By Tim Biggs
A year ago, Nvidia was one of Wall Street’s hottest stocks. The semiconductor giant, already a dominant force in the video game graphics market (and, inadvertently, the market for mining cryptocurrencies) successfully expanded to become the sole player in ultra-high-end AI processors, riding a wave of hype to briefly become the world’s most valuable company in June 2024.
As major businesses worldwide announced huge capital expenditure on AI data centres and processing, Nvidia was a major beneficiary, giving it extraordinary weight in major indexes and attracting significant investment from retail shareholders and super funds.
Jensen Huang has had a lot of explaining to do at and around Nvidia’s GPU Technology Conference.Credit: Bloomberg
But in January, the market was spooked by Chinese AI firm DeepSeek, previously a relative unknown, which released a conversational AI model it said could reason as well as the big American models but was created at a fraction of the cost. Nvidia’s market cap dropped by $938 billion in a single day. There was something of a rebound in February, followed by another dip.
Chief executive Jensen Huang this month made a case for a strongly resurgent Nvidia, at the keynote for the company’s annual GPU Technology Conference (GTC) and in investor interviews. It’s not that the US multinational has stopped making money; its most recent earnings update in February would indicate it’s making more than $3500 per second in pure profit thanks to its truly gargantuan trade in AI and networking tech for data centres. But that means little to investors if the volume of profit isn’t growing, and that’s where analysts have concerns.
First, there’s the question of whether companies will maintain or increase their investment in AI hardware, or if the big spending will be reconsidered. Then there’s the impact of DeepSeek, which implies that competitive AI will soon be available for less. And finally, there have been fears that US President Donald Trump’s tariff war could increase costs of manufacturing and shipping.
Huang, of course, had answers for each of these concerns at GTC.
On demand, he said that AI was the operating system of every industry going forward, and that even in the face of recession companies would increase investment because AI would be their key growth area.
Companies would continue to buy from Nvidia, he said, because even the biggest of the tech giants have failed to make chips that rival the company’s last-generation Hopper designs. The latest generation, called Blackwell, had seen more orders from cloud companies than Hopper did at a comparable point in its life cycle, Huang added.
On DeepSeek, Huang praised the company’s R1 model but said its implications had been misunderstood by the market. A reasoning model of its type will require much more computing power to use than older models, despite claims that less energy had gone into its training, he said, which means higher demand for Nvidia chips and not lower.
A so-called Superpod of Blackwell Ultra units will supply 288 CPUs, 576 GPUs and 11.5 exaflops of computing, with 300TB of memory.
The industry needs “100 times more [AI computing power] than we thought we needed this time last year”, he said, and indeed expenditure at data centres is climbing faster than analysts have expected, for now.
On tariffs, he said the impact would be small and short term as Nvidia was moving its most important manufacturing to the US. In fact, some analysts think the tariffs could end up becoming a competitive benefit.
“Current US semiconductor export restrictions put an absolute ceiling on AI compute [for China], at a degraded Hopper level,” said Richard Clode, portfolio manager at Janus Henderson.
“Over the next few years, new Chinese AI models will be constrained by that compute ceiling, while globally AI models are likely to be training on exponentially higher-performance AI infrastructure.”
Despite it all, it’s impossible to ignore how differently received this year’s Nvidia announcements were compared to those of the previous two years, when hype around AI applications was massively accelerating. Huang merely mentioning Dell at the 2024 GTC led to a rally in that company’s value, whereas an announcement this year about supplying chips to General Motors didn’t seem to move the needle for the automaker.
In previous announcements, Huang painted a picture of massive economic and social benefits that would be achieved through AI, but despite huge investments and many announcements from OpenAI, Amazon, Google, Microsoft and others, those just don’t seem to even be on the horizon.
Demand for the Blackwell chip is ramping up, and no viable competitor appears to be forthcoming.
Now Huang is saying industry needs to invest more and that a wider pool of industries needs to get behind the technology to see benefits.
Nvidia is down more than 10 per cent from the January peak, erasing more than $1 trillion in market value. So, was ill-founded AI hype pumping up the stock, meaning Nvidia will return to a stable value based on its still-formidable chips, or is it set to see a rebound as fears around competition and trade wars subside?
“The market is very sceptical on the stock,” said Alec Young, chief investment strategist at Mapsignals, as reported by Bloomberg. Nvidia’s current valuation, coupled with its high expected topline growth, is a sign that the “market thinks the growth’s not going to happen”, he said.
Yet demand for the Blackwell chip is ramping up, and no viable competitor appears to be forthcoming. A Blackwell Ultra chip will be launched this year, offering the same 20 petaflops of AI performance but with around 50 per cent more memory. Then next year the company will switch to a new architecture called Rubin, with a full stack that should provide 3.3 times the performance of Blackwell.
These are designed to be used in clusters as part of big data centres, but for smaller operations Nvidia has also introduced desktop PCs with Blackwell chips inside, starting at around $5000.
“In essence, Nvidia’s chips remain the new oil or gold in this world for the tech ecosystem, as there is only one chip in the world fuelling this AI foundation, and it is Nvidia,” said Wedbush analyst Dan Ives.
“This is now about the ripple impact for tech. We estimate for every $1 spent on an Nvidia chip there is a $8 to $10 multiplier across the tech ecosystem with the hyperscalers, software, data centre buildouts, cybersecurity, and energy demand all benefiting from this $2 trillion of AI CapEx set to take place over the next three years.”
Get news and reviews on technology, gadgets and gaming in our Technology newsletter every Friday. Sign up here.