DeepSeek isn’t the AI game-changer the Chinese think it is
The arrival of China’s market-crashing chatbot has started a dash for AI supremacy that could rival the “space race” of the 1960s – but, as with Sputnik vs Nasa, it isn’t inevitable that the ultimate winner will be the one who is first to make a big splash, says Sean O’Grady
It is no wonder that the launch of DeepSeek, the upstart Chinese artificial intelligence (AI) outfit that wiped a trillion dollars off US tech stocks yesterday, has caused so much turmoil on the financial markets. At least at first glance, it has re-engineered the AI business model. As a result, the high-end advanced chip producer Nvidia alone lost more than $590 billion from its valuation.
Between them, Alphabet, the owner of Google, and Microsoft shed another $160bn. Things move fast and furious in the emerging and perplexing world of AI, and we are witnessing the beginning of a race between the United States and China for AI supremacy to rival the “space race” of the 1960s, between the US and its then rival superpower, the Soviet Union. Hence why people are calling the arrival of DeepSeek as another “Sputnik moment”.
That’s a reference to the successful – and surprise – launch of the Sputnik (it means “satellite” in Russian) by the USSR in 1957. No one in the West expected it. The breakthrough so alarmed the US that the then president, Dwight Eisenhower, ordered an immediate accelerated programme to get an American satellite up as soon as possible. It was followed up by President Kennedy’s pledge in 1961 to get a man on the moon by the end of that decade (Elon Musk now has similar ambitions for Mars).
In tangible terms, the space race gave us, legendarily, non-stick saucepans and a space telescope that can spot tumours and some other benefits from the costly Apollo programme. But the impact of AI will be far more pervasive and dramatic, from simple chatbots that can answer queries, write essays and sort out your banking and finances in seconds, to vast AI computers that can crunch huge quantities of NHS data and suggest life-saving treatments.
This is also a world of “unknown unknowns” – things that AI can create and accomplish that are unimaginable, such as entirely new genres of music, literature and art. AI could one day diagnose and cure your illnesses, test your eyesight, manage your pension fund, drive you home, organise your shopping… Who knows?
The question now is whether the Americans can do it again on this new technological frontier; or will the 21st-century prove to be one of Chinese industrial dominance?
The shock and the contrast is palpable. Last week, Donald Trump – golf player, friend to the billionaire tech bros and president of the United States – announced in typical grandiloquent style the Stargate Project, a $500bn infrastructure investment to support the AI revolution. He was flanked at the White House by some of the leading figures at the centre of the AI revolution – Sam Altman, the 39-year old chief executive of ChatGPT firm OpenAI we’ll hear much more about; Larry Ellison, executive chairman of rival Oracle (he’s older and worth about $200bn); and the boss of the major specialist investor SoftBank, Masayoshi Son.
Trump declared that the project was the “largest AI infrastructure project in history”, and would guarantee American pre-eminence. Microsoft has committed $80bn, and Mark Zuckerberg’s Meta (Facebook-Instagram-WhatsApp) some $65bn. All this is needed for building massive data centres, employing programmes and paying the sizeable electricity bills.
Enter the dragon. DeepSeek is already the most downloaded app in the Apple Store and, according to its founder Liang Wenfeng, who runs the Chinese hedge fund High-Flyer, was built for just $5.6 million – one hundred-thousandth of the $500bn cost of the Yankee version.
Even allowing for some exaggerations and hidden costs, that’s quite the discount. The Americans, and their investors, have been and are wasting their money. So, for that matter, is the UK, with its more modest £14bn plan to boost the country’s AI sector and create some 13,000 new jobs. The Chinese have won the race without anyone noticing, haven’t they?
We shall see. One aspect of China’s Great AI Leap Forward is that it is, ironically, partly the result of official US restrictions on technology transfer. So the story goes, DeepSeek had therefore to use a stockpile of older Nvidia chips, cheaper and less expensive to power than the upgraded versions available in the West. Allied to the generally lower Chinese cost base, it looks like California cannot compete, or at least the rules of the game have altered.
Recent analogies would be, for example, how China’s BYD is challenging Musk’s Tesla for electric car supremacy, Huawei smartphones are a credible alternative to an Apple iPhone or a Samsung, and TikTok virtually invented a whole new social medium that’s so popular, America can’t ban it.
In the case of DeepSeek, a radically different approach has, potentially, altered the paradigm. They claim that their R1 large language model (LLM) can process the endless quantity of books, newspapers, web pages, social media posts and heaven knows what else is needed to make its computers learn to “read”, understand language(s), “think” and converse with humans and other machines – not unlike how a baby grows up and interacts with people and the world around it. If the Chinese have had to find a way to do that with downgraded, cheaper chips, good for them – necessity being the mother of invention.
As Trump says, this has been a “wake-up call” for America, just as Sputnik was for Eisenhower all those decades ago, who launched Nasa in its wake. These developments in China – and there are other emerging players – will affect other parts of the AI sector, which explains, if not necessarily justifies, the slide in equity values.
According to Altman, posting on X (so excuse the typos…), “epseek’s r1 is an impressive model, particularly around what they’re able to deliver for the price. we will obviously deliver much better models and also it’s legit invigorating to have a new competitor! we will pull up some releases.”
The obvious danger with DeepSeek is cultural and political. As has been well noted, ask it about Taiwan or Tiananmen Square, and it will rather pathetically reply that “I am an AI assistant designed to provide helpful and harmless responses”, before trying to change the subject. Such biases could infect its entire output, albeit in a more insidious way. Users may worry that every keystroke they make, smartphone payment, email and website will be logged and transmitted to Beijing – or, even, that the software could turn malicious. It is certainly conceivable.
AI, it hardly needs a chatbot to tell you, is very new, and we are in the earliest stages of the global race for leadership. Some companies and countries will eventually gain the edge, and they may end up with a surprisingly dominant market position, just as Amazon, eBay and the Google search engine did in the past. That, in principle, would not be in America’s strategic economic and security interests. But, if the tech giants can’t put up a fight and win the AI race, and the world wants to use the Chinese AI software, what can be done about DeepSeek?
Waiting for an answer here…
Join our commenting forum
Join thought-provoking conversations, follow other Independent readers and see their replies
16Comments