The bursting of an AI bubble is just a matter of time and could hurt the telecom players doing deals with Nvidia.
Jensen Huang, the boss of Nvidia, is among a dwindling band of people who rubbish talk of AI boom and bust. “There’s been a lot of talk about an AI bubble,” he said on the chipmaker’s third-quarter earnings call this week. “From our vantage point, we see something very different,” he continued, reeling off examples of why a transition from older computing platforms to Nvidia’s graphics processing units (GPUs) makes perfect sense. It’s a ridiculous but unsurprising assertion by the AI billionaire as Nvidia’s market cap crests previously undiscovered summits while the rest of the world economy crawls along.
Not even Nokia CEO Justin Hotard, Nvidia’s biggest telecom partner, was prepared to dismiss bubble talk when asked about it on a recent call, instead drawing a parallel with the dotcom boom 25 years ago. “There was a bubble burst, but then go look at Cisco’s leadership in networking for the following ten years,” he said. Financial experts have stopped speculating about the if and started asking when the bubble will burst, how damaging it will be and which companies will most quickly recover.
There might be some way to go after Nvidia’s latest financials sent its share price up 5% in pre-market trading. After several years of exploding sales, revenues grew another 62% year-over-year, to $57 billion, while Nvidia’s net income was up 65%, to around $31.8 billion. The absence of viable competitors is signified by Nvidia’s obese gross margin, still more than 73%.
Right place, right time
None of this seems by design. Nvidia was founded in 1993, almost 30 years before the birth of generative AI, but has spent most of its adult life as an ordinary tech stock, catering to the games community for which its GPUs were originally designed. In November 2022, days before the release of ChatGPT, its share price was worth $16.27 on the Nasdaq. Today, at the time of writing, it is valued at more than $186. Those games chips, with their ability to do multiple calculations at once, serendipitously turned out to be good for AI, too. Nvidia’s market cap reached $5 trillion recently, although it has subsequently fallen back to around $4.5 trillion, giving Nvidia a price-to-earnings ratio of about 53.
Even after the drop, it remains wildly out of kilter with non-technology stocks. Warren Buffett’s Berkshire Hathaway, the world’s most highly valued non-tech company in November, has seen its share price grow 120% in the last five years. That of Walmart, a more recognizable bricks-and-mortar company, has roughly doubled. Nvidia’s stock has soared 1,325%.
The disconnect between Nvidia and the real economy is even more glaring. The global GDP growth rate was about 6.4% in 2021 as the world bounced back from the initial lockdowns of the pandemic. It fell to about 3.4% in 2022, when GenAI appeared, dropping below 3% for each of the following two years. This forecast this year is for about 3%. Nobody, as data-center loads of online coverage have established, is doing fabulously well out of AI except the people directly involved in it.
Company bosses in sectors including telecom (especially telecom, it seems) increasingly refer to it as an explanation for the job cuts scything through staff ranks. But it’s likely to have caused only a small fraction of those and has not, in many cases, done anything to improve company sales and profits. When AI costs as much as it does, and gobbles so much energy, that should hardly come as a surprise.
Increasingly dubious, too, is Nvidia’s habit of making multibillion-dollar investments in customers or partners, behavior that smacks of old-fashioned vendor financing to critics. The most outrageous example is the $100 billion that Nvidia is pouring into OpenAI, the Sam Altman-founded and loss-making company behind ChatGPT. Struggling to afford GPUs, it now has a ready source of cash from a generous donor to use. What Nvidia pays out of one pocket, it ultimately puts back in the other. Circular is the apt word analysts have been using to describe this arrangement.
Faustian pact?
It is evident, albeit to a lesser extent, in the more recent deal with Nokia. To recap, Huang’s company is investing $1 billion in the Finnish equipment vendor in exchange for a 3% stake, which will seemingly make Nvidia the second-biggest shareholder in Nokia. This gives it an influence over Nokia’s strategy that no other supplier enjoys, and Nokia’s strategy has suddenly become all about Nvidia. At the time of the deal, it revealed it would build a range of 5G and 6G network products on Nvidia’s GPUs, providing an “AI-RAN” option for customers. But the relationship with Nvidia turns out to be far more exclusive for Nokia – if not Nvidia – than it originally appeared.
This only became obvious this week at Nokia’s capital markets day in New York, where Hotard heralded the beginning of the end for custom silicon in Nokia’s mobile business with a “shift from proprietary to general-purpose hardware.” Until then, it was unclear how much of its research and development Nokia would commit to Nvidia’s “general-purpose” GPUs and how much it would continue to spend on that proprietary tech, based heavily on Marvell Technology’s silicon. The answer now looks clear.
It’s a huge gamble by Hotard that assumes Nokia’s 5G customers will share his enthusiasm for a GPU-based RAN. Numerous telcos that have spoken with Light Reading currently do not, and they include big hitters such as Verizon in the US and Orange in Europe. For many of these service providers, general purpose continues to mean lower-cost central processing units from Intel, AMD and even Arm licensees, not Nvidia’s “expensive” GPUs, as they were recently described by Yago Tenorio, Verizon’s chief technology officer. A bursting of the AI bubble that coincided with the launch of Nokia’s first GPU-based products could also leave Nokia exposed if it had no competitive alternatives at the time.
Much worse, however, would be a similar pivot from custom silicon to GPUs by the other big 5G vendors available to European and North American telcos, and there are only two viable options besides Nokia on the baseband (RAN computing) side of the equation – Ericsson and Samsung. It would be a surprise if Nvidia had not held talks with those companies, especially Ericsson, about a similarly “circular” arrangement to the one it has with Nokia. On his call with analysts about the tie-up, Hotard had to acknowledge that Nvidia was free under the terms of the deal to do the same thing with his competitors.
This clearly presents another risk for Nokia – that of losing the competitive edge it might gain as the sole vendor of GPU-based RAN products. But if Ericsson and Samsung similarly decided to retreat from their current silicon platforms and fully commit to Nvidia, the whole market for critical mobile infrastructure outside China would be dependent on one overvalued chipmaker.
That scenario may, of course, never materialize, but Nvidia’s big bucks will be hard for struggling vendors and telcos to resist, and Nvidia’s desire to sell more chips to new customer groups looks irrepressible. Two years of talking about AI-RAN got it almost nowhere. It took that investment in Nokia to make a big RAN developer start to build GPU-compatible products. The logical next step is surely an investment in a telco worried about the expense of deploying a GPU-based RAN. T-Mobile US is a possible target.
Nvidia must be admired for what it has achieved. It obviously employs some of the smartest minds on the planet – before they are overtaken by AI superintelligence, that is – and has continued to innovate. But it also benefited from being in the right place at the right time with a computing platform not initially meant for AI. As it grows, it owes success increasingly to its market power and burgeoning ecosystem – the moat of its CUDA software platform that is hard for any capable rival to breach. Some cheerleaders, meanwhile, are bizarrely celebrating the latest results as evidence there is no AI bubble. Even the most naïve child knows a balloon can’t inflate forever.
https://www.lightreading.com/ai-machine-learning/nvidia-is-eating-the-world-and-telecom-is-part-of-the-meal
