- AT&T has set an internal goal of reaching Level 4 network autonomy by 2030
- The operator is moving its RAN, core, IP and optical networks toward open, disaggregated and programmable architectures designed to support AI-driven operations
- Elbaz says AI traffic volumes have not exploded yet, but AT&T is already seeing signs of new traffic patterns, including more continuous and uplink-heavy demand
AT&T isn’t waiting for AI traffic to overwhelm its network before figuring out how to handle it. Instead, the Tier 1 is rebuilding multiple layers of its network around openness, programmability and automation, creating an architecture that Yigal Elbaz, SVP and network CTO at AT&T, said can support a much more autonomous network as AI adoption accelerates.
“Pretty much in every layer of our network, we are modernizing and moving to an open, disaggregated, programmable architecture,” Elbaz said during a fireside chat with Roy Chua, founder and principal at AvidThink at FutureNet North America in New York City last week.
That effort stretches from the radio access network (RAN) and standalone 5G core through AT&T’s IP and optical infrastructure. The company is also shifting cell-site management to its service management and orchestration, or SMO, platform, which Elbaz described as an important foundation for applications and “most importantly for the agentic era.”
Autonomy becomes a real target
AI has also changed AT&T’s thinking about autonomous networks. For years, Elbaz regarded Level 4 autonomy as “an audacious goal” that the technology wasn’t ready to deliver. But advances in AI have changed that calculation. AT&T has now made network autonomy a priority and set an internal vision of getting to Level 4 by 2030, he said.
The operator isn’t starting from scratch. Elbaz said AT&T already uses AI for tasks including cell-site energy savings and wireless network planning and design. It has also built its own network foundation model around network-specific numbers and KPIs.
But he warned against treating TM Forum's network automation scale scale as an industry scorecard. “Don’t focus on the number,” Elbaz said. “If you have a 3.9 or 4.1, trust me, in the real world, no one really cares.” What matters, he said, are customer outcomes.
There are still significant barriers between today’s automation and genuinely autonomous networks, particularly because telecom infrastructure leaves little room for AI mistakes. Elbaz pointed to data integrity, resolving conflicts when multiple AI agents reach different conclusions and, above all, trust in both AI models and their outcomes.
AI traffic starts to look different
The network also has to be ready for AI as a traffic generator, and here Elbaz drew a distinction between actual volume and early signals.
“The volume is not there yet,” he said. But AT&T is already seeing indications of how AI could reshape network usage, including workloads that run around the clock instead of following the busy-hour patterns networks have traditionally been engineered around.
Physical AI could produce an even bigger architectural shift. Robotaxis, drones and other machines send significant amounts of data upstream, challenging networks historically dominated by downstream traffic. Elbaz said AT&T has observed a robotaxi on its network with roughly 60% to 65% of its traffic traveling upstream.
As a result, AT&T is framing AI readiness as more than a wireless capacity problem. Elbaz pointed to the combination of wireless, network slicing, fiber, transport and cloud connectivity as a way of creating a more predictable end-to-end experience between physical AI devices and the models and data they need to reach.
Taking convergence into space
That idea of an intelligent, converged network also extends beyond AT&T’s terrestrial footprint. Satellite is another piece of AT&T’s broader converged-network strategy, said Elbaz.
AT&T and AST SpaceMobile announced that they successfully handed off a connection between AT&T’s terrestrial cellular network and AST SpaceMobile’s direct-to-device satellite network. The companies called it an industry first and said technology from Ericsson and Nokia supported the handoff. The milestone moves the companies closer to beta and eventual commercial satellite service.
Ultimately, customers don’t care which network is carrying their connection, Elbaz said. They simply want it to work. “It is our role to make sure that they are always seamlessly connected to the best available network,” he said.
He said AT&T and AST SpaceMobile specifically worked to smooth what has traditionally been a difficult transition between terrestrial and non-terrestrial coverage.
“We wanted to make sure that customers understand that their benefits and their seamless connectivity is top of mind,” Elbaz said. “So we already solved that.”
Read more about AT&T on Fierce Network:
The Five Nine: How AT&T and GSMA are building AI for telecom networks
AT&T beats Q3 estimates but SpaceX threat looms
AT&T sells Mexico wireless biz to Grupo Televisa and Fintech Holdings for $2B