DeepSeek, which is free and costs far less to develop, appears to be as good as ChatGPT.Dado Ruvic/Reuters
John Rapley is an author and academic who divides his time among London, Johannesburg and Ottawa. His books include Why Empires Fall (Yale University Press, 2023) and Twilight of the Money Gods (Simon and Schuster, 2017).
If Nvidia was the Goliath of artificial intelligence, a wee David in China knocked it for an eight-count. While it’ll take some time to fully digest what has happened, the events of the last couple of weeks may comprise a major turning point.
The simplest way to put what happened is to say that a Chinese firm little-known outside industry circles, called DeepSeek, produced a knockoff AI model that appears to be as good as, if not better, than ChatGPT, but is available for free and costs far less to develop than other emerging AI models. Moreover, rather than using proprietary technology, the makers of the new DeepSeek model, called R1, have made it open source, allowing developers to build upon it or integrate it into their own software at little cost.
Investors are still trying to sort through DeepSeek’s claims, asking if they’re too good to be true. There’s no question the technology is real, since developers are able to play with it and are pretty impressed. But Western firms and market analysts have raised doubts that it really was done on the cheap, with some suggestions that the Chinese government was secretly subsidizing DeepSeek or that it used a lot more Nvidia chips than supposed. OpenAI, the owner of ChatGPT, says the Chinese firm may have stolen some of its intellectual property, something repeated by Donald Trump’s ‘AI Tsar’, David Sacks.
The United States economy isn’t as tough as Donald Trump thinks
But until such evidence emerges, these counterclaims deserve to be treated with the same skepticism as DeepSeek’s own statements. Not only would folks such as OpenAI’s chief executive Sam Altman and Mr. Sacks say that, since they stand to lose so much, but OpenAI’s claims of property theft have elicited no small amount of laughter, since it stands widely accused of itself stealing reams of data to develop its large languages models. What’s sauce for the goose is sauce for the gander. Besides, other startups, including Mistral, have already made progress in the direction DeepSeek has gone in developing AI models efficiently, so the Chinese company’s claims are not implausible.
Assuming DeepSeek is all that it says it is, it’s good news for most of us. Being able to access cheap AI models means mostly everyone, from governments to small businesses, can start integrating AI into their operations in low-cost ways that can make marginal improvements to their efficiency. Since the cost remains so low, the barriers to entry have tumbled. AI may now really begin to transform our lives, improving efficiency in many fields from the public sector to medicine, in small ways and large.
But there may be a few big, indeed very big losers here – namely the American tech oligarchs and possibly the U.S. economy itself. Until now, the presumption in global markets was that AI would be hugely costly to develop. This is what drove Nvidia to make up nearly 15 per cent of total U.S. stock-market capitalization – the assumption that the massive demand for its chips, of which it was essentially a monopoly-supplier, made it a money-printing machine.
The same went for the companies, such as OpenAI and the Magnificent Seven (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, Tesla), which were developing the AI models that were expected to dominate the future. Their enormous share-price valuations were justified on the grounds that they needed huge amounts of capital to break through into the new age of artificial general intelligence, but that the returns on doing so would pay off to investors.
Suddenly, all this is in doubt. With private equity stumping up so much of the money behind the U.S. tech startups, the immediate impact on stock markets will remain muted for now: Falling share prices won’t themselves stanch the flow of capital. After plunging on Monday, U.S. markets gained back some lost ground on Monday, led by a rally in Nvidia’s share.
Still, the Magnificent Seven, which have been able to retain the faith of investors that their promises will come good, are now under pressure to show their AI really will pay dividends, and be sufficiently unrivalled to justify their premium. Meta’s Mark Zuckerberg started the clock ticking this week when he told his investors they’d see results this year. If he disappoints and the private valuations on AI startups then begin to fall, it will have ripple effects, as private equity is forced to cover losses by selling other assets into the public domain.
Rather than a crash, what could result would probably be a bear market in U.S. stocks. The narrative of ‘American exceptionalism’ would get rewritten. The U.S. has been sucking in capital from around the world, as investors abandoned other markets in the belief America was pulling away from the rest of the developed countries. If that faith evaporates and the flow reverses, the U.S. could start to mimic their moribund economies pretty quickly.