LAS VEGAS -- Cisco has made a significant move in supplying network infrastructure to support generative AI models, signaling a strong commitment to this burgeoning enterprise technology.
At the Cisco Live conference on Tuesday, the company unveiled Cisco Nexus HyperFabric, developed with AI chipmaker Nvidia. The prebuilt product will include Nvidia GPUs and AI software, Cisco Ethernet switching and a Vast data store. Cisco plans to make HyperFabric available for testing by select customers in the fourth quarter.
When public cloud use began growing in the enterprise, Cisco failed to participate on the infrastructure side, Cisco CEO Chuck Robbins said. The company doesn't intend to make the same mistake with AI in the enterprise data center.
"When the cloud transition happened, we were on our heels," Robbins said during a meeting with reporters and analysts. "We're leaning in on the AI transition, and we're well prepared to actually help our customers."
Cisco has partnered with Nvidia in the AI infrastructure market at a time when some customers are still trying to understand how GenAI can help them manage their networks.
We're leaning in on the AI transition, and we're well prepared to actually help our customers.
Chuck RobbinsCEO, Cisco
"I'm not 100% sure how we think about it because there hasn't been enough in the infrastructure space to really touch, feel and get a true understanding," said Michael Blake, vice president of IT at PTC. PTC develops computer-aided design applications and product lifecycle management software.
Nexus HyperFabric includes a new Cisco 6000 series switch that delivers Ethernet speeds of 400 Gbps and 800 Gbps. The product uses Nvidia's H200 Tensor Core GPUs and the Nvidia BlueField-3 SuperNIC and data processing unit to accelerate security, networking and data access.
Nvidia's AI Enterprise software with the NIM inference microservice handles the deployment and fine-tuning of the model. Vast Data Platform is a network-attached storage system built to support AI computing architectures.
What's unclear is how HyperFabric will help customers like PTC.
"We're waiting to see it and understand it," Blake said of the GenAI models that would run on HyperFabric. "The hype is at such a high level that we need to bring it down to get to the different use cases and then apply those in each of the business units. In our case, that would be within the infrastructure."
Cisco executives answer questions from reporters and analysts at the Cisco Live conference. From left to right is Chief Communications Officer Maria Poveromo, Chief Customer Experience Officer Liz Centoni, CEO Chuck Robbins, President Gary Steele, General Manager of Security and Collaboration Jeetu Patel, and General Manager of Networking Jonathan Davidson.
Cisco as an AI model infrastructure provider
Nvidia has partnered with a number of infrastructure providers recently to expand their AI offerings. The company approached Cisco to help develop HyperFabric, Robbins said. Cisco's large enterprise customer base and extensive reseller network make it a valuable partner for AI product developers.
"Many of these partners don't have enterprise global markets, and they actually are going to lean on us, which gives us an opportunity to play a very strategic role for enterprises as they think through how to deploy a lot of this emerging technology," Robbins said.
To help bolster AI development, Cisco launched on Tuesday a $1 billion startup investment fund. Startups that had received money from the fund included Cohere, Mistral AI and Scale AI.
What Blake was particularly interested in was not just GenAI, but also Cisco's strategic plans for Splunk, a big data analytics company that Cisco acquired in March for $28 billion. Cisco has already integrated its security data into Splunk, and Blake is eager to see real-time machine data for networking, servers and storage also being incorporated with AI.
How far Cisco can take AI into the infrastructure it provides remains to be seen, said Jim Frey, analyst for TechTarget's Enterprise Strategy Group.
"The overarching promise of AI is looking across all of the technology domains and correlating data so that you can do root cause analysis across multiple parts of the stack," Frey said.
To have AI analyze a problem and then deliver a conclusion that network operators can go back and verify before moving forward is a complex scenario to create with technology.
"That's really hard because you're bringing even more data together," Frey said. "Also, every customer environment is a snowflake. They're all unique."
In other Cisco Live news, Cisco extended its network visibility platform ThousandEyes deeper into public clouds, on-premises networks, and Cisco Meraki Wi-Fi and LANs.
ThousandEyes maps and monitors infrastructure across the internet to troubleshoot problems that affect applications running on the WAN.
The latest capabilities include the following:
Extending the platform to provide topological mappings of a company's AWS environment, including service connectivity, configuration changes and traffic characteristics.
Pushing ThousandEyes deeper into on-premises networks by collecting and correlating traffic flows within synthetic measurements, which involve simulating user interactions with network services and applications to identify performance issues.
Adding Meraki Wi-Fi and LAN telemetry and device information to the platform to provide real-time and proactive monitoring of network connectivity from laptops and desktops.
Antone Gonsalves is an editor at large for TechTarget Editorial, reporting on industry trends critical to enterprise tech buyers. He has worked in tech journalism for 25 years and is based in San Francisco.
