Can AI rescue 5G?Can AI rescue 5G?

In the early days of 5G, there was plenty of hype around how the technology might cultivate related opportunities including private wireless networking and edge computing.

But those opportunities haven't yet played out like many had hoped. Private wireless networking over 5G has been gaining steam, but it remains a relatively small piece of the overall 5G pie. 

And edge computing is perhaps an even bigger disappointment. Most edge computing efforts in the telecom industry remain in the planning or trial stages.

Now, though, some 5G players are hoping demand for AI services will change all that.

The latest: Verizon on Tuesday announced it will work with AI superstar Nvidia to inject Nvida's AI Enterprise software platform and its NIM microservices into Verizon's private 5G and private edge computing offering.

It's all part of Verizon's efforts to leverage 5G as a way to sell more than just connectivity.

"We don't just offer connectivity; we deliver cutting-edge solutions that empower our customers to anticipate and adapt to change," Srini Kalapala, Verizon's SVP of technology and product development, explained in a release.

Verizon isn't alone in using AI to put a shine on its existing services. For example, content delivery company Fastly recently took the wraps off its AI Accelerator, which leverages the company's edge network to give enterprises faster and cheaper access to AI services like OpenAI's ChatGPT.

But there is plenty of skepticism that AI can supercharge demand for struggling services like private wireless and edge computing.

We've been burned before

"We're entering a new era of telco #edgecomputing hype," Dean Bubley, analyst with Disruptive Wireless, wrote in a brief post to social media. "It is essentially a replay of the MEC [multi-access edge compute] debacle from 5-6 years ago, but this time updated for the AI era."

Bubley argued that there are plenty of questions over whether enterprise demand for AI services will drive computing business to telecom operators. He wrote that most developers don't want to spread their computing workloads across sites run by different telecom operators, and that most telecom network engineers won't want to share their computing resources with third-party customers.

"Telcos have demonstrated only a minimal role in edge computing services, either as localised low-latency cloud computing suppliers, or even in terms of just offering colocation space in exchanges, or mobile towers / aggregation sites," Bubley wrote.

Indeed, most telecom operators haven't even managed to improve the latency on their networks. That's noteworthy considering lowering the latency of AI services is one of the key selling points in most telecom AI sales pitches.

William Webb, an independent consultant and former Ofcom executive, wrote on social media that "5G latency in the US is only a little better than 4G," despite years of hype about the lower latency that 5G connections would support.

Webb acknowledged that private 5G networks can support 10-15 millisecond latency – much better than the 30-50ms latency on public networks – but he said that's still not very compelling.

"While 10 ms is better than public 5G, it is still worse than a decent Wi-Fi network. Those who need very low latency might want to look elsewhere than 5G," he argued.

Undeterred

But Verizon, T-Mobile and others remain undaunted. That's partly because early investments into AI services have been primarily focused on training AI models in massive data centers. When those AI systems are distributed for use in an "inference" model, that might generate more demand for edge computing services that are closer to users and therefore could speed up the performance of AI systems.

"Verizon's Private 5G Network provides the ultra-low latency required for real-time AI processing at the edge, critical for applications like robotics, and augmented reality," the operator noted of its new deal with Nvidia.

"The AI Accelerator can deliver a much better user experience, much more human-like chat experience, by lowering latency of responses," explained Fastly CEO Todd Nightingale at a recent investor event. 

Nightingale said Fastly's edge service could speed up ChatGPT responses from 5-6 seconds to less than a second while also reducing transit costs customers might incur sending their data to massive AI data centers.

"So, there's a lot of benefits to the AI Accelerator," Nightingale said.

Nvidia is obviously keen to stoke such hopes. Ronnie Vasishta, SVP of telecom at Nvidia, recently estimated that a mobile network operator running AI services across 6,000 base stations could generate up to $1 billion per year in revenues.

Nvidia is now one of the world's most valuable companies thanks to demand for its AI-friendly chips.

The AI network

Nvidia is also one of several founding members of the AI-RAN Alliance. The group is in the early stages of developing operator standards to inject AI services into future mobile networks.

One reason to do that is to leverage AI technologies to improve the performance of wireless networks. According to trade associations like ATIS and 5G Americans, there are lots and lots of benefits in leveraging AI technologies for networking operations.

But the other big goal of the AI-RAN Alliance is to create a new revenue stream for telecom operators by hosting enterprise customers' AI operations inside telecom networks.

"The group aims to increase platform utilization while creating monetization opportunities by running AI on shared computational resources," The AI-RAN Alliance wrote in its first white paper. The alliance released the 13-page document shortly after its first in-person meeting, which was hosted in Ericsson's Santa Clara, California, offices.

Interest in the AI-RAN Alliance is growing. The group launched earlier this year with just two network operator members, T-Mobile and SoftBank. Today it counts two more: EchoStar's Boost Mobile and South Korea's SK Telecom.

But AI may just be telecom's newest fad. For example, it's worth noting that SK Telecom this month reorganized its whole business around AI – just in time to shutter its once-hyped metaverse platform, Ifland.