AT&T and Deutsche Telekom Build 6G Foundations on Today's 5G Networks

AT&T is already running an 18 billion parameter AI model on its radio access network, and Deutsche Telekom operates a multi-agent system that optimises its network for events without human intervention. Both operators told the TIP Fyuz 2026 conference in Seattle that these projects, running on today's 5G networks, form the practical starting point for 6G.

Rob Soni, AT&T's VP of RAN technology and chair of the TIP board, said 6G has already begun. In his view it is "a continuous evolution" rather than a disruptive upgrade.

What role does sensing play?

Integrated sensing and communication (ISAC) is "the big shiny object in 6G," Soni said. AT&T is not waiting for a new generation to test it. Several months ago the operator demonstrated sensing technology at AT&T Stadium, home of the Dallas Cowboys.

Outside defence and military applications, the commercial case remains open. "Who pays the bill?" Soni asked, calling ISAC more a business question than a technology one.

The engineering details point to a distributed future. Trials have placed servers beside the baseband unit, and AT&T expects to keep that design until ISAC and commercial processors make a single-server form factor possible.

"We can't imagine streaming all of the data from a radio head to some centralised location and processing it. It's just too much data," Soni said.

FirstNet partners and US government groups have expressed interest in ISAC. Soni sees a role for TIP in those conversations, because smaller vendors and operators risk being excluded. "The table's flat," he said.

Is 6G really about fibre?

For AT&T, 6G "doesn't just look like an interface change. It's a fibre build," Soni said, one that also ties together satellite, Wi-Fi, cellular and compute.

The numbers behind that framing are concrete. Soni expects about 1,000 data centres to be built in the US over the next two to three years, with 97% of them located within 10 miles of AT&T fibre.

Cell-edge performance is the harder problem. Soni cited Ookla data putting typical cell-edge speeds at 1 Mb/s to 5 Mb/s, against the 10 to 20 Mb/s that a self-driving car needs.

"So, what does that mean when a self-driving car goes to the edge of the cell? It drops its connection," he said. "Can it still drive? Can it still get where it needs to go?"

As AT&T works to free up capacity for sensing and explores new spectrum, it must also improve cell-edge performance. Soni described it as a big problem, one that will take creative thinking and, in the end, a lot of capital.

How cautious is Deutsche Telekom?

Thomas Lips, SVP group technology at Deutsche Telekom, took a more measured line. Sensing and compute get more attention than connectivity, he argued, while customers still need gains in uplink, capacity and at the cell edge.

He also pushed back on optimism around defence funding. Some people assume defence money will "fly into the window" for ISAC use cases. "I can guarantee you it is not that easy," Lips said. He urged operators to start instead with their own business and consumer customers.

On AI, Lips wants 6G to have its own AI domain, with support for multiple agents working together and making decisions automatically. Today's AI functions are "bolted left and right," he said, adding: "The data layer is not yet where it is supposed to be."

Truly integrated compute, sensing and connectivity will only arrive with 6G, he noted, though 5G-Advanced can carry much of the testing until then.

What comes next?

Alongside its foundation models, AT&T runs homegrown agents built on LangChain that trace the root cause of dropped calls. Soni expects the next stage, in which an agent spots a drop before any customer complaint arrives, to be in place next year.

Lips closed with advice for the industry: "bring early use cases, test with your customers." Both operators' roadmaps point to 6G work intensifying over the next two to three years as the US data centre build-out and 5G-Advanced trials converge.