NVIDIA has quietly dominated this type of chip - 36Kr

NVIDIA has quietly dominated the data center Ethernet switch market, with its network business experiencing rapid growth. According to IDC data, NVIDIA became the top-revenue provider of data center E... Read original →

NVIDIA’s rise to dominance in data center Ethernet switches via Spectrum-X reflects a fundamental shift in AI infrastructure architecture. The technical ripple effect is profound: large-scale GPU clusters demand co-optimized NICs, DPUs, and optical interconnects, enabling NVIDIA—through its Mellanox acquisition—to vertically define the entire communication stack, including protocols. This integration erects formidable barriers for Arista and Cisco, whose horizontal models lack GPU-level synergy. On compliance, tightening U.S. export controls on 3nm EUV tools heighten supply chain risk, while reports of NVIDIA pressuring cloud providers to exclude rival accelerators have drawn scrutiny from EU and Korean antitrust regulators. Over the next 12–24 months, HPE and Huawei may accelerate in-house switch and SmartNIC development, but without a cohesive GPU ecosystem, scale remains elusive. The long-tail implication is clear: control over AI infrastructure is migrating from general-purpose compute to integrated 'compute-network-memory' platforms—and NVIDIA is positioning itself as the de facto OS for the AI data center.