scale, data movement becomes critical for efficient AI computing; the network itself has become part of the computation. This is not just about faster GPUs; the network must provide scalable bandwidth and reliable, congestion-free data transfer,” said Martin Lunde, Executive Vice President of Cisco’s General Hardware Group.

He added, “The Cisco Silicon One G300 powering our new Cisco N9000 and 8000 systems offers high performance, programmable, and deterministic network experience, enabling every customer to fully utilize their computing resources and securely and reliably scale AI in production environments.”

Last month, NVIDIA released its next-generation AI computing platform Vera Rubin, which includes several key network and infrastructure components. In June 2025, Broadcom began shipping its Tomahawk 6 series switch chips.

Meanwhile, Cisco also announced a series of features to help enterprises adopt AI securely while maintaining the integrity of agents and control over agent interactions.

These features include:

• AI Bill of Materials: Provides centralized visibility and governance for AI software assets (including model context protocol servers and third-party dependencies) to ensure AI supply chain security.

• MCP Directory: Discovers, inventories, and helps manage risks associated with MCP servers and registries across public and private platforms, strengthening AI governance.

• Advanced Algorithm Red Team Testing: Extends the scope of AI security assessments.

• Real-time Agent Guardrails: Ensures the security of agents and applications.