China had a Sputnik moment this month when an obscure startup company, DeepSeek, released an artificial intelligence model that outperformed the biggest and best models US giants such as Meta and OpenAI had to offer.
The R1 reasoning model from DeepSeek is up to 50 times more efficient in terms of computer processing required. The company claims it produced R1 for US$6 million, a fraction of the cost spent on building and training the US large language models currently in vogue.
As the retired Microsoft engineer who runs the popular YouTube channel Dave’s Garage put it: “If you can throw together a Ferrari in your shop on your own and it’s really just as good as a regular Ferrari, what do you think that does to Ferrari prices?”
DeepSeek is the circuit breaker AI needed. It has lowered the barriers to entry in a field dominated by a handful of powerful US companies. R1 is an open-source model, unlike most US rivals. It means anyone can download the software and run the model to perform AI tasks relatively cheaply.
New Zealand is about to create a new public research organisation tasked with undertaking research and development into AI. Before DeepSeek, it seemed a tough proposition. However, the Chinese model proves that useful AI can be developed without requiring massive computing capacity. DeepSeek and similar open-source models will now flood the market, giving us affordable tools to tackle big challenges facing the country. We can also build our own models.
It doesn’t mean an end to the AI arms race and the dominance of the small handful of American companies that have poured hundreds of billions of dollars into building AI data centres to train their large language models.
They have access to the world’s highest-performing computer chips, which are off limits to the Chinese due to a US trade embargo. They will use what they’ve learnt from the Chinese to advance the field of AI yet again.
Nor will it erode the power of the tech giants which, with Donald Trump’s return to the White House, are set to enjoy a return to the low-regulation, laissez-faire regime that characterised the last 25 years of Silicon Valley innovation.
In his new book Techno Feudalism: What Killed Capitalism, Greece’s former finance minister suggests the likes of Amazon, Meta and Google have created billions of cloud serfs who volunteer their labour in return for free services such as the Alexa digital assistant and Google search engine.
Yanis Varoufakis told Wired, “Every time you upload a video on TikTok, on Facebook, on Instagram, you’re adding to the capital stock of these companies.”
“In that way, we’re serfs, but we are more than serfs, we’re cloud serfs producing capital. And that has never happened in the history of the world.”
However, DeepSeek’s disruptive arrival poses challenges to the existing business models underpinning AI. We saw a flavour of that with the trillion-dollar drop in market value of tech companies following DeepSeek’s release.
Premium tools such as ChatGPT Plus, Perplexity Pro and Copilot Pro cost from $25 to $38 for a monthly subscription. That’s a hefty sum.
If the price of a powerful AI assistant drops to $10 a month, that’s a win for consumers, putting sophisticated tools in the hands of many more people. But it isn’t good for the likes of Microsoft, which this year alone plans to invest US$80 billion in data centres to meet the growing demand for AI. It faces the prospect of being undercut by cheaper competitors turning to open-source models.
Kiwis know better than most that necessity is the mother of invention. DeepSeek reminds us that working on a small scale with limited resources can yield big results.
