InfiniBand Switch Sales Soar in Q2 2025, Yet Ethernet Remains Market Leader in AI Back-end Networks

InfiniBand switch sales surged in Q2 2025, driven by NVIDIA's Blackwell Ultra platform. However, Ethernet maintains the lead in AI back-end networks, with Celestica, NVIDIA, and Arista accounting for nearly two-thirds of sales. The report highlights the rapid adoption of 800 Gbps switches in AI back-end networks.

In Q2 2025, NVIDIA's Blackwell Ultra platform significantly boosted InfiniBand switch sales, reflecting the company's expanding influence in AI infrastructure. According to The Cube Research, NVIDIA's Q2 FY26 earnings call underscored the company's transformation into a leading AI infrastructure provider, with revenue hitting $46.7 billion and data center sales accelerating [1]. The report highlights that InfiniBand switch sales surged due to the rapid adoption of 800 Gbps switches in AI back-end networks.

The report further notes that while InfiniBand switch sales surged, Ethernet remains the dominant technology in AI back-end networks. Celestica, NVIDIA, and Arista accounted for nearly two-thirds of these sales, indicating a strong market presence and competition among these major players [1]. The rapid adoption of 800 Gbps switches in AI back-end networks reflects the increasing demand for high-speed, low-latency networking solutions to support the growing computational needs of AI workloads.

NVIDIA's Blackwell Ultra platform, which includes the GB200 and GB300 racks, has become a benchmark for training, inference, and reasoning workloads. These servers are deployed by hyperscalers such as OpenAI, Meta, and Mistral, contributing significantly to NVIDIA's Q2 revenue [1]. The platform's NVLink 72 technology allows entire racks to behave as unified computers, enhancing performance per watt and driving efficiency in AI inference and training.

In addition to the Blackwell Ultra platform, NVIDIA's networking solutions, such as Spectrum-X Ethernet and InfiniBand, are crucial components of its AI infrastructure strategy. These interconnects enable memory bandwidth amplification essential for reasoning AI, turning racks of GPUs into unified AI factories [1]. The company's annual GPU cadence, integrated networking, and full-stack systems are key factors in its strategy to maintain a strong position in the AI infrastructure market.

The report also highlights the growing importance of enterprise and edge AI, with NVIDIA extending its reach beyond hyperscale environments. NVIDIA's RTX PRO servers, Jetson, and Omniverse are key components of its strategy to support AI workloads in enterprise and edge environments, including factories and robotics [1]. The company's digital twin strategy, supported by Omniverse with Cosmos, is a notable example of its efforts to expand its AI infrastructure offerings.

In conclusion, NVIDIA's Blackwell Ultra platform has driven a significant surge in InfiniBand switch sales in Q2 2025, reflecting the company's growing influence in AI infrastructure. While Ethernet remains the dominant technology in AI back-end networks, InfiniBand's rapid adoption indicates the increasing demand for high-speed, low-latency networking solutions. NVIDIA's comprehensive AI infrastructure strategy, including integrated networking and full-stack systems, positions it well to capitalize on the growing AI market.

References:
[1] The Cube Research. (2025). Decoding NVIDIA's AI Factory Product Maze. Retrieved from https://thecuberesearch.com/decoding-nvidias-ai-factory-product-maze/