In late January, DeepSeek publicly released a generative AI tool that creates text and code from user prompts, at a quality comparable to American tech companies, but at only a fraction of the production cost.Dado Ruvic/Reuters
For the past few years, some of the biggest names in the tech industry have tried to cast a spell on society. “Artificial intelligence is the most transformative technology of all time!” they have cried. “You must give us vast sums of money and all the energy we need, or the economy will be doomed!”
And then along came a Chinese startup that – like Toto pulling back the curtain in the Wizard of Oz – broke the illusion. In doing so, it allows for a more clear-eyed look at the actual benefits and costs of generative AI.
In late January, an obscure tech company called DeepSeek publicly released a generative AI tool that creates text and code from user prompts, at a quality comparable to American tech companies, but at only a fraction of the production cost.
There is likely some illusion in the estimate that training DeepSeek’s AI model was done at only US$5.6-million. But the public release of the model’s code showed it made novel innovations in processing efficiency, requiring less hardware and power than U.S. models.
It stood in stark contrast to an announcement from leading U.S. firm OpenAI and Japanese investing giant SoftBank that they would build a US$500-billion (yes, billion with a b) data-centre project in Texas to “secure American leadership in AI.”
The project – called “Stargate” – was merely the latest in a long line of very big-ticket pledges for AI spending. Meta founder Mark Zuckerberg, for example, said his company planned to spend US$60-billion this year alone on AI infrastructure.
Implicit in all these promises – along with asks of investors and governments to join in the spending spree – was that the technology must be truly transformative if it demanded this many resources. The Canadian government, for one, promised in December to invest $240-million in a new AI data centre.
The field’s leading voices, such as Sam Altman of OpenAI, have always said some stunning new use of the tech – one that would justify the huge financial and energy costs – is just around the corner.
(It should also be noted that these companies have brushed aside well founded concerns that their models were trained on copyrighted material, which makes Mr. Altman’s concern that DeepSeek could have ripped off OpenAI a little rich.)
The emergence of DeepSeek challenges, first, the narrative that the industry needs all these resources. Perhaps these innovative companies can, if they put their brilliant minds to it, do more with less – just like most other sectors have to.
It also pulls back some of the mystique of these companies and allows us to evaluate them more clearly. Just what are they, in the end, delivering?
Goldman Sachs researchers last summer were some of the few to point out that generative AI had not yet generated either the revenue or productivity gains to justify such massive private and public investments.
But if their expenses go down, maybe the return on investment will start to make sense.
Artificial intelligence has been shown to have some useful applications. (And the use of the term “artificial intelligence” is being used to mean just about anything with an algorithm these days.)
For example, one AI product is helping farmers optimize lighting, irrigation and harvest timing to get the most out of their crops.
It is being used in medical imaging to process pictures faster and more accurately so more patients can be seen in a day.
And it is being used to deal with low-level customer-service complaints so that human agents can focus on trickier cases.
Those are all ways to increase efficiency and productivity, of the sort our economy badly needs. But are they the sort of innovations that require dedicated nuclear reactors, especially as our electrical grid becomes strained with more pressing needs in future years? Well, maybe not.
As this space has noted before, the near future of AI has always looked more like Clippy, the anthropomorphic paper clip in Microsoft Office, luckily, than it does the Terminator.
And lately it looks less like a wizard and more like an ordinary man behind the curtain.