C
Cisco
2026-04-30
Architecture Shift Impact: Important Strength: High Conf: 85%

Cisco Launches Liquid-Cooled Network Switch, Extending Cooling Architecture to AI Infrastructure Core

Summary

Cisco has officially launched its N9000 and 8000 systems with direct-to-chip liquid cooling, extending liquid cooling from GPU servers to network switches. The product doubles bandwidth density and reduces energy consumption by nearly 70%, addressing the thermal challenges of high-power AI clusters. This move signals a shift in data center cooling architecture from component-level optimization to systemic redesign.

Key Takeaways

Building on its 2023 prototype of a 51.2T liquid-cooled switch, Cisco launched its next-generation product in February 2026. The new system, powered by the Silicon One G300 chip, delivers 102.4Tbps throughput, doubling performance in the same footprint. A key innovation is solving the thermal challenge for pluggable optics (e.g., 800G OSFP) with a pioneering 2×8 optics cooling design.
Cisco emphasizes its liquid cooling is systemic, with chip, optics, and cooling co-designed, not bolted on. The company validates solutions through its Engineering Alliances program and addresses regulations like Germany's EnEfG, focusing on waste heat capture and reuse. Cisco is already exploring immersion cooling and extending liquid-cooling to storage and power supplies.

Why It Matters

【Technology Breakthrough】This signals the acceleration of the cost-performance inflection point for data center cooling, shifting from servers/GPUs to full-stack infrastructure including networking and storage. The power density of AI clusters (projected to reach 1MW per rack by 2030) makes liquid cooling mandatory. As the core interconnect layer, network cooling efficiency directly limits the performance and energy efficiency of the entire AI infrastructure. Cisco's systemic design lowers integration barriers and technical risks for enterprises deploying full-stack liquid-cooled AI infrastructure.

PRO Decision

**Technology Breakthrough**
**Vendors**: Must treat liquid cooling as a core element of full-stack system design (chip, optics, fabric), not an optional add-on. Vendors failing to invest in systemic liquid cooling R&D risk marginalization in the high-power AI infrastructure market.
**Enterprises**: When evaluating AI infrastructure, treat the cooling architecture of networking equipment with the same priority as GPU servers. Demand full-stack liquid cooling solutions and clear integration paths from vendors, and initiate pilot projects within the next 12-18 months.
**Investors**: Monitor the substitution curve of liquid cooling technology as it penetrates from servers to networking, storage, and power supplies. Track growth metrics for liquid cooling components (cold plates, pumps, fluids, connectors) and system integrators. The market share of traditional air-cooling solutions will decline rapidly.
Source: Cisco Blog
View Original →

💬 Comments (0)