Will data sovereignty dictate AI infrastructure investments?

“AI is expensive. It requires investment, and if you correlate that investment to actual use cases and governance that focus on the priority areas that will impact your profitability, your revenue, your cost structure, it is very likely that it will get a very strong return, and the infrastructure decisions and the places you invest will actually produce value as opposed to just cost,” explains John Roese, Global Chief Technology Officer & Chief AI Officer at Dell Technologies.

Dell’s AI Factory was designed to make the AI journey less complicated for customers. Since announcing its partnership with NVIDIA about two years ago, John Roese, Global Chief Technology Officer & Chief AI Officer at Dell Technologies explained that the AI factory could push the burden of making it work on Dell and NVIDIA and their partners so that customers could simply consume it and not have to invent an AI factory to get there.

For Roese, the end result from the AI factory is that its been working well with most customers feeling they do not have to build their own servers or even design their own stack. Instead, the AI Factory enables them to use the templates and reference architectures that has been created, and they're able to move significantly faster.

“If you go the AI factory route, it is much more rational because what you're buying and implementing is something that is guaranteed to actually work as opposed to an experiment. That is incredibly important,” said Roese in a media briefing before the Christmas break.

However, he also pointed out that there is the second part to AI development which depends on the infrastructure.

“What you choose to run on it matters as much as the infrastructure itself. Just because you have hundreds or thousands of GPUs, if you don't have governance, if you do not have prioritization, if you do not focus on return on investment, you may use all of those GPUs for just interesting work that does not impact your business. That is a very bad thing,” he said.

“AI is expensive. It requires investment, and if you correlate that investment to actual use cases and governance that focus on the priority areas that will impact your profitability, your revenue, your cost structure, it is very likely that that will get a very strong return, and the infrastructure decisions and the places you invest will actually produce value as opposed to just cost,” explained Roese.

Roese pointed out that when a company picks up the right use cases and builds the right infrastructure, they get production and end up with a positive ROI. He said the return on that investment could be 10 to 1, or in some cases it can be 30 to 1.

“What that translates into, we invest $1 and we get $30 of P&L impact by doing that. That is what every customer wants to do, and it is a combination of both governance and prioritization but also putting in place the right kind of infrastructure because if you cannot make the use case work because you created a snowflake, that is almost as bad as not having a use case at all. But if you get them both right, you get an economic impact from your infrastructure that is net positive, which is exactly what we want from infrastructure,” added Roese.

Sovereign AI and the partner ecosystem

According to Roese, the sovereign discussion is becoming significantly more interesting, especially since cloud service providers (CSPs) are a big piece of sovereign. Roese pointed out that when it comes to sovereign requirements, it is not usually the government itself building the infrastructure.

“They are usually partnering with CSPs in the country to help build them out. Macquarie is a great example in Australia where the infrastructure is built kind of with the government to build out the sovereign environment,” said Roese.

However, Roese also pointed out that most partners are not in the sovereign business. While some of them are, the reality is that there is probably not enough space for everybody. As such, the opportunity is going back to where customers need help, which is up and down the stack.

“If your expertise is around vertical industry, you can help customers find the highest value use cases. You can help them understand and clean their data tied to that particular industry. If your expertise is around infrastructure, you can help them accelerate their path to production by essentially giving them the additional capabilities which they may not have internally to build and operate an AI factory,” explained Roese.

The most interesting part for Roese though is agents. He explained that if partners are going to apply agents to any industry, it requires not just technical work, but it requires the ability to understand business processes, to understand how work is done and to be able to diagnose where the right place is to apply them.

“More importantly, because it's brand new technology, if you invest, and this is my recommendation to all of you, if you are a partner of Dell, you should be investing in your skill set, because if you know more about the AI technology and have more experience working with agents, you are intrinsically valuable to every customer, because if they know less, they need partners to help them know more,” he said.

“We do see that our partner base, it is to some extent contingent on them investing in their own skill sets. We are glad to help them do that, but if you know more than your customers, whether it's a government helping them with sovereign AI, or whether it's an enterprise trying to navigate agentic, you are in a very, very strong position. The one area that we do not recommend people do is stand still, because this is a fast-moving topic and expertise is valuable,” said Roese.

Put simply, Roese believes partners should learn faster than customers so that they can transfer that knowledge and help them move forward faster. They will appreciate it, and they will absolutely leverage their partners.