AT&T is tightening its ties with hyperscale cloud partners as it readies its network for the AI era – embedding last-mile fiber and 5G directly into AWS environments, pushing AI into the RAN, and expanding its edge ecosystem with Microsoft.
In sum – what to know:
Last-mile AI workloads – AT&T has launched a preview of AWS Interconnect with AWS, embedding fiber and 5G FWA directly into AWS workflows to support latency-sensitive AI.
AI-native cloud RAN – AT&T and Ericsson have demoed AI-native link adaptation on an Intel-based cloud RAN stack, claiming efficiency gains of up to 20% over legacy systems.
Enterprise AI edge – AT&T is extending Connected Spaces via Azure, positioning its network and edge assets as an AI-enabled architecture for enterprise insights.
As AI workloads become more distributed – stretching from branch sites and factory floors to metro data centres and hyperscale regions – operators are under pressure to collapse complexity between access networks and cloud platforms. AT&T has a deliberate strategy: fuse last-mile connectivity into cloud provisioning processes, virtualise and open up the RAN to AI-driven optimisation, and anchor enterprise edge services in hyperscale ecosystems.
It is betting that the next phase of telecoms growth will hinge less on raw bandwidth, and more on how seamlessly networks mesh with the AI engines they are designed to feed.
At MWC, AT&T has announced a “preview”, available from the second quarter, of its last-mile enterprise AI connectivity offering with AWS, which brings its fiber and 5G systems “directly” into AWS environments. The service, called AWS Interconnect, offers a site-to-cloud channel for AI workloads via AT&T’s 5G-based fixed wireless access (FWA) solution in enterprise offices, and its fiber metro and long-haul systems.
AT&T said the solution will reduce network complexity and latency for real‑time analytics, machine learning, and agentic AI use cases. The firm is building an “AI‑ready network”, it said, to scale network network performance for spiralling enterprise workloads. It said it will expand its fiber capacity to 1.6Tbps across key metro and long‑haul routes. “Connectivity is becoming ever more critical,” it said.
Last week, AT&T said it is working with AWS to provide high-capacity interconnect fiber to AWS data centers, and also to migrate AT&T workloads at its data center facilities to AWS Outposts, the hyperscaler’s edge-cloud hardware, installed on customer sites and in co-location facilities. AT&T is using agentic AI services from AWS to migrate network service enablement to AWS. AT&T will also look to use Amazon Leo, Amazon’s low Earth orbit network.
The forthcoming AWS Interconnect proposition, for last-mile access, “embeds AT&T‑delivered connectivity directly into AWS workflows”, it said. It “lays the foundation for the use of AI agents to monitor and manage the AI experience from the user to the cloud”, it went on. It is presented as a “streamlined, self‑managed approach” to help enterprises reduce network complexity while maintaining control of their extended enterprise network.
The service is geared for metro-level / last-mile delivery of latency‑ and data‑intensive use cases – such as real‑time analytics, machine learning, and agentic AI, the pair said. A statement said: “AT&T provides the last‑mile and connectivity linking enterprise locations to AWS, while AWS integrates the service into its cloud platform.”
Shawn Hakl, senior vice president and head of product at AT&T Business, said: “AI does not just need more compute; it needs flatter networks and faster connections. By bringing high‑capacity connectivity closer to cloud platforms, integrating the management of the networks directly into the cloud provisioning process and engineering for resiliency at the metro level, AT&T is helping enterprises… scale AI with confidence.”
Robert Kennedy, vice president of network services at AWS, said: “We’ve spent years reimagining how customers compute in the AWS Cloud, and now we’re bringing that same thinking to connectivity itself. With AT&T, we’re extending that transformation all the way to the last mile, giving customers a simpler, more resilient network experience that’s built for whatever comes next.”
Separately, AT&T and Ericsson claimed a “significant milestone” at MWC in Barcelona this week by demonstrating the vendor’s AI-native Link Adaptation on its target cloud RAN stack. They claimed “measurable gains” in terms of efficiency and experience. The test setup, using a Intel Xeon 6 system-on-chip (SoC), utilised AT&T’s frequency bands and propagation characteristics, and used AI models to respond to changing channel conditions in ‘real time’.
The demo showed progression from initial call – the first with portable Ericsson software – to feature capability. The pair said the tests achieved up to a 20 percent increase compared to legacy rule-based link adaptation, along with corresponding improvements in spectral efficiency and they serve as a collaborative benchmark for AI in RAN. Ericsson and Intel are jointly benchmarking AI models on commercial off-the-shelf hardware.
AT&T explained: “Open architecture allows AT&T to integrate innovations more quickly and flexibly run Ericsson software across multiple platforms, avoiding lock-in. This flexibility and portability enable faster scaling of new capabilities from the lab to the live network. First-mover advantage. AT&T is the first in the industry to demonstrate the portability of a commercial-ready AI feature on a Xeon 6 SoC- based cloud platform.”
Writing in a blog post, Rob Soni, vice president of RAN technology at AT&T, wrote in a blog post that AT&T is making “steady progress on its path toward a more open, programmable network”, The company has advanced “key elements that support open RAN readiness” over the past year, he said. “We’ve completed more than 50 percent of radio replacements, reflecting execution at scale and ongoing progress toward a more open capable RAN footprint.”
He cited work with Ericsson and 1Finity late last year to make the first call on its commercial network using Ericsson basebands and 1Finity radios, and its status as the first operator anywhere to deploy a third-party rApp to optimise its live network. More than 50 percent of AT&T’s traffic is carried on open-capable hardware, he said; the company is “live” with cloud RAN in two cities in the US, and on track to complete integrations and start scaling this quarter.
He said: “Open RAN readiness also depends on the software platforms that orchestrate, automate, and extend the network. We’ve [deployed] Ericsson’s Intelligent Automation Platform, which includes key open RAN components for cloud orchestration and a new RIC… This positions us… [also] to operate the RAN in a more programmable way, with a clearer framework for automation and policy-driven behavior. We’re advancing multiple fronts at once.”
On the business side, AT&T said at MWC its enterprise edge platform, Connected Spaces, is now available via Microsoft Azure. The platform uses Azure AI to bring together equipment, sensors, cameras, and devices into a single architecture. It enables enterprises to “turn near real-time data into actionable insights”, it said. It is geared for industries including retail, quick-service restaurants, hospitality, and more.
Cameron Coursey, vice president of connected solutions at AT&T, said: “Businesses want real outcomes from their physical environments — better operations, stronger safety and security, and improved customer experiences. Connected Spaces for Enterprise brings IoT devices, cameras, and analytics into a single, highly secure architecture and pairs that with Azure-powered analytics, so customers can turn near real-time signals into decisions at scale.”
