The AI arms race is frequently compared to the space race and nuclear stockpiling that took place during the Cold War – a ‘Moloch trap’ where two opposing governments race to develop the most powerful tools, prioritising victory over anything else. For a long time, the US seemed to be winning the race, courtesy of pioneering AI companies like OpenAI, Google DeepMind, and Anthropic. This week, though, a Chinese company called DeepSeek appeared to blow them out of the water with the public release of a “reasoning model” called DeepSeek-R1.

Reportedly, DeepSeek’s debut wiped $1 trillion off the value of US stocks, in what Donald Trump called a “wake-up call” for Silicon Valley. But what actually happened? Is the hype really worth it? Did China just win the AI arms race? We dig deeper in the the DeepSeek debacle, below.

WHAT IS DEEPSEEK, EXACTLY?

Bearing a small blue whale as its logo, DeepSeek-R1 burst onto the scene without much warning. The company that created the model – also called DeepSeek – was founded in December 2023, and its latest version of the app was released on January 20. As a large language model (or LLM) it’s built on similar technology to familiar US-made chatbots like ChatGPT, Claude, or Gemini. In its first week, though, it shot past the other competitors to hit the top of the Apple App Store charts.

WHAT’S SO GOOD ABOUT IT, THEN?

R1 performs as well as OpenAI’s most advanced, publicly-available reasoning model on certain AI benchmarks – which are a bit like exams, but for AI. However, it supposedly does so at a fraction of the energy and cost, using fewer and less advanced computer chips thanks to innovations in the ‘training’ process. On Monday (January 27) this caused leading chip manufacturer Nvidia to lose almost $600 billion in market value, AKA the biggest one-day loss in stock market history.

R1 is also open-source, and free, which probably explains why it’s hit the top of the App Store charts (for its comparable model, OpenAI charges $20 per month).

THE US-CHINA TECH BATTLE CONTINUES

We probably don’t need to remind you that the DeepSeek drama comes hot on the heels of a (very brief) US TikTok ban. This was largely tied to security concerns about TikTok creator ByteDance’s links to the Chinese government.

People using R1 have already reported running up against instances of censorship as a result of oversight by the Chinese government, which says that AI systems must “embody core socialist values” and not produce content that “damages the national image”. R1 can’t generate responses about Tiananmen Square, or Taiwanese autonomy, for example.

That said, smaller and slightly less powerful versions of R1 can be run on local machines, including something as small as a laptop. In this case, as AI researcher Dean Ball points out, it can be run “far from the eyes of any top-down control regime” – in other words, its responses can’t be filtered or moderated by Chinese authorities.

THE LAUNCH SPARKED NUMEROUS CONSPIRACY THEORIES

It’s unsurprising that DeepSeek has caused some controversy already – wiping out $1 trillion in US stock value will do that. Some believe that the company simply “ripped off” OpenAI (although OpenAI arguably “ripped off” tonnes of copyrighted material to train ChatGPT as well) while others say that its efficiency claims are a “CCP state psyop” designed to undermine US companies.

Experts have also warned against trusting R1 with sensitive data, because you don’t know where it might end up (again, the same could be said of any US company – it’s probably best to avoid feeding your whole identity into any AI chatbot, just to be safe).

At the other end of the scale, DeepSeek has reported being targeted by “large-scale malicious attacks” since making headlines, causing it to limit new signups for its services. It’s not yet revealed the suspected source of the cyberattacks.

WILL DEEPSEEK CHANGE THE AI INDUSTRY FOR GOOD?

Marc Andreessen, a venture capitalist, self-proclaimed accelerationist, and adviser to Donald Trump, called the launch of DeepSeek-R1 AI’s “Sputnik moment”, making a direct reference to the US-USSR space race. Whether or not it proves to be of the same historical importance, R1 might have punctured the Western AI bubble for good, driving down investments that many already believed to be overblown.

OpenAI’s Sam Altman has asked for trillions of dollars to help overhaul the AI chip industry over the last few years, while Trump announced a $500 billion ‘Stargate’ project in the first days of his presidency, aimed at building “the largest AI infrastructure project by far in history”. But will investors be so keen to invest, if a Chinese chatbot can do the same job for a fraction of the cost? Probably not! This leaves companies like OpenAI with some catching up to do. Let’s just hope that they don’t throw caution to the wind in the process – that’s how you get the Terminator future.