Why DeepSeek matters (or doesn't) for telecomWhy DeepSeek matters (or doesn't) for telecom

Telcos anticipate lower-cost models and edge computing opportunities, while equipment vendors have suffered in the wake of the DeepSeek news.

Iain Morris, International Editor

January 29, 2025

7 Min Read
Artificial intelligence in the palm of your hand
(Source: Pitinan Piyavatin/Alamy Stock Photo)

Glugging mulled wine and exchanging gifts, most Christmas partygoers would not have even heard of a Chinese artificial intelligence developer called DeepSeek. That has changed dramatically after a flight through the moderate stock market turbulence of the last few days. A huge volume of news stories, LinkedIn posts and X updates lays claim to being an authoritative take on the DeepSeek story. The human experts have multiplied faster than any generative AI gibberish. Some are undoubtedly hallucinating.

The only real certainty is that DeepSeek has given the West the willies. The worst-case scenario, for certain US AI stakeholders, is that it comes with supremely efficient software requiring far less processor power than Western rivals like OpenAI. Naysayers scoff, insisting it cannot possibly be that much better. It probably drew on boatloads of Nvidia's most advanced graphical processing units (GPUs), stockpiled before tougher sanctions cut China off, they say. Accusations of intellectual property theft are now circulating, too.

Ignorance of all the facts is reflected in what has happened to Nvidia's share price this week. On January 27, it plunged 17%, "wiping" hundreds of billions of dollars off Nvidia's market capitalization, headlines screamed, as investors imagined slower production lines at the GPU factory. A day later, the stock clawed back some of what it had lost, rising 9% after what was judged to be an epidemic of overreaction.

None of this strictly has much to do with telecom, but there is a spillover effect with obvious repercussions for AI users, telcos included. Asked specifically about DeepSeek on his company's recent earnings call, John Stankey, AT&T's cost-cutting CEO, evidently hoped it might boost his profits.

In such a nascent market, he somewhat breathlessly told analysts, "we should expect there's going to be days we wake up like this one when somebody comes in and says they figured out a way to get as much benefit out of the model by consuming less power or using less processing capability or they fine-tuned the models to work in a particular domain area more effectively where they can run locally as opposed to in the cloud, which is going to open up and facilitate new applications and business models."

Taking robodog for a walk

Generative AI first poked its head through the door of telco operations with the emergence of ChatGPT more than two years ago. It has subsequently clambered inside the building, ushered some employees out and forced others to change. A 30% year-over-year drop in call volumes at AT&T, attributed to AI, is just one example of its impact.

And yet, no telco – at least none in the Western world – seems to be doing very well out of generative AI. AT&T's share price was this morning worth 14% less than it was five years ago. Its headline annual sales have flatlined in the last couple of years and plummeted a third in the last five after a retreat from some markets. Despite successive waves of cuts, its operating expenses rose 4% last year, to $103.3 billion, due mainly to much higher impairment and restructuring charges. Operating profits consequently sank 19%, to about $19 billion.

The story is even worse for telcos in Europe's small and overcrowded markets. In the UK, Vodafone and Three claim to have such poor returns they cannot even afford to roll out a decent 5G network. Telco talk of models that would "run locally as opposed to in the cloud," as Stankey put it, sounds like an attempt to resurrect ideas about edge computing, which would host user applications at telco premises dotted around a country rather than inside a giant hyperscaler facility. Yet telco edge computing, discussed for many years, has gone nowhere.

The logic was that local hosting would shorten the distance to the user for the network signal and reduce "latency," which measures that journey time in milliseconds. Lower latency was needed for more advanced applications, said advocates. But there has been no nationwide edge refurbishment of thousands, hundreds or even dozens of telco premises, because no application has emerged that would justify it.

If it does exactly what it says on the tin, DeepSeek could theoretically improve the economics for a telco of an edge computing deployment to support AI applications. AT&T competitor Verizon now seems to think it can earn money by acting as a kind of landlord for AI tenants. Nvidia is promoting a similar arrangement as part of its AI RAN (for radio access network) strategy. The idea is that telcos run network software on locally hosted GPUs and sell the extra capacity for inference, allowing applications to make use of AI models after they have been trained.

But as with pre-AI edge computing, the role for telcos, as opposed to dedicated GPU-as-a-Service (GPUaaS) companies like Nscale, is unclear. Voice-based AI is not heavily dependent on latency. The latency-sensitive example Nvidia and Japan's SoftBank recently provided was that of a robot dog following a person. When inferencing happened at the telco edge, robodog immediately reacted to its human handler. With inferencing in the cloud, it "struggled to keep up," said the companies. But walkie time for robodogs does not sound like it would have mainstream appeal – especially after footage appeared of them machine gunning targets.

A bumpy week for vendors

Telco share prices barely flickered in response to the DeepSeek news, probably because no serious investor thinks telcos have much skin in the game. The Nvidia-like turbulence instead affected the companies providing connectivity for and between the GPUaaS factories and AI data centers, some of which are better known as telco vendors. If less processor power is needed than people originally thought, it could weaken demand for those connectivity products.

Accordingly, one of the biggest losers this week was Ciena. Its share price tumbled 21% on January 27 as investors panicked about a possible slump in demand for its optical equipment. At the time of writing, it had not staged much of a recovery and was still trading at a 17% discount to its closing price on January 24. Some 54% of its revenues came from "non-telco" customers during its recent first quarter, with sales to cloud providers up 38% year-over-year. It is exposed because it boasts a 50% plus market share in data center interconnect (DCI). It is also trying to expand from doing the connectivity between data centers to making the "pluggables" used within them.

All this means that companies like Arista, Cisco and Juniper with their Ethernet switches and routers also suffered this week. To connect the clusters of GPUs in a data center, two main options are available. The first is a technology called Infiniband, monopolized by Nvidia since a $7 billion takeover of Mellanox in 2019. Companies like Nscale are therefore gravitating toward Ethernet-based connectivity, where there is more competition. As with DCI, weaker demand for GPUs would naturally hurt the outlook for these products.

To briefly convey evidence of the damage, Arista has fallen 17.6% over a five-day period, while Cisco is down 2.6% and Juniper 5.9%. This is certainly no rout. Arista's share price remains 58% higher than it was this time last year, Cisco is up 13.7% over that period, and Ciena has gained 51.7%. But the week's bumpiness serves to illustrate which companies in the connectivity products sector may be chiefly at risk.

One other is Nokia. The Finnish equipment vendor has had a tough time in a shrinking 5G market over the last couple of years and is looking to its smaller optical and Internet Protocol units for near-term growth. Last June, it announced a $2.3 billion acquisition of Infinera, another optical equipment maker, in anticipation of more AI-fueled data center business. Nokia's share price over a five-day period is down 3.7% on the New York Stock Exchange, but it remains 15.8% higher than it was a year ago. So far, the DeepSeek damage is limited.

Read more about:

AI

About the Author

Iain Morris

International Editor, Light Reading

Iain Morris joined Light Reading as News Editor at the start of 2015 -- and we mean, right at the start. His friends and family were still singing Auld Lang Syne as Iain started sourcing New Year's Eve UK mobile network congestion statistics. Prior to boosting Light Reading's UK-based editorial team numbers (he is based in London, south of the river), Iain was a successful freelance writer and editor who had been covering the telecoms sector for the past 15 years. His work has appeared in publications including The Economist (classy!) and The Observer, besides a variety of trade and business journals. He was previously the lead telecoms analyst for the Economist Intelligence Unit, and before that worked as a features editor at Telecommunications magazine. Iain started out in telecoms as an editor at consulting and market-research company Analysys (now Analysys Mason).

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like