Nvidia has been officially anointed the company to beat in AI network fabrics, according to a new report from Gartner. The analyst firm’s endorsement validates CEO Jensen Huang’s boast at GTC that the chip giant is now the “largest networking company in the world.” Gartner praised Nvidia’s dominance in AI accelerators combined with its broad data center networking portfolio, particularly highlighting its proprietary NVLink protocol for delivering high performance and reliability in both scale-out and scale-up AI cluster networking. However, the report also warns that Nvidia’s throne may not be as secure as it appears.
Why It Matters: Network as the Hidden Bottleneck in AI Infrastructure
As AI workloads—especially large language model training and inference—scale to thousands of GPUs, network speed has become a critical performance variable. Nvidia’s push into both InfiniBand and Ethernet-based Spectrum-X switches reflects the strategic importance of inter-chip communication. According to IDC, Nvidia has already captured a 21.5% share of the Ethernet switch data center segment, with Spectrum-X seeing “significant traction” among hyperscalers and cloud providers building AI factories. This positions Nvidia not just as a GPU supplier but as a gatekeeper of the entire data center infrastructure.
Our Analysis: Cracks in the Crown
Gartner’s report identifies three key vulnerabilities in Nvidia’s position. First, the shift from AI training to inference, along with the rise of agentic use cases, could reshape the vendor landscape. Inference workloads are less sensitive to network latency, making open Ethernet-based alternatives more attractive than Nvidia’s proprietary protocols. Second, large hyperscale customers are increasingly demanding open Ethernet alternatives. Third, Nvidia’s product and go-to-market strategy is described as “lacking alignment” with the mainstream enterprise market. If Nvidia loses the hyperscale race to open alternatives, traditional networking rivals Arista, Cisco, and Marvell could seize the opportunity.
- Key beneficiary: Marvell was named by Gartner as the company to beat for AI data center optical connectivity. Huang himself called Marvell the “next trillion-dollar company.” Marvell’s portfolio spans DSP-based pluggables and analog optical components, and it plans to expand into co-packaged optics (CPO) for scale-up networking, aided by its $3.25 billion acquisition of Celestial AI.
- Risk factors: Marvell’s CPO technology is still emerging, and execution risks remain. Microsoft is hiring engineering teams for photonic interconnects, and alternative architectures like near-packaged optics (NPO) could challenge Marvell’s approach. Broadcom and Ayar Labs are also competitors.
Competitive Landscape: AMD’s CPU Throne and Networking Dynamics
Separately, Gartner named AMD the company to beat for enterprise AI server CPUs, citing its consolidation capabilities, agentic AI orchestration tools, and I/O bandwidth. AMD’s EPYC line has gained traction among VM vendors, including Microsoft Azure’s Da/Ea/Fasv7-series VMs. AMD also offers edge-centric Embedded 4005 series for inference near applications. This puts pressure on Intel in the x86 market and adds competition for Nvidia’s Grace CPU.
Uncertainty and Counter-Scenarios: Nvidia’s Counterattack
Nvidia is not standing still. Its Spectrum-X Ethernet switches are already gaining traction, and if it successfully launches inference-optimized networking solutions and realigns its enterprise GTM strategy, it could solidify its lead. Nvidia’s proprietary NVLink remains a high-performance barrier that competitors will find hard to match. However, if hyperscalers standardize open Ethernet through initiatives like the Ultra Ethernet Consortium, Nvidia’s reliance on proprietary protocols could become a liability.
Investors should watch: Nvidia’s data center networking revenue share (currently small relative to GPU sales), Spectrum-X revenue growth, Marvell’s CPO commercialization roadmap and customer wins, AMD’s EPYC market share trends, and the adoption pace of open networking standards like UEC. The AI networking market is still in its early stages, but the competitive landscape could shift significantly in the next 2–3 years.
#AINetworking #Nvidia #Marvell #AMD #Gartner #DataCenter #OpticalConnectivity #CPO #Ethernet #AIInfrastructure
Sources
- Nvidia crowned AI networking’s top dog, but its throne isn’t quite safe — SDxCentral · News coverage · Tue, 21 Jul 2026 18:52:26 +0000
Written by: XPLAIN AI Editorial Team · Reviewed by: XPLAIN AI Editorial Desk
This content was drafted with AI assistance based on publicly available sources and reviewed under XPLAIN AI's editorial standards.