‘Hybrid AI really gives you flexibility of model independence,’ says Cagnazzi, chief innovation officer at Presidio. ‘We think [hybrid] gives you this efficient use of tokens.’
Solution provider powerhouse Presidio is betting that the future of AI is hybrid, as organizations seek more flexibility around model usage along with reduced cost, Presidio’s Chris Cagnazzi told CRN.
“We do strongly believe—when we think about the future and our priorities in the future of AI—that it really is a hybrid world. It’ll continue to be very much a hybrid world,” said Cagnazzi, chief innovation officer at New York-based Presidio, No. 26 on CRN’s Solution Provider 500 for 2026.
[Related: 5 Things To Know On OpenAI Hugging Face Autonomous Hack]
The strategy builds on Presidio’s deep experience around enabling hybrid environments, including in building on-premises capabilities and providing tools for reducing cloud costs during the initial adoption of cloud infrastructure years ago—a shift that is now mirrored in the current AI surge, he said.
Crucially, “hybrid AI really gives you flexibility of model independence,” Cagnazzi said. “Everybody’s talking about token costs, token economics—we think [hybrid] gives you this efficient use of tokens. And it really allows you to keep some of your data closer to private [control].”
Ultimately, “we firmly believe that it’s totally going to be a hybrid world going forward with [AI],” he said.
What follows is more of CRN’s interview with Cagnazzi.
In terms of your biggest priorities right now, what would those be?
I look at 2026 so far as a year of accelerated releases. Every time you turn around, there’s something new. But when I think about the opportunities and the areas that Presidio is focusing on—our background is very much that we built out on-prem infrastructure and data centers and compute for years. And then we built out a fairly robust, cloud-forward digital business. We do strongly believe—when we think about the future and our priorities in the future of AI—that it really is a hybrid world. It’ll continue to be very much a hybrid world. There are some key components that we’re seeing in the marketplace when we think about hybrid. There’s model diversity. And, if you look back to when DeepSeek was first released, when R1 was first released, we really saw that there were models that could compete with the frontier model providers, all the big ones. But hybrid AI really gives you flexibility of model independence. Everybody’s talking about token costs, token economics—we think [hybrid] gives you this efficient use of tokens. And it really allows you to keep some of your data closer to private [control]. But we’re seeing it across the board. So we firmly believe that it’s totally going to be a hybrid world going forward with all of this.
How does repatriation factor into this?
Repatriation has been a big buzzword lately, outside of the cost of it all. What we’re seeing is repatriation is, it’s a bit of a cost thing, it’s a bit of a risk and latency lever. And it’s not really just about security by any means. But that was part of our reasoning to build out our AI Blueprint and introduce our Lighthouse consulting group, and how we tie that all together with our Presidio AI technology hub. That really helps our enterprise clients through that journey in a much more effortless way. So that’s a really big area. I think the other big area, if I think about it, is around security.
What are some of the biggest focus areas in security right now?
We’re working with a lot of clients across really three main areas. [One is] guardrails and control points for AI use. What are we doing around governing what data flows in and out of these models and credentials? Then [it’s about] how they’re using prompts, how our users are prompting, what they’re doing to access the models. And I think the second part of that is really around what we’re terming as ‘agentic identity.’ So think of that as the agents start acting on systems on their own credentials. So that’s a whole new identity world that’s surfacing that needs to be secured. And the third thing—and this does tie back to token visibility or model cost visibility—[is that] we’re spending a lot of time around visibility and governance across all AI solutions broadly. [That is] whether it’s commercial platforms, whether it’s Anthropic or OpenAI, or it’s an AI embedded platform in some existing tool—or things that we’re doing around our own AI accelerator platform that we’re delivering to customers.
Could you say a bit more about model cost and how Presidio is approaching helping customers with that?
Think of it as sometimes the best use cases are the ones that come from internal transformation. So we think about our own business, and we think about some of the models that we’re using. We built what we call AI Studio. It’s an agentic AI platform and it runs internally. It runs at scale. But it’s production-ready orchestration, model routing. A bunch of MCP connector builds. We have guardrails, observability, all these things that we built for our own internal transformation and to drive fluency within our workforce. So when we think of the costing side of it, back in 2017-2018, we rolled out a whole cloud costing platform, which we called PRISM. So we’re almost taking a very similar approach to that as we did to that costing platform. And that approach really is, first, we need a visibility tool. So if you’re using Claude, you’re using ChatGPT, you’re using some internal on-prem, what is your cost per month as far as consumption of those tokens in use? And then, how do we then provide not only visibility, but actionable items that come out of that visibility? Because I think that’s where customers struggle, and will continue to struggle, for a good period of time. So we’re building [it] out, and we’ve launched a couple of pilots on this with some key customers. We’ve developed a visibility platform—so it doesn’t matter what tool you’re using. We get by-user, how much they’re spending across the board.
Secondarily to that, we get an understanding of what are they prompting for? Is it those repeatable, easy things, or is it more around complexity—solving some kind of complexity? And then ultimately, what we do believe is that we’re going to build and create for our customers [a centralized platform]. When you ask for something, there’s a pane of glass that will then go out and match the best model to service your request. So I would say broadly, we’re doing a few pilots with customers on a lot of this [capability] that we’ve built internally, our own IP. But we do think that it’s an area that customers have to focus on. And we’ll continue to look around that cost optimization of the tokens and the spend.
What should customers be prioritizing as they develop their AI road maps?
Right now, you have a lot of customers that are focused on use cases. But I think [we should] flip the narrative a bit and think about knowledge worker fluency. I mentioned earlier what we did with our own AI Studio platform. Most organizations probably have a list of AI use cases that everybody’s thinking of. They’re trying to figure out what’s the issue on how we roll them out? They might be struggling with their own blueprint. But I think if they look at it and say, ‘What are the ways that we can build AI fluency across the organization that will help us develop the users, who then will deploy and use the use cases?’ And then—[after] you wrap the governance layer around that—I think you’ll have a much better outcome in the long run as a business, if you look at it from the fluency stage versus just the use cases.
