Cool heads, big plans – AT&T ties private 5G to AI grid

By James Blackman April 9, 2026

Collected at: https://www.rcrwireless.com/20260409/internet-of-things/att-private-5g-ai-grid

AT&T is positioning private 5G as a necessary enterprise capability rather than a breakout revenue engine, tying its future instead to edge AI, spectrum pragmatism, and a broader “AI grid” vision with partners like Cisco and Nvidia.

In sum – what to know:

Hype reset – AT&T acknowledges private 5G has fallen short of earlier revenue expectations, but insists it remains critical for in-building coverage, security, and on-premise data control.

Dual strategy – AT&T is working with two vendors on large hybrid deployments and smaller standalone enterprise systems, while integrating management into existing IoT platforms.

IoT evolution – AT&T says private 5G works as an enterprise platform for GPUs, and a far-edge extension of its “AI grid” with Cisco and Nvidia – which offers global enterprise support.

Here’s what we wrote at the top of the newsletter yesterday (April 8) about AT&T’s rarely-heard position on private 5G – as scooped, last minute, from the cutting room floor following yesterday’s long write-up about its AI grid strategy with Cisco (which is also worth a read). It was offered as a sneak peak ahead of a fuller article today, as chopped up and served up here. It serves as a fine introduction; direct commentary from AT&T about its private 5G adventures follows at the bottom. Note, the original newsletter tease has been edited and expanded since yesterday. 

Rarely-heard; realistic, pragmatic

Private networks are important, says AT&T, but have not seen the revenue boom that was predicted a couple of years ago – as witnessed in Nokia’s weird about-turn, and also Verizon’s reset. So, AT&T is dialling back the hype; but it also maintains that private 5G is a necessary capability, if not a transformational growth engine. The firm has deals with two vendors, it says: a big one and a small one – for big projects and small projects. It won’t give names, although RCR has chased again. Given the uncertainty around Nokia, Ericsson and Celona might fit the bill.

Note: AT&T has a historical relationship with Nokia in the US, dating back to 2020, to sell the Finnish firm’s DAC and MPW private-network solutions (now on the chopping block) to enterprises with CBRS licenses. These were the industry’s salad days, of course, when Nokia was the chief agitator in the market – when CBRS was still new, and when MEC was still a thing. Coverage elsewhere, as recent as late 2025, suggests AT&T has relationships with both Nokia and Ericsson – plus with Microsoft, IBM, and Google (“to provide the various shadings of its offerings”). 

In the context of AT&T’s dual vendor strategy for complex systems with “hybrid” public/private support and simpler edge deployments with standalone functionality, plus its more liberal support of CBRS, it is likely AT&T is working with Celona, as well – for less-demanding setups, sometimes using shared spectrum. Celona said last year it had a deal with AT&T for US neutral-host deployments. (T-Mobile, its original partner, has since reneged on its outward support for CBRS – and since then also backtracked on its reneging to take a more pragmatic position.)

But as above, nothing is confirmed; this is just speculation and guesswork. (More hearsay, from a separate conversation yesterday about Nokia’s DAC/MPW proposition: Nokia’s enterprise customers have been calling its rivals about the longevity of their Nokia systems – or, more specifically, the management of them; so the story goes.) But back to the story: AT&T’s position on shared non-telco spectrum, locally licensed, is a clear point of difference. It would prefer enterprises to use its own frequencies, but it appears to acknowledge that CBRS has its place.

AT&T wants control and differentiation via its own spectrum, but it is pragmatic – more so than T-Mobile, and way more than Verizon. But one thing: it (AT&T Business, now incorporating AT&T IoT) wants customers to manage private networks using the same tools as public cellular. It specifically references IoT platforms like Control Center and Connection Manager, and implicitly suggests a bigger vision, where private networks are (inevitably from a telco-side view) an extension of the public general-purpose offering. Like 5G slices and AI grids, and whatever else

AT&T is positioning private 4G/5G around specific needs: coverage gaps where macro networks fall short; scenarios where DAS and Wi-Fi solutions do the same. Which is familiar, of course: the same drivers, as always – on-prem control, security, and privacy, plus deterministic performance. Its go-to-market is cautious, under-publicised. It has few/no public references (versus Verizon, say; even versus T-Mobile, lately). It claims deals with big automotive manufacturers, including at least one ‘hybrid’ private/public use case, plus various public sector / government sites. 

