How Mark III Won ‘One Of The Largest’ Nvidia AI Factory Enterprise Deals In The Channel

After TD Synnex revealed that Mark III Systems is set to deliver ‘one of the largest’ enterprise AI factory deployments in the channel, an executive at the systems integrator explained to CRN how its AI factory expertise and enterprise trends helped it seal the deal.

As hyperscalers and neoclouds begin to launch offerings based on Nvidia’s new flagship Vera Rubin rack-scale AI platform, one systems integrator is set to deliver what could be the largest deployment of such platforms for an enterprise customer in the channel.

This was revealed last week by TD Synnex CEO Patrick Zammit, who said that the Fremont, Calif.-based distribution giant and Houston-based systems integrator Mark III Systems recently signed an agreement “to support an Nvidia AI factory powered by Vera Rubin NVL72 systems.” CRN has learned the customer is in the Fortune 100.

Nvidia has said that Vera Rubin, which links 72 Nvidia GPUs and 36 Nvidia Vera CPUs to create an AI supercomputer, is expected to represent the fastest product ramp in the company’s history after it started shipping over the summer.

“This is one of the largest enterprise AI factory infrastructure deployments expected to be delivered through the channel bringing together the design, integration, deployment, Day 2 co-admin operations, financing and supply chain capabilities needed to operationalize a sophisticated Nvidia-based AI factory platform for a large enterprise,” Zammit said during the distributor’s third-quarter earnings call last Thursday.

The deal falls in line with the growing demand that TD Synnex is seeing “for partners that can simplify complexity and accelerate implementation” of next-generation platforms under evaluation by enterprises, according to the CEO.

To Andy Lin, CTO and vice president of strategy and innovation at Mark III, the customer win serves as another validation point for his company’s “extremely unique” capabilities in not just delivering and integrating systems but also in making those systems operational so that customers can derive useful applications for all of their users.

It’s this kind of expertise that has allowed Mark III to deliver around eight AI server clusters based on Nvidia’s DGX SuperPod designs with TD Synnex since 2023, according to Lin (pictured above). The company has also won Nvidia Partner Network awards multiple years in a row.

“You can only learn it through running clusters. This has been the biggest factor as to what’s uniquely driven us in the market. People hear about it, and they come to us because they need help,” he said in an interview this week.

While the executive didn’t disclose the specifics of what led Mark III to win the enterprise AI factory deal with what he described as a global Fortune 100 company based in the United States, he said the project aligns with broader enterprise AI trends.

One of the biggest issues is that enterprises are looking for cheaper alternatives to closed frontier models from the likes of OpenAI, Anthropic and other firms, according to Lin.

Many enterprises have “unleashed” these frontier models “on their general population and encouraged” their employees to use the models, he said. But they have not been prepared for how many employees are using the models “for everything,” which is causing the bills for such models to get “out of control very quickly.”

“Not only is it extremely expensive, but it’s extremely unpredictable in what they might be, which are two very difficult combinations for an enterprise, which relies on rigid budgeting, financial planning and forecasting,” he said.

This is leading enterprises to ask whether they need frontier models for every application; “many of them” are concluding that such models are only needed for 10 to 20 percent of use cases where high precision is required, according to Lin.

In turn, these enterprises see value in moving the remaining use cases to open models they can run in their own data centers. But the size and complexity of open models have grown exponentially too, which is driving customers like the Fortune 100 company to make big purchases on expensive AI infrastructure like Nvidia’s Vera Rubin NVL72.

Having the resources to purchase and house such infrastructure, however, does not come with any guarantees that a customer will get the best bang for its buck, which is why Mark III puts so much emphasis on the operationalization of these AI factories, Lin said.

“That’s actually the hardest part of this, and what’s prevented most enterprises from adequately adopting these AI factories at scale because people assume that it just works, and it doesn’t. It requires constant focus and solving issue after issue in the environment, and that’s what we’ve built out over the last few years,” he said.