AT&T is increasing its reliance on Nvidia's artificial intelligence (AI) to power internal data processing, service-fleet routing, and employee support and training.

The carrier is the first to tap into Nvidia’s breadth of AI services. This includes Nvidia’s AI Enterprise software, which taps its Rapids Accelerator for Apache Spark; Nvidia’s cuOpt to support real-time vehicle routing and usage optimization; Nvidia’s Omniverse Avatar Cloud Engine and Tokkio; and Riva for conversational AI.

Andy Markus, chief data officer at AT&T, explained in an interview with SDxCentral, that the carrier’s AI push is designed to permeate throughout AT&T.

“We intend for everybody at AT&T to really lean in and to use AI to do their job better and more effectively,” Markus said. “We're really doubling down on what I call mobilizing the citizen data user, data scientist. … We're creating functionality where you don't have to be a professional data scientist to read AI. And our goal here is that we'll have smart people with the right tools to create AI to self-service in a way that we've never seen before and that we can grow the number of people creating AI – not just using AI, but creating AI – by 5x in the next two and a half, three years.”

Nvidia has for years been pushing its AI capabilities into cloud environments. This includes work with hyperscalers and virtual platform providers to support its AI services off premises.

Chris Penrose, who leads Nvidia’s telecom business development and spent more than 30 years as an executive at AT&T, added that the AT&T work highlights Nvidia’s work in expanding its extensive AI work to the telecom space.

“We're really leaning in hard around how can we really AI-power telcos and specifically in this area around transforming operations to generate cost savings, enhance customer experiences, and improve sustainability,” he said.

Nvidia launched its AI Enterprise software platform in early 2021. It packaged AI tools and frameworks that could run in virtualized environments.

The vendor initially touted AT&T’s testing of GPU-powered servers using Nvidia’s Rapids Accelerator for Apache Spark last year. At that time it noted the ability to process 2.8 trillion rows of information in just five hours, which was more than three-times faster and at a 60% lower cost than any prior test.

Markus noted that this performance gain has allowed the carrier to dramatically reduce cloud spend.

AT&T is also drawing cost savings through the cuOpt platform.

Nvidia unveiled cuOpt in 2021. It combines the vendor’s Rapids software with local search heuristics algorithms and metaheuristics such as Tabu search to analyze data and provide real-time optimization for route delivery.

Markus explained that AT&T was using cuOpt to help route its approximately 30,000 technicians on the most efficient driving path to reduce wasted mileage and help more customers during their workday. He added that trials have shown a dramatic reduction in the time to produce optimized routing schedules, which helps the operator more dynamically react to employee issues.

“Things change overnight,” Markus said, noting that “trucks break down, people get sick, emergencies happen. So we needed to be able to run it faster.”

AT&T is also using Nvidia’s Omniverse Avatar Cloud Engine and Tokkio platform to power employee productivity tools, and its Riva software to provide conversational AL. These provide interactive avatars that employees can converse with to answer internal corporate questions and to provide customer service recommendations.

Markus said that AT&T was also looking at these platforms to power digital twin modeling that can be used to help with network management.

“Going to market faster with how we administer the network, how we tune the network, how we built the network, understanding where it's good, where it's bad, this has so much potential for AT&T,” Markus said. “The building out of the network is super expensive, but just faster deployment and faster improvements for our customers, it's exciting for us and doing it digitally I think is a real game changer. We are super excited about this. I would say it’s a bit in the earlier stages, but one of the most exciting things that we're working on.”

Markus added that AT&T’s AI expansion timing is based around use case needs. He noted that the dispatch automation was already in place today, boosted by the moving of workloads to the more efficient processing paths within its data center environments.

“And then the other use cases are similar,” Markus said. “We just have to look at the tasks, determine the right type of analytics to apply, and then we look at our toolbox for the tools needed to execute the job.”