But, as with Ericsson (versus Nokia) to an extent, and per a rush of new analyst forecasts about a 2030 boom-time, AT&T’s timing may also be good.

Private 5G, from the horse’s mouth

But what does AT&T actually say? Well, here is an extended quote, from the horse’s mouth – taken from a wider conversation with Cameron Coursey, vice president of connected solutions, about the company’s AI-grid work with Cisco and Nvidia, as covered here. It explains most of the above commentary.

“Everybody had a view that [private networks] was going to be the next big thing to drive revenue and things like that. That’s not really been the case – as we saw with Nokia and, to some extent, with Verizon. But it is also very important. Because you need to provide the capability inside buildings and campuses – where you may not have the ability to provide macro network coverage, or where in-building antenna systems are not adequate, [or just] because the enterprise wants their data to be on-premise and private. So that is where private 4G and 5G comes into play. 

“We have split [the work between] one large private network provider [to deliver] large enterprise solutions that may need hybrid capability, in the factory and out in the wild, and another where the enterprise requires a smaller instance, and doesn’t necessarily need that hybrid capability.

“In both cases, the goal is to provide customers the ability to control their connections inside the private network in the same mechanism they use in the macro network – like with Control Center and Connection Manager in the IoT space. Making it simple for the customer is the goal.

“We are primarily focused on using cellular spectrum for this, but if a customer needs CBRS, [and] neutral host, then we work with that as well. But if we are going to deliver something from scratch for a customer, it would be [with] one of these private network [vendors], either on the higher end or the lower end using our own spectrum. We work closely with our local radio access network folks to get that done. 

“Generally, we haven’t been promoting this, [and] talking about customers. We cover some large plants with private 5G for some large automotive manufacturers. We are working on a very unique solution now that begins in the factory and then goes out to the wild. That will be a very interesting one when we are able to announce it. And we have a number with the public sector and government – which are being worked also. But I can’t give specific references at this time.”

Local extension of global AI grid

There is more from Coursey besides. The emerging conversation around private 5G networks in enterprise settings is, in many respects, a continuation of the long-running IoT and M2M narrative – albeit underpinned by a localized infrastructure layer. And really, the new AI-grid discussion, to put accelerated compute into the regional core network for certain IoT applications (like surveillance cameras), is the same, again. There’s a new piece of infrastructure to set down, but these are mostly high-end IoT use cases, right? “They are,” he says, and he adds some colour.

“A growing use case is AI at the private network edge – where, like a switched location on the Cisco network in a macro sense, you have a GPU in the private network for third-party developers and application providers. That’s an area we are very interested in.” Which is not a million miles away from what Nokia is cooking up with InfiniG, to place AI-RAN into private-5G extensions of neutral-host set-ups in US enterprises. Of course, there is a question, as well, about whether IoT/OT cases require GPUs, or whether CPUs will do it – per Nokia’s MXIE model, now up for sale. 

Coursey talks as well about the wider Cisco/Nvidia view of the AI-grid concept – that it is not just about local, regional, or national network infrastructure, but about global ecosystems. “Their view is more expansive – in that it is global,” he says. “It’s not just a telco solution for them. Others can be a part of the grid too.” But that international remit also extends AT&T’s local scope with multinational organizations – and it is only a skip and a jump, presumably, for it to tap partners to serve up the same telco-geared enterprise 5G and IoT to its customers in foreign markets. 

He comments: “We are looking at how we can use it for our own customer base. If multinationals need this capability at the edge of the network in the US and elsewhere, where we don’t offer a network, or where it is not appropriate for us to provide a roaming solution, then we would work with other operators that are part of the AI grid to support those customers.” In this way, the AI grid becomes a federated platform, extending beyond the footprint of any individual network provider – to be patched up on private networks, where public 5G, private DAS, and enterprise Wi-Fi fail.

Call it private 5G, call it edge AI, call it physical AI, call it whatever – in the end, it is like everything comes back to IoT. Coursey responds: “It is an evolution of IoT in our minds. The next phase of IoT has a lot to do with AI. That’s why we are making the plays we are. We have a strategy to provide a network-centric AI capability, and to continue to have a dedicated IoT core, which can scale and has security, and which can do a portion of inference – and to expand our deep enterprise and industry focus, which we’ve had so long, into this AI realm.”