Exclusive: Nvidia Networking Rival Cornelis Gives Channel Supply Reprieve For AI Buildout

The decision by Cornelis Networks to dedicate supply to the channel is seen as a differentiator when so many vendors are under pressure to keep hyperscalers fulfilled for AI buildouts, sometimes at the expense of channel partners.

As the AI data center boom continues to put a damper on hardware availability, an upstart rival to Nvidia in the networking space said it’s allocating supply for channel partners who are working on AI and high-performance computing projects.

Lisa Spelman, CEO of Intel spin-off Cornelis Networks, made the disclosure to CRN in an exclusive interview about the company’s growing channel roster, several months after it launched the 400-Gbps CN5000 family of scale-out networking solutions that it said has better performance and price-performance than InfiniBand- or Ethernet-based offerings.

“We want to put our partners in a position where they can ship and deliver system within weeks, not multiple quarters of delay. And so we’re continuing to prioritize them from a supply perspective,” said Spelman, a former Intel executive.

The Channel Is ‘Significant Part’ Of Cornelis’ Growth Plan

The move is part of the Wayne, Pa.-based startup’s channel-driven strategy, which she said is a “foundation of our go-to-market.” Cornelis generates close to 50 percent of its annual revenue from channel partners already.

“We actually see [the channel] as a significant part of our growth plan and our revenue, so you can see that the business opportunity matches the effort and investment we’re putting in as well,” said Spelman (pictured above).

Since launching its partner program last year, Cornelis has racked up 26 channel and OEM partners in the Americas, according to its website. These range from system integrators like Ace Computers and Microway to system vendors such as Dell Technologies and HPE.

Cornelis Partner Program Set Up To Be ‘Win-Win-Win’

Among the more recent channel recruits are TVAR Solutions and CTG Federal, two Washington, D.C.-area system integrators focused on serving the federal government sector. And up until recently, the startup had two authorized distributors in the Americas: Ingram Micro and TD Synnex. But on Tuesday, it announced a third: ASI.

Spelman said Cornelis has set up its partner program to be a “win-win-win” situation between partners, customers and the company.

“We want really well-trained partners. We want them to feel fully sales-enabled as well as support-enabled,” said the chief executive. “We put in structures around pricing and margin support so that they’re competing on an even playing field and then working very tightly with them on early-stage demand generation.”

How Cornelis Is Filling Multiple Gaps In The Channel

Sam O’Daniel, president of TVAR Solutions, told CRN that Cornelis is filling a gap in the market in multiple ways. First, the vendor’s Omni-Path architecture is designed specifically to handle “high-speed, transactional workloads that AI and HPC environments really need.”

“We saw a need for something that was really hyper-focused [on] that workload, which is what Cornelis has really been focused on,” he said.

The decision by Cornelis to dedicate supply to the channel is also seen as a differentiator when so many vendors are under pressure to keep hyperscalers fulfilled for AI buildouts, sometimes at the expense of channel partners.

“Our ability to deliver is our reputation,” O’Daniel said. “And while, in many cases, customers understand we’re in a tough position, being able to have an OEM that is adamant about ensuring the channel has access to supply and can deliver on timely basis not only sets us apart, but obviously it sets [Cornelis] apart as well.”

This is especially meaningful for TVAR Solutions’ federal customers.

“When you go into a conversation with a customer, it is definitely a nice thing to have in your back pocket when you say, ‘OK, you have this timeframe, and it’s mission critical, we know exactly where to go,’ and that is a good peace of mind when you’re going into a discussion with someone that you know has an urgent timeline,” he said.

Cornelis Is Opening Access For Partners Not In Nvidia’s Program

Cornelis is also giving a broader swath of channel partners the opportunity to work directly with another vendor that sells scale-out networking solutions for HPC and AI deployments.

This is seen as important by O’Daniel because his company, TVAR Solutions, is not part of the Nvidia Partner Network, as much as he wishes it was. And while his company sells Nvidia-based products through OEMs, it doesn’t get all the benefits that come with being part of the AI infrastructure giant’s partner program, which he said is closed.

“We don’t have the scale of engineers on staff that you really need to have trained up on everything and the certifications that are required [for Nvidia],” O’Daniel said.

Peter Chen, vice president of product management at ASI, told CRN that the new partnership with Cornelis “fills a pretty decent gap” in its line card for high-speed networking solutions for data centers running AI and HPC workloads.

While ASI sells Nvidia products through OEMs and other vendors such as PNY, the company is not an authorized distributor for the GPU giant.

“We do a lot of GPUs, but where we don’t have a large breadth of solutions is really on the networking side, and Cornelis [does] with their product offerings,” he said.

Omni-Path’s Openness, Flexibility Seen As Key

The TVAR Solutions and ASI executives see two other benefits to the channel in Cornelis’ Omni-Path architecture: support for a wide range of processors and the plan to support Ethernet compatibility with future products.

The latter is set to happen with Cornelis’ 800-Gbps CN6000 products, which is expected to launch this year and will allow Ethernet-based networks to access some of Omni-Path’s features through RDMA over Converged Ethernet, shortened as RoCE.

The following generation, the 1.6-Tbps CN7000 family, will then integrate standards from the widely supported Ultra Ethernet Consortium into Omni-Path next year.

O’Daniel said the move towards interoperability will not only give customers more choice in which vendors and products they use to build new data centers; it will also make a difference in the face of any hardware shortages.

“That flexible environment that is going to be very important for a lot of these different AI and HPC environments moving forward,” the system integration leader said. “In today’s supply chain ecosystem, you have to be able to work with what you can get at that time, and I think that’s going to be very important for the CN6000.”

Kent Tibbils, vice president of marketing at ASI, said Cornelis’ openness is timely because of how inference is becoming more mainstream across businesses.

“The ability to have that openness across the entire inferencing ecosystem is really a key advantage for our customers and for the channel,” he said.