Squiz Shortcuts - DeepSeek and the US/China AI race

Your Shortcut to DeepSeek and the US/China AI race…

A new Chinese artificial intelligence app sent the tech world into a spin this week when investors realised a competitor to ChatGPT had been produced at a fraction of the cost… Almost a trillion US dollars was wiped from global markets within days, and questions are now being asked about the huge sums of money spent on AI by US tech giants. So in this Squiz Shortcut, we’ll look at:

• what the Chinese model - called DeepSeek R1 - does

• who’s behind it

• and what it could mean for the AI race between the US and China.

Prefer this in your ears?

Listen to our podcast 🎧

Listen time: 9 minutes

Let your smile shine

Smiling is one of our favourite things to do, but it helps when your teeth are at their best. The Smilie Teeth Whitening Kit offers visible results from the first treatment without the hefty price tag, delivered straight to your door. It'll even lift long-lasting stains without causing your toothy-pegs pain. Find out more here.

Squiz the Shortcut

What is DeepSeek R1 and why is everyone talking about it?
It’s the new AI chatbot on the block. It’s made by a Chinese company called DeepSeek, which, until a couple of days ago, wasn’t a very big name, even within tech circles. But that all changed this week.

What happened?
DeepSeek revealed that it built R1, which has similar capabilities as Open AI’s ChatGPT, but it’s much cheaper to make. 

How cheap are we talking?
They reckon it cost them US$5.6 million to build. That’s a whole heap less than the hundreds of millions of dollars invested into more well-known models like Open AI’s ChatGPT-4, Google’s Gemini, and Meta’s Llama. 

But, is it actually good?
Reports say it can do the same things as ChatGPT, so it has reasoning powers and it can generate text, summarise large passages and build reports - and it’s free to download, so it’s proving to be super popular in app stores. 

Exactly how popular?
As word spread, R1 shot to the top of app stores around the world, and by Tuesday, its downloads had passed 3 million. And that sudden surge in interest had a huge impact on tech stocks, as investors dumped those shares in droves. 

Can you explain how a major development in tech caused those stocks to tank?
In a nutshell, investors freaked out. Tech stocks relating to AI and data storage fell across the board - including on the ASX - but one US company was hit particularly hard. 

Which one was that?
Nvidia, which makes computer chips specifically for use in AI, had nearly US$600 billion wiped from its bottom line on Monday. That amount set a new record for the biggest loss on the US stock exchange in a single day, and it was because some of its chips with simpler designs were used by DeepSeek to make the R1 chatbot - just a whole lot less of them than it usually takes… 

How many are we talking about?
While it usually takes more than 16,000 chips to make a chatbot such as ChatGPT, DeepSeek says they’ve done it using only 2,000 chips. But far from being intimidated by the Chinese app, Nvidia said R1 was an "excellent AI advancement", and to meet with demand, DeepSeek would be needing many more of its chips in future. 

Staying positive then…
And potentially for good reason - experts say that demand for chips could see Nvidia’s value bounce back - which is starting to happen already. The other important piece of all this is that Nvidia’s skyrocketing value last year also had a lot to do with its key role in US AI tech… 

What was that?
Former US president Joe Biden introduced policies around AI that prevented American companies selling or manufacturing those chips in China, which the US views as its main rival in the field. It’s an emerging, fast-growing field and he said he wanted to make sure the US was leading the game. But it seems that the Chinese - who’ve also had domestic policies in place to fund AI development - have found ways around the restrictions. 

How did they do it?
They say they’ve created R1 by using Nvidia’s widely available, lower-powered “export-compliant” chips. The thing about this is that those chips cost a lot less than the advanced ones and they use a lot less power. The timing of DeepSeek’s launch of R1 has also got tongues wagging…

Why is that?
Some analysts have said it wasn’t a coincidence that R1 was launched on the same day as Donald Trump’s inauguration. They say that China was sending a clear message to Trump and his tech exec pals that trade restrictions and tariffs won’t stop their progress.

Who’s behind DeepSeek?
It’s owned by Chinese billionaire Liang Wenfeng, a Chinese billionaire who’s used his hedge fund, High-Flyer, to invest in AI. And while it’s just over a year old, it’s made fast progress. It launched other large language models last year - aka chatbots with similar capabilities as ChatGPT - but R1 has really shaken things up.

Is it just down to the low cost?
That’s a big part of it, but another reason it’s seen as an industry “disruptor” is that it’s built on an open source, meaning anyone can download the code for free, test it and build on the model to improve it… 

Have those figures been tested?
Not yet. The numbers quoted by DeepSeek haven’t been scrutinised and analysts say they don’t include the costs of training the chatbot, and there could be research and development numbers also not accounted for. But even so, reports say those costs aren’t likely to reach into the hundreds of millions. One other thing on that - Open AI says it’s looking into claims that DeepSeek might have harvested its data - without authorisation - to make R1. 

What else should I know?
There are some big blind spots in the app’s responses when it comes to topics sensitive to China like Tiananmen Square and Taiwan that it can’t or won’t answer, which could point to censorship. 

Got it… So, what does DeepSeek’s R1 mean for the AI race between America and China? 
If the figures stack up, reports say it could be a game-changer in that it means AI can be built for a lot less moolah than the US tech giants thought - and it could require a lot less energy. It comes at a time when both Open AI and Meta have made recent funding announcements to the tune of billions of dollars - even buying nuclear power companies to help run their AI models which they say are power-hungry and need a lot of electricity. 

And didn’t Trump make a recent AI announcement too?
Yep. He announced his US$500 billion Stargate partnership with Open AI, Oracle, and SoftBank, last week to fund AI development and build huge data storage centres across the US.

What was Trump’s reaction to the R1 launch?
He says it "should be a wake-up call” for US industries. But he also said it "could be a positive development" by paving the way for new innovations, and that instead of spending billions, tech companies would spend less, and “come up with, hopefully, the same solution.”

What happens in the meantime?
Until then, there’s some talk that Trump might extend the US government’s export bans to include even simpler computer chips to obstruct China’s AI production… And there are warnings for any Aussie users downloading the app to be aware of privacy and data security concerns associated with Chinese-owned tech…

What sort of concerns?
Not be be a Debbie Downer, but they’re the same jitters our government and cybersecurity experts have raised in the past about TikTok and Huawei - they say because the firms are linked to the Chinese state, it could lead to people’s information being harvested, so they’re urging Aussies to “be very careful” using the app. 

Sounds like the AI race is heating up?
Yep, analysts say the US tech giants have got some work to do, so watch this space…

Recent Shortcuts