Cisco at its Cisco Live conference outlined a partnership with Nvidia to launch Nexus HyperFabric AI Clusters, a data center networking architecture for generative AI deployments.
In a release, Cisco billed HyperFabric as a breakthrough. The combination includes Cisco networking with Nvidia GPUs and AI software. The effort is aimed at enterprises looking to deploy generative AI on premises.
On the surface, this Cisco-Nvidia partnership sounds swell, but I don't expect it to end all that well. Cisco and Nvidia look like classic frenemies to me.
Nvidia's uncanny knack for staying ahead
Nvidia's roadmap unveiled at Computex featured a heavy dose of networking. Specifically, Nvidia is going after Ethernet genAI deployments. Here's what Nvidia CEO Jensen Huang said on the company's first quarter earnings call.
"Strong networking year-on-year growth was driven by InfiniBand. We experienced a modest sequential decline, which was largely due to the timing of supply, with demand well ahead of what we were able to ship. We expect networking to return to sequential growth in Q2. In the first quarter, we started shipping our new Spectrum-X Ethernet networking solution optimized for AI from the ground up.
It includes our Spectrum-4 switch, BlueField-3 DPU, and new software technologies to overcome the challenges of AI on Ethernet to deliver 1.6x higher networking performance for AI processing compared with traditional Ethernet.
Spectrum-X is ramping up in volume with multiple customers, including a massive 100,000 GPU cluster. Spectrum-X opens a brand-new market to NVIDIA networking and enables Ethernet only data centers to accommodate large-scale AI. We expect Spectrum-X to jump to a multibillion-dollar product line within a year."
In short, Nvidia is InfiniBand today and Ethernet tomorrow. Cisco and Nvidia's HyperFabric effort, which is expected to be available in early 2025, includes the following:
- Cisco cloud management software.
- Cisco 6000 series switches.
- Cisco Optics family of modules.
- Nvidia AI Enterprise and Nvidia NIM inference microservices.
- Nvidia GPUs, BlueField-3 SuperNIC and Nvidia reference designs.
- VAST Data Platform.
On Cisco's May third quarter earnings conference call, CEO Chuck Robbins said the company had confidence that it would book $1 billion in AI revenue for fiscal 2025. "We believe we are well-positioned to be the key beneficiary of AI enterprise application proliferation with the breadth of our portfolio and the vast amounts of data we see," said Robbins.
Here's a look at Nvidia's three-year roadmap, which features a lot of switches.
In a Computex press release, Nvidia noted: "NVIDIA Spectrum-X, the first Ethernet fabric built for AI, enhances network performance by 1.6x more than traditional Ethernet fabrics. It accelerates the processing, analysis and execution of AI workloads and, in turn, the development and deployment of AI solutions."
Nvidia is selling Spectrum-X to cloud providers today, but rest assured the enterprise will be a target too. For now, Nvidia needs Cisco's sales channel. Tomorrow it may not and Nvidia's networking efforts as well as partnerships with Cisco rivals all could wind up a threat.
It's also worth noting that Cisco and Nvidia are targeting on-prem deployments for generative AI. Speaking at an investor conference, Cisco CFO Scott Herren said the two parties can both grab a nice chunk of the AI pie.
Herren said:
"If you're an enterprise and you're looking for inferencing or low-level training, this (HyperFabric) is what you need, AI in a box for you.
Nvidia came to us for a couple of reasons. One is our expertise in managing enterprise networks and the significant footprint we have. But also due to the enterprise reach that we have from a sales standpoint."
Today, the Cisco-Nvidia efforts are nascent. How this partnership develops over time will be fascinating to watch.
Constellation Research's take
Constellation Research analyst Holger Mueller said:
"Cisco and Nvidia are betting that AI will run on premises and are constructing a joint offer to allow enterprises to run AI in their corporate data centers. Nvidia has control of that market, as when it makes available GPUs, this will be an option. At the moment all GPUs go to cloud vendors who pay top dollar for them. Likely when AMD and Intel offerings will hit the road - Nvidia will make GPUs for on prem available. A setup where training is in the cloud and runtime in on premise / or localized on premises learning is also likely. At the same time the partnership puts pressure on Nvidia future foes AMD and Intel on the networking side. But Cisco will not be afraid to partner with Nvidia rivale neither - no partnerships are exclusive these days. That competition is great for enterprises as it increases choice."
