Telecom vendors differ in how they structure AI

AI is the latest buzz-worthy technology to hit the telecom market, following in the footsteps of once hot topics like blockchain, virtualization and cloud native. And as with those other technologies, vendors are struggling to incorporate AI into their existing operations.

"What we're doing now [in AI] is not what we were doing three months ago," said Jonathan Davidson, EVP and GM of Cisco Networking. He made his comments during a media event last week at the MWC Barcelona trade show. 

Davidson said Cisco is taking a structurally flat approach to its embrace of AI technologies. "It's not a centralized team," he said, noting that most of Cisco's AI experts hold doctorate degrees.

However, Davidson said Cisco has a slightly more centralized strategy regarding AI within its networking division.

Overall, Cisco's AI experts are working to share their insights across the company for different products and divisions. "Those teams collaborate together," Davidson said.

Cisco has said it expects $1 billion in AI-related orders in its upcoming fiscal year. The company also has a separate agreement with Nvidia to sell Nvidia's AI-capable graphical processing units (GPUs) combined with Cisco's Ethernet offerings to enterprises and others, with products scheduled to start shipping later this year.

Ericsson's approach

Cisco's rival Ericsson has a slightly different approach to AI.

Elena Fersman is the head of Ericsson's Global AI Accelerator effort, which is the company's centralized repository for AI expertise. During an interview at MWC, Fersman told Light Reading that Ericsson employs about 350 AI experts in the office of the company's CTO.

"The ambition is that we transfer the knowledge" across the company, she said.

Fersman said Ericsson is deploying several flavors of AI technology including generative AI (GenAI); trustworthy AI; cognitive, intent-based AI; and machine learning operations.

"We build it in-house, normally," she said, explaining that Ericsson obtains its core AI platforms from a variety of open source options and then trains the platforms with its own data. Ericsson has a "couple dozen" flavors of AI that are built from one core, according to Fersman.

"Ericsson is very good on structured data," she said, noting that the vendor has been training its AI models using its user manuals, network trouble ticket data and other data sources. Fersman echoed other executives in the industry in arguing that AI is only as good as the data it is trained on.

When asked to describe the kinds of AI platforms Ericsson is developing, Fersman said: "They don't compose poems." Then she clarified that, actually, they might be able to compose poems, but they're not designed to do that. Instead, they're designed to improve the company's internal processes and communications as well as to enhance Ericsson's customer-facing products and services.

Investing in the future

Ericsson has purchased a range of AI equipment to support its efforts, including GPUs from Nvidia, according to Fersman. She said Ericsson runs its own internal corporate cloud – Ericsson Web Services – that can power AI applications, but the company also works with major hyperscale companies to test out their AI platforms.

"We've tried all of them," Ferman said of large language models (LLMs) like those from OpenAI. "It's a good opportunity to work with hyperscalers."

But Fersman said Ericsson is careful to protect its data sources and not "leak" them into other, potentially insecure data pools. "Telecom knowledge is so sensitive, it needs to be in-house," she said.

Cisco's Davidson agreed. "We don't want that [data] to get out and train third-party models," he said, arguing that the data held by mobile network operators and vendors like Cisco will be valuable in an age of AI.

IBM advises on training

IBM is another big vendor playing heavily in the AI industry. Indeed, the company said during its most recent quarterly earnings report that its AI sales and bookings roughly doubled sequentially in the fourth quarter.

In the mobile industry, the company continues to push its watsonx AI platform for training and other programs. Stephen Rose, IBM's GM for global industries and the company's top telecom executive, told Light Reading that the vendor is working to help its telecom customers navigate the complex AI marketplace.

Speaking last week at the MWC Barcelona trade show, Rose explained that there are hundreds of open source AI platforms for various activities, and IBM works with its customers to help them select the proper models for their goals. Then the company can advise on training those models with various data sources. IBM can also help provide the proper data guardrails for AI models so they don't stray into unwanted operations, Rose added.

Staffing rush

IBM recently suggested it would lay off some employees in 2024 but added that it will also hire more AI-centric positions. The end result will likely be an unchanged headcount.

Fersman, the Ericsson executive, said it's difficult to recruit top AI researchers because of high demand for such positions. She said she moved to California – a hotbed of tech – specifically in order to be able to recruit AI experts.

"They're expensive," agreed Davidson, the Cisco executive, in a nod to some sky-high AI salaries.

However, he said Cisco has been able to recruit some of its AI workforce based on the data the company can apply to its AI models.

"We are doing billions of measurements every single day," he said. "We have a view into all of these networks that no one else has."

Davidson added: "People say, 'Oh, I wish I could work with that data set. People get pretty excited about the mission" of creating connections.

"I think it's extremely valuable," Davidson said in discussing the data sets Cisco has access to. "You can't derive conclusions without an extremely large amount of data."