- Comcast’s Elad Nafshi thinks the operator’s network can use AI better than its fiber competitors, particularly when it comes to identifying network issues
- The ways Comcast is improving its network “may be underappreciated” among investors, said NSR’s Jonathan Chaplin
- AI isn’t increasing Comcast’s network traffic yet, Nafshi said, because applications are mostly textual
When it comes to using AI, Comcast firmly believes it has a leg up on the fiber competition due to its DOCSIS 4.0 technology, according to chief network officer Elad Nafshi.
Speaking at New Street Research and BCG’s Future of Connectivity conference, Nafshi argued that because Comcast has embedded AI that’s “literally feet away from the customer” with real-time pattern detection capabilities, the operator can pinpoint “exactly” where there is interference in the network.
Comcast started rolling out DOCSIS 4.0 in fall 2023, offering consumers symmetrical multi-gig speeds of up to 2 Gbps. DOCSIS 4.0 has also allowed Comcast to “take advantage of a new generation of edge compute devices” that are built into its nodes and amplifiers, Nafshi said. He further touted Comcast’s low lag technology (which supports applications from companies like Meta and Nvidia), leverages AI to automate “over 99.7% of all software changes that we're making on the network.”
“When you look at some of our telco competitors that are still reliant on vendor proprietary hardware that all is hand configured, it's that operator at the end of day that makes a mistake and causes a national outage that we've bypassed,” Nafshi said.
How operators are dabbling in AI
We’re around two years into the AI hype cycle, yet telcos for the most part are just starting to dabble with the technology to see if it’s worth the money.
AT&T told us earlier this year AI is already delivering plenty of ROI, particularly in software development and customer care. Verizon, which like Comcast has been using AI to detect network issues, recently launched AI Connect to take advantage of enterprise demand for more network bandwidth. Comcast and other operators have been shifting to a more cloud-based networking approach to better implement AI tools into their operations.
Nafshi went on to say other providers typically have to dispatch a technician to troubleshoot network issues, and that technician would use a device to “basically shoot slight[ly] down the fiber to try and locate where the fiber cut is.” But that doesn’t give a precise location on where the cut is located, which can take “several hours” to figure out.
As he talked up how Comcast is well-positioned against fiber competitors, Nafshi stressed there’s already a ton of fiber across its backbone and access network – coax is just a part of it. “These catch words around [fiber versus coax], it's meaningless if the technology behind it makes it meaningless,” he said.
According to New Street’s Jonathan Chaplin, Nafshi “made an impassioned case for the capabilities of his network being equal, and in some cases superior to fiber.”
“The quality of the network that Comcast is building, at a fraction of the cost that it takes to build fiber, may be underappreciated among the investment community,” Chaplin wrote in a note this week.
Notably, Comcast has said it costs around $200 per passing to upgrade homes to DOCSIS 4.0 tech, which is significantly lower than some of the costs we’ve seen to deploy fiber (especially in rural areas).
Is AI driving network traffic?
While AI is helping Comcast with network operations, Nafshi said it’s not increasing traffic growth – yet.
Sure, more Comcast subscribers are using the likes of ChatGPT and OpenAI because they’re embedded into Google’s search engine, but AI applications are “mainly textual today,” he said.
AvidThink’s Roy Chua similarly told us how latency is not yet a huge constraint of AI applications, because people are mostly using it to chat with bots, generate images as well as for backend inferencing and reasoning.
AT&T predicted its network traffic will double by 2028 due to the rise of AI applications, and it’s preparing accordingly through a recent trial of Ciena’s 1.6 Tbps coherent optics.
The future could see larger AI workloads being run over wireline networks, Chua added, which could even be AI-powered augmented reality (AR) or virtual reality (VR) experiences.
“The world of Avatar is coming,” Nafshi concluded. “I think that the world where you talk to Siri and Siri doesn't look back at you, is probably not going to take long.”