AI workloads force network upgrades from AT&T, robotics firms and the Celtics

Gartner projects that by 2028, more than two-thirds of enterprise-managed data will be created and processed outside the data center or cloud, and speakers at an AWS event in Boston said enterprise networks are not ready for it. The consensus among presenters and attendees at Inside AWS Boston on Nov. 4 was that scaling AI requires more than new models and better data infrastructure: organizations must expand network capacity connecting data centers, cloud providers and workers in the field.

Three enterprises at very different stages illustrated the problem. They included a three-year-old robotics company whose business model depends on AI, an 80-year-old professional sports team deploying AI selectively across hundreds of employees, and a global telecom giant in the middle of a digital transformation.

What did AT&T say about its connectivity backbone?

AT&T, a 150-year-old company shaped by mergers and acquisitions, faces a legacy estate that complicates AI delivery. Luke Corbin, assistant vice president of digital transformation and artificial intelligence at AT&T, laid out the problem in an event presentation.

"We're a 150-year-old company that's been through a lot of M&A. You can imagine what our data infrastructure looks like. We have a lot of legacy application infrastructure," Corbin said. "How do you build all of that infrastructure and integration that's going to allow you to build AI even faster? What is your actual connectivity backbone?"

Field workers increasingly depend on AI agents, making reliable edge connectivity essential. Corbin said a field technician or salesperson dependent on an agentic layer must be able to connect over mobility or satellite. AT&T Business disclosed plans on Sept. 8 to integrate Amazon's Leo satellite network into its enterprise and public service connectivity architecture for remote areas.

"The whole thing becomes so critical. It's not like you missed an email. You can't actually do your job," Corbin said.

IDC's FutureScape: Worldwide IT Industry 2026 Predictions adds urgency: 50% of all enterprise AI inference workloads (the act of running a trained model on new data) will run on endpoints or edge nodes by 2030.

How does a robotics firm handle bandwidth constraints?

Luminous Robotics, founded roughly three years ago, deploys fleets of autonomous LUMI robots that build solar energy infrastructure alongside human crews. The robots use vision-language-action models for edge inference, which lets them operate without constant cloud communication. All data eventually returns to AWS for processing, but latency requirements vary by purpose.

"Some things are much more sensitive in terms of performance. Those get pushed to the cloud several times per second," CEO Jay Wong said in an interview with TechTarget. "We're bandwidth constrained. We're in the middle of nowhere typically. These deployments aren't like building a solar power plant in the middle of Boston."

The company splits its connectivity:

• High-speed LTE wireless carries error logs, safety information and other transmissions requiring higher frequency.
• Starlink satellite links handle workloads needing less frequent communication with AWS data centers but higher bandwidth.

Luminous batch-uploads robot training data to fine-tune AI models in the cloud using AWS Bedrock, then pushes over-the-air updates to robots. The company avoids updating during active deployments. "It's typically when there's something problematic, we would try to do it overnight, or if a release has a ton of great features, then we'll do it once every month or so," Wong said.

Bandwidth constraints will become more pressing as the company scales. Luminous announced two to three weeks before the event that it is sizing up manufacturing to produce hundreds of robots. It currently operates from an 11,000-square-foot facility. Wong was uncertain whether Luminous would partner with Amazon on networking but said he is open to exploring Amazon Leo. "We're fairly agnostic in terms of that requirement for our robots," he said.

Why is an NBA team expanding network security?

The Boston Celtics, an 80-year-old franchise with hundreds of employees, are scaling network infrastructure and monitoring to match a business on what CTO Jay Wessland called a "tremendous ramp-up." Growing data volume on and off the court is the catalyst.

"Every game is on television five different ways, and every stat can be delivered to you five different ways," Wessland said in an interview with TechTarget.

The team runs a layered security stack. "We have a significant bunch of cyber tools at the network level and at the user level that feed into our third-party security partner," Wessland said. The tools include ReliaQuest's GreyMatter, which filters security data using AI and forwards alerts to humans for review when appropriate.

"Sometimes it just auto-generates information about packets or things that are happening at the network edge," Wessland said. "We see those alerts and our third party sees them, so it's a significant amount of work at the network level."

The through-line across a telecom giant, a robotics startup and a sports franchise: AI processing at the edge has turned network capacity from background plumbing into a board-level dependency, and vendors from Amazon to ReliaQuest are positioning to carry that load.