What the $500B Stargate AI plan could mean for telecomWhat the $500B Stargate AI plan could mean for telecom

The flood of investment into AI data centers powered by Nvidia chips could have major implications for 5G silicon developers.

Iain Morris, International Editor

January 22, 2025

7 Min Read
Still from Stargate movie
Stargate: No, not that one.(Source: Everett Collection Inc/Alamy Stock Photo)

A crack squad of US soldiers and a single nerdy scientist travel through a wormhole to a distant planet, where an alien superintelligence has enslaved humans it captured in Egypt thousands of years ago, after first using them to build a few pyramids. Fans of early 1990s science-fiction movies will recognize the plot of Stargate. Evidently drawn to the story of human subjugation by a higher intelligence, several of the world's tech bros and Donald Trump have stolen the name for their latest AI plans.

They are to involve hurling half a trillion dollars (yes, half a trillion) over the next four years at the construction of new AI data centers in the US. If it seemed as though the last thing the US needed was more AI data centers, these ones prioritize OpenAI, the original star of the generative AI show, whose models sit behind the famous ChatGPT app. Sam Altman, its sinister founder of adolescent appearance, has been scurrying around like a billionaire beggar pleading for trillion-dollar investments to create an artificial superintelligence far smarter than today's somewhat underwhelming AI. Stargate could be it.

Its list of other investors, several of whom lined up alongside Trump after his epic session with executive orders, intriguingly includes Japan's SoftBank, Oracle (a throwback to an earlier software age) and MGX, which appears to be an AI-focused offshoot of Mubadala, Abu Dhabi's sovereign wealth fund. Unsurprisingly, given the billions it has pumped into OpenAI, Microsoft is also on the list. Arm, the SoftBank-owned but UK-headquartered chip designer, and Nvidia, whose graphical processing units (GPUs) are the hardware engines of AI, are described as key partners. Their growing involvement in telecom makes this an interesting watch for telcos and their traditional vendors.

The Son also rises

But it's SoftBank and OpenAI that will be directing the show, with SoftBank taking financial responsibility and OpenAI operational responsibility for Stargate. SoftBank founder Masayoshi Son, whose reputation for business acumen has suffered after several disastrous bets in recent years, is chairman. His role proves the sun is yet to set on the 67-year-old Son. Like Trump, he can bounce back when he's down.

It's not the newest news, The Information having first reported on plans for a $100 billion investment in a "Stargate AI supercomputer" last March. But that $100 billion is to be spent immediately, according to the official announcement. Buildout has apparently started in Texas. What takes shape will presumably be a Microsoft data center replete with Nvidia chips (GPUs and more humdrum Arm-based central processing units, or CPUs) and OpenAI algorithms. Where Oracle fits in is less clear. Larry Ellison, its 80-year-old founder, was photographed next to Trump at the official announcement, but he seems to spend most of his time these days sailing yachts and watching tennis.

The combined age of three men in that same photo is 225, meaning only Altman – looking like someone out with his grandfathers and their Japanese pal – must really worry about the consequences of artificial super intelligence, or even the intermediate step of artificial general intelligence (AGI). Altman, though, displays not a shred of concern about his crusade. In the choice of Stargate as a name, does he recognize enslavement as a possible consequence without caring?

This is obviously not how the project has been sold to the public. "All of us look forward to continuing to build and develop AI – and in particular AGI – for the benefit of all of humanity," said the companies in their release. "We believe that this new step is critical on the path, and will enable creative people to figure out how to use AI to elevate humanity." The language is that of altruism, of a saintly effort to raise living standards for the world's population.

What has the Trump administration got to do with any of this? It's clear from the release that all the equity is to come from the capacious wallets of the project partners. This is Make America Great Again with a massive injection of Asian and Middle Eastern money. But the government's support was obviously needed, and Trump was ready to show it just minutes after tearing up another Biden policy. The relevant one meant AI companies had to share the results of safety tests with authorities. Under Trump, that's no longer deemed necessary.

Dawn of the AI RAN

What has any of this got to do with telecom, is the more pertinent question that a Light Reading audience may be asking. And the answer is partly that some of Stargate's main partners are the same companies now pitching to replace older technologies in telecom networks. An Nvidia supercomputer housing a GPU and an Arm-based CPU is at the heart of the AI RAN (for radio access network) concept, which envisages using them for AI applications as well as the compute side of 5G and 6G.

The potential worry for anyone involved in the design and manufacture of customized silicon for a relatively small market such as the RAN is a lack of manufacturing capacity. TSMC, the Taiwanese foundry responsible for a majority of the world's most advanced chips, is dedicating a bigger share of its resources to AI chips. The revenues it earns from making those AI chips already accounted for about 15% of last year's total and that proportion is expected to double this year.

Quizzed about AI demand constraints on his company's last earnings call a few days ago, TSMC CEO CC Wei told analysts that "I know we have a very tight capacity to support the AI demand. I don't want to say I'm the bottleneck. TSMC is always working very hard with customers to meet their requirements." In this sort of environment, the big AI spenders like Nvidia will clearly have the priority in the pecking order over smallish designers of application-specific integrated circuits (ASICs).

This is partly the rationale for virtual RAN – the idea that telecom could piggyback on the much bigger ecosystem for general-purpose, x86-based CPUs and realize economies of scale. Rather than worrying about supply constraints and committing budgets to specialized silicon, RAN vendors could lean on the IT market for hardware and common software tools. This would free up resources for important things like the design of radio algorithms.

But virtual RAN has not taken off amid growing industry concern about the outlook for x86 giant Intel, which stumbled to a $16.6 billion loss for its recent third quarter, after recording a small profit of $300 million a year before. Its sales fell 6% over that period, to $13.3 billion, while Nvidia's rose from $18.1 billion to $35.1 billion. Investment is flooding into AI data centers and the chips they use on a scale IT has not seen for a long time.

Meanwhile, the market for RAN products is shrinking as telcos tighten their belts. After spending $45 billion in 2022, they coughed up $40 billion in 2023 and are expected to have spent about $35 billion last year, according to the midpoint of the latest forecast from Omdia, a Light Reading sister company. In this context, Nvidia may seem like an attractive destination.

Not everyone agrees the market for customized RAN chips is suddenly in peril. One knowledgeable industry source, speaking on condition of anonymity, insists TSMC capacity is available if companies commit to it and that many TSMC-reliant industries, automotive and smartphones among them, will not pivot to GPUs. There is concern about the economic case for using notoriously power-hungry GPUs in the RAN. Ericsson insists purpose-built baseband will remain "the most energy-efficient and compact hardware for radio site deployments."

What's interesting is that both Ericsson and rival Nokia are already known to be exploring the redesign of RAN software for compute unified device architecture (CUDA), the platform used by Nvidia. Both are part of the Nvidia-backed AI RAN Alliance and feature in a separate initiative that involves Nvidia and T-Mobile US. Nokia already runs less computationally demanding RAN software on Arm-based chips in some of its latest 5G products. Moving that to an Nvidia CPU would seemingly be quite straightforward.

Would a dominant Nvidia in AI, as critical to multiple industries as Intel has been in data centers and PCs, be desirable? A criticism of virtual RAN is that at the processor level it has continued to lack alternatives to Intel, despite much talk in the last two or three years about the emergence of AMD, an x86 rival, and Arm as viable alternatives. Nvidia as another option in a healthily competitive market, rather than a direct successor to Intel, would surely look much better.

Read more about:

AIOmdia

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