Data centre, uranium casualties on ASX amid DeepSeek-induced rout

Australian shares have largely avoided the tech-induced wreck across the US and Japan after investor panic at the emergence of Chinese artificial intelligence start-up DeepSeek.

Revelations the company had supposedly built a machine-learning model with fewer resources than the likes of OpenAI, Microsoft and Google spooked investors, and caused a record-breaking plunge for NASDAQ-listed US chipmaker Nvidia.

About $US593 billion ($944b) was wiped out from the hugely popular stock’s market value and marked the deepest ever one-day loss for a company on Wall Street. That dragged down the Nasdaq index down 3 per cent, its biggest one-day percentage drop since December 18.

Japan’s tech-laden Nikkei lost more than 1.45 per cent on Tuesday.

But the ASX largely survived the sell-off, Capital.com senior market analyst Kyle Rodda said — the exchange’s lack of large tech and AI players had been its saving grace.

“The ASX200 has been relatively well insulated from the sell-off on Wall Street, largely for the same reason it has underperformed US indices in recent years: a lack of large tech companies and artificial intelligence exposure,” he said.

The S&P/ASX200 lost 0.12 per cent, or 9.8 points on Tuesday to 8399.1.

Despite a dire session for US markets, Betashares senior investment strategist Cameron Gleesonsaid there was “no guarantees” that this was the end of the road for a record run that had benefited Australian investors with exposure to the sector through their super funds.

“The game isn’t certainly over,” he said.

Companies associated with data centres and uranium fell hard.

Mr Gleeson said the likes of industrial property and digital infrastructure player Goodman Group — which fell more than 8.18 per cent to $35 on Tuesday — had been affected because of the link between data centres and AI as a huge consumer of the storage they provide.

“Before the weekend, the base assumption was that we’re going to see exponential growth in the build out of of data centres, because that’s what was required to build more advanced AI models,” he said.

“And since the weekend, the question’s been asked. Perhaps we will still need to see a build out of data centres in order to house computing power to build more advanced models. But perhaps we won’t.”

He suggested Deepseek’s discovery of a new path for training AI models could lower the trajectory of growth in data centres from “exponential to somewhat more normalised.”

“Some of these data centre companies were priced off high expectations.”

Digico Infrastructure was down 11.53 per cent to $4.22, and NextDC fell 7.23 per cent to be $14.75 by market close.

Australian uranium stocks copped the biggest losses of the day, with WA-based Deep Yellow falling 15.46 per cent to $1.23, followed by South Australian yellowcake miner Boss Energy, down 10.41 per cent to $2.84 and Paladin Energy down 9.44 per cent to $8.15.

Nuclear power has been floated as a potential way to fuel energy hungry data centres, with both Microsoft and Google signing deals back in to potentially use the energy source for their AI and data centre ambitions.

“There was quite a lot of interest from these AI players,” Mr Gleeson said, but he questioned whether it would be feasible for a country like the US to build and house data centres compared with the likes of China and India.

Mr Gleeson said the rise of a potentially cheaper artificial intelligence model that requires fewer graphics processing units (GPUs), or chips, could have implications for big tech players in the US.

“We’ve seen a bit of a bottleneck the demand for GPUs from Nvidia and other semiconductor companies, really creating an inflation in their price. And we’ve got to a point where there’s been an innovation that’s come through that’s meant you don’t need as much of these GPUs, these processing units, in order to train a model,” he said.

“This is actually a good thing for AI. It means it’s cheaper to train, therefore it will be easier to create more applications that can be put to use.”