NVIDIA's Spectrum-X platform crossed 21.5% market share in Q1 2026, displacing Cisco and Arista at the top of datacenter Ethernet switching by revenue , a move that extends its AI infrastructure grip well beyond GPUs.
For decades, if you were wiring up a serious data center, you called Cisco or Arista. That's no longer the opening assumption. According to IDC data for Q1 2026, NVIDIA has claimed the number one position in datacenter Ethernet switching by revenue, with its Spectrum-X platform generating $2.1 billion in quarterly revenue, up 192.7% year over year. Two years ago, NVIDIA held less than 4% of the datacenter switching market. It now holds 21.5%. Arista sits at 20.7%. Cisco, which still dominates the broader Ethernet switching market including campus and enterprise, has a 17.8% share of the datacenter segment.
The overall Ethernet switch market hit a record $15.4 billion in Q1, with the datacenter segment alone reaching $10 billion, up 61% year over year. Those are extraordinary figures, and most of the acceleration traces back to one thing: the buildout of large-scale GPU clusters to train and run AI models. That infrastructure isn't being designed the way enterprise networks were. Cisco and Arista built their franchises by selling networking platforms into enterprise and cloud environments where general-purpose switching was good enough. NVIDIA is doing something structurally different.
Spectrum-X isn't just a switch. It's an end-to-end networking fabric built specifically for GPU clusters, combining Spectrum Ethernet switches with BlueField DPUs and NVIDIA LinkX cables into a tightly integrated system that handles lossless networking, adaptive routing, and congestion control at scale. NVIDIA says the platform delivers roughly 1.6 times better AI performance compared to standard Ethernet configurations. Its latest BlueField-4 DPU pairs a 64-core Grace CPU with a ConnectX-9 NIC, doubles bandwidth over its predecessor, and is designed to support clusters four times larger than BlueField-3 could handle.
The key word in all of this is integration. When hyperscalers and cloud providers are standing up AI factories , the term NVIDIA actively promotes for its GPU cluster deployments , they're not mixing and matching components from different vendors. They're buying a coherent fabric where the switches, the DPUs, and the cables are all tuned together and sold as a system. That's a different procurement conversation than the one Cisco and Arista have historically won.
Arista isn't standing still. The company reported 37.3% year-over-year revenue growth in Q1 2026 and is pushing its 800G technology hard, with its HyperPort hardware targeting AI workloads. Arista has also pursued partnership arrangements with NVIDIA to offer verified fabrics that span DPUs and switches, which is both a defensive hedge and an acknowledgment that customers are demanding NVIDIA's stack. Cisco, meanwhile, grew its datacenter Ethernet revenues just 16.9% year over year in the most recent comparable quarter , fast by any historical standard, but slow against a market running at 61%.
The real concern isn't market share
Frankly, the market share number matters less than what it represents structurally. NVIDIA has already locked in the GPU layer , there is no credible alternative for large-scale AI training at the frontier. It controls the CUDA software ecosystem, which has proven remarkably sticky even as AMD, Intel, and various custom silicon efforts have tried to chip away at it. Spectrum-X is the next ring of that moat. If hyperscalers and cloud builders are running NVIDIA GPUs connected by NVIDIA networking managed by NVIDIA DPUs, they're running an NVIDIA factory, not a multi-vendor infrastructure. The switching revenue is real, but the strategic leverage is what should concern Cisco and Arista most.
As Data Center Knowledge reported on the IDC findings, the purchasing model itself has shifted. Traditional switching vendors sold platforms. NVIDIA is selling an architecture. Buyers aren't evaluating Spectrum-X switches the way they'd evaluate a comparable Arista box , on port density, latency, and feature sets. They're asking whether the fabric is optimized for what NVIDIA GPUs need, and the answer is that it was designed from the ground up for exactly that. Everything else is retrofitting.
None of this means Cisco and Arista are finished. Enterprise networking, campus deployments, and the parts of cloud infrastructure that don't involve GPU clusters are still enormous markets, and both companies are growing into them. Arista in particular has built a strong position with the big hyperscalers outside of AI-specific workloads. But the fastest-growing and highest-capital-expenditure segment of datacenter infrastructure right now is AI, and in that segment NVIDIA has just planted a flag that nobody expected it to reach this quickly. The question from here isn't whether NVIDIA belongs in networking. It's whether networking vendors can stay relevant in AI.
Also read: Oracle cut 21,000 jobs in a year and put the blame directly on AI • The United Nations is building its own cloud and you should be paying attention • The AI boom is keeping inflation alive and interest rates higher than founders want