Nvidia has turned in another set of sizzling results on the back of AI-driven demand for its products, yet industry watchers are increasingly wondering how long this substantial growth can continue, or whether the bubble is going to burst.

The GPU giant reported revenue for its fiscal Q1 2025 (ended April 28, 2024) of $26 billion, up by a staggering 262 percent from the same period a year ago. Its datacenter revenue makes up $22.6 billion of that figure, up an even more eye-opening 427 percent year-on-year.

These figures, which exceeded the expectations of financial analysts and Nvidia alike, were primarily driven by strong demand for the Nvidia Hopper GPU computing platform, the company said. Its overall revenue was also up sequentially - rising 18 percent from the previous quarter, which had likewise shown an impressive jump from a year ago.

Looking ahead to the next quarter, Nvidia said it is "currently in production with shipments on track for Q2" for its H200 GPU, while products based on its next-gen Blackwell GPU architecture will also start to ship in Q2 and ramp in Q3.

Perhaps based on demand for these, the company forecasted revenue for fiscal Q2 2025 to step up again, to $28 billion, plus or minus two percent. If accurate, this would be lower than the 23 percent increase it saw between the previous quarter and this one, but it is not unthinkable that Nvidia will beat expectations again.

But how long can the GPU maestro continue to beat expectations and show staggering year-on-year growth each and every quarter? CEO Jensen Huang certainly sees no cause for concern, at least in his public statements.

"The demand for GPUs in all the datacenters is incredible. We're racing every single day," he told analysts on a conference call to announce the latest results.

"Customers are putting a lot of pressure on us to deliver the systems and stand it up as quickly as possible. And of course, I haven't even mentioned all of the Sovereign AIs who would like to train their regional models," Huang said, adding "anyhow, the demand, I think, is really, really high and it outstrips our supply."

That, of course, is one reason for Nvidia's profits - its accelerators are like gold dust, and not being able to make enough to meet demand pushes up prices.

Yet high prices also attract competition, and the company now faces growing competition from in-house accelerators developed by the hyperscalers and cloud companies, which are also Nvidia's biggest customers.

Google, for example, has been outfitting its cloud with millions of its custom-built Tensor Processing Units (TPUs), as The Register recently reported. These are largely used to power Google's services at present, with Nvidia GPUs offered to cloud customers, but this could change.

"Nvidia's growth is strong, but it mostly comes from just 6 customers," said IDC's Senior Research Director in EMEA, Andrew Buss.

"A year ago, no single customer made up more than 10 percent of its revenue, now they do, and that's a risk for Nvidia," he explained.

There is also the sense that we are currently in a "panic-driven build-out" period at the moment, Buss added, where hyperscalers and enterprises alike are desperate not to be left behind in all the AI hype, and this situation won't last forever.

Nvidia also faces competition in its software stack, according to TechMarketView Principal Analyst Simon Baxter.

"Some of Nvidia's biggest customers are taking aim at its Cuda software platform (a key part of its AI dominance alongside the GPU chips), by helping to develop Triton, software that was first released by OpenAI in 2021 and designed to make code run software on a wide range of AI chips," Baxter said.

"Rival chipmakers Intel, AMD and Qualcomm are all looking to use Triton to help lure customers away from Nvidia, whilst AI suppliers such as Meta, Microsoft and Google, which have all spent billions of dollars on Nvidia chips, are also contributing to Triton at the same time as producing their own AI hardware," he added.

Additionally, Nvidia's continued success is tied to the perception that all of this investment in AI is going to pay dividends somewhere down the line - yet there are growing doubts about this notion.

According to the Wall Street Journal, investment outfit Sequoia Capital estimates that the AI industry spent $50 billion on Nvidia chips to train advanced AI models last year, but to date this has resulted in only $3 billion in revenue.

The Financial Times also noted recently: "The uncertain take-up of generative AI stands in stark contrast to the money being poured into the tech infrastructure needed to support it."

Microsoft's success with OpenAI also added six percentage points of growth to its Azure cloud platform, as customers test out the technology. While that likely translates to nearly $3 billion extra revenue a year, that's little more than one percent of Redmond's total revenue.

Realistically, Nvidia is unlikely to be boasting revenue increases of 262 percent for much longer. These year-on-year increases are in comparison with last year, before the generative AI wave started to take effect.

However, Huang moved to dispel concerns among investors about whether Nvidia will see its momentum slow, at least for now. "We see increasing demand of Hopper through this quarter," he said. ®

https://www.theregister.com//2024/05/23/nvidia_beats_market_expectations_again/