By Paul Smith and Joanne Tran
Big investors say that the sudden emergence of a cheaper Chinese artificial intelligence model that claims to require less computing power has raised doubts about the long-term growth prospects of high-powered chip manufacturer Nvidia and other American tech giants.
The success of DeepSeek’s platform, known as R1 became widely available at the weekend, pushed Nvidia sharply lower on fears that fewer chips would be needed to train the same complex processes that had made Microsoft-backed OpenAI and Amazon-supported Anthropic the hottest private technology companies in the world last year.
“I’ve always been a bit sceptical that Nvidia was going to be the only winner,” said Garry Laurence, a former head of global equities at Perpetual who now runs his own firm, Profeta Investments. Mr Laurence, whose fund owns Microsoft and Google shares, said the introduction of DeepSeek had made him “quite mindful” of investing in companies solely making chips.
“The valuations did get stretched in the sector,” he added. “Nvidia’s valuations are about 40 times earnings. Those earnings are assuming that people keep buying the chips at the rate that they are buying, and that there are no competitors. I just think that there’s a lot of optimism baked in.”
Since the end of 2022, when OpenAI introduced its ChatGPT model, Silicon Valley has been the undisputed leader in developing AI models. DeepSeek, which is owned by Chinese hedge fund High-Flyer, promises to be able to provide similarly smart models for free, and claims to use significantly fewer costly Nvidia chips than its rivals in the United States.
Australians have invested at least $28 billion in Nvidia through managed funds and their superannuation savings, according to publicly disclosed positions analysed by figures by global equities manager VanEck.
Marc Andreessen, a veteran venture capitalist and the founder of Andreessen Horowitz, described the sudden emergence of DeepSeek’s RI model as AI’s “Sputnik moment” – a reference to the shock felt in the US during the Cold War when the Soviet Union displayed its satellite prowess.
But Tesla chief executive Elon Musk, a confidant of US President Donald Trump and the founder of artificial intelligence start-up xAI, said on X that DeepSeek “obviously” had more Nvidia chips than it had claimed and had skirted US trade restrictions which prevented their sale to Chinese companies.
Mr Musk, who has also been accumulating Nvidia chips for xAI, made the comments in response to an interview with Alexandr Wang, whose Scale AI provides data to train the platforms.
Mr Wang said DeepSeek must have access to 50,000 high-powered chips rather than the 10,000 less-powerful A100s it has disclosed. If those claims are true, it would mean the Chinese model requires far more power than it claims, a revelation that would ease fears by US tech investors.
Mr Trump, who on Tuesday said the launch of DeepSeek was a “wake-up call” for US companies, also described it as a “positive because you won’t be spending as much, and you’ll get the same result”.
“Instead of spending billions and billions, you’ll spend less, and you’ll come up with, hopefully, the same solution,” Mr Trump said.
Both OpenAI and Nvidia were also optimistic, with the former’s chief executive, Sam Altman, writing in a post on X that it was “legit invigorating to have a new competitor”. “The world is going to want to use a lot of AI, and really be quite amazed by the next gen models coming,” he wrote.
The uncertainty over DeepSeek sent the shares of ASX-listed companies exposed to AI lower on Tuesday. NextDC, which operates data centres used by big US tech companies, fell 6.4 per cent. Another group which has a big pipeline of data centres, Goodman Group, was down 7.8 per cent. The broader market closed marginally lower.
Local investors were split on the longer-term implications of an AI model that is free, and could be developed on far fewer Nvidia chips.
Wilson Asset Management’s senior analyst William Liu said he was concerned about the lack of clarity around capital expenditure of big tech companies on AI infrastructure. WAM’s global fund does not hold Nvidia, but is now seeing it as a potential target due to the price drop.
“[Nvidia]’s product is well ahead of its competitors, but we need to get more understanding of what AI data centre capex is going to be, in light of DeepSeek’s breakthroughs in software-driven resource optimisation,” he said. “If the Western AI companies adopt some of the breakthroughs DeepSeek has shown the market, they can run large language models more cost efficiently, potentially making AI development more accessible.”
The slump in Nasdaq made way for the single biggest day of US trading on investment platforms popular with retail shareholders. Superhero, a popular app, said that “significant amounts of money was piled into Nvidia … with one week’s worth of trading volume in a single day”.
“I think the market reaction is overblown,” said Thomas Rice, a portfolio manager at Minotaur Capital, a fund which owns Nvidia and Microsoft shares. “We see this as a buying opportunity … nobody is looking at the efficiency improvements from DeepSeek and saying, ‘That’s it, AI is solved, pack it up’.” Mr Rice said he was buying more Nvidia shares.
The Nasdaq-listed chipmaker, meanwhile, said the emergence of DeepSeek was positive for its prospects. “DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely available models and compute that is fully export control compliant,” Nvidia said in a statement.
“Inference requires significant numbers of Nvidia GPUs and high-performance networking. We are now scaling in three dimensions: pre-training, post-training, and the new test-time scaling.”
Inference is the word used to describe the way AI models make decisions. DeepSeek is a reasoning model, meaning it uses an approach which allows it to break down complex queries into smaller, manageable tasks. The AI works through problems more like a human in a chain of thought.
With the surge of interest, DeepSeek said on Tuesday that it was restricting new sign-ups. The company also claimed that it had been the target of “large-scale malicious attacks” without providing details.
Licensed by Copyright Agency. You must not copy this work without permission.
