�׸���Ƭ

Before January 27 2025, it’s fair to say that Chinese tech company DeepSeek was flying under the radar. And then it came dramatically into view.

Suddenly, everyone was talking about it – not least the shareholders and executives at US tech firms like Nvidia, Microsoft and Google, which all saw their thanks to the success of this AI startup research lab.

Founded by a successful Chinese hedge fund manager, the lab has taken a different approach to artificial intelligence. One of the major differences is cost.

The development costs for Open AI’s ChatGPT-4 were said to be (£81 million). DeepSeek’s R1 model – which is used to generate content, solve logic problems and create computer code – was reportedly made using much fewer, less powerful computer chips than the likes of GPT-4, resulting in costs claimed (but unverified) to be as low as .

This has both financial and geopolitical effects. China is on importing the most advanced computer chips. But the fact that a Chinese startup has been able to build such an advanced model raises questions about the effectiveness of these sanctions, and whether Chinese innovators can work around them.

The timing of DeepSeek’s new release on January 20, as Donald Trump was being sworn in as president, in AI. Trump responded by describing the moment as a .

From a financial point of view, the most noticeable effect may be on consumers. Unlike rivals such as OpenAI, which recently began charging for access to their premium models, DeepSeek’s comparable tools are currently free. They are also “open source”, allowing anyone to poke around in the code and reconfigure things as they wish.

Low costs of development and efficient use of hardware seem to have afforded DeepSeek this cost advantage, and have already forced some Chinese rivals to . Consumers should anticipate lower costs from other AI services too.

Artificial investment

Longer term – which, in the AI industry, can still be remarkably soon – the success of DeepSeek could have a big impact on AI investment.

This is because so far, almost all of the big AI companies – OpenAI, Meta, Google – have been struggling to and be profitable.

Until now, this was not necessarily a problem. Companies like X (formerly Twitter) and Uber went years without making profits, prioritising a commanding market share (lots of users) instead.

And companies like . In exchange for continuous investment from hedge funds and other organisations, they promise to build even more powerful models.

These models, the business pitch probably goes, will massively boost productivity and then profitability for businesses, which will end up happy to pay for AI products. In the meantime, all the tech companies need to do is collect more data, buy more powerful chips (and more of them), and develop their models for longer.

But this costs a lot of money.

Nvidia’s Blackwell chip – the world’s most powerful AI chip to date – costs around , and AI companies often need tens of thousands of them. But up to now, AI companies haven’t really struggled to attract the necessary investment, even if the sums are huge.

DeepSeek might change all this.

By demonstrating that innovations with existing (and perhaps less advanced) hardware can achieve similar performance, it has given a warning that throwing money at AI is not guaranteed to pay off.

For example, prior to January 20, it may have been assumed that the most advanced AI models require massive data centres and other infrastructure. This meant the likes of Google, Microsoft and OpenAI would face limited competition because of the high barriers (the vast expense) to enter this industry.

Money worries

But if those barriers to entry are much lower than everyone thinks – as DeepSeek’s success suggests – then many massive AI investments suddenly look a lot riskier. Hence the abrupt effect on big tech share prices.

Shares in chip maker Nvidia and ASML, which creates the machines needed to manufacture advanced chips, also saw its share price fall. (While there has been a slight bounceback in Nvidia’s stock price, it appears to have settled below its previous highs, reflecting a new market reality.)

Nvidia and ASML are that make the tools necessary to create a product, rather than the product itself. (The term comes from the idea that in a gold rush, the only person guaranteed to make money is the one selling the picks and shovels.)

The “shovels” they sell are chips and chip-making equipment. The fall in their share prices came from the sense that if DeepSeek’s much cheaper approach works, the billions of dollars of future sales that investors have priced into these companies may not materialise.

For the likes of Microsoft, Google and Meta (OpenAI is not publicly traded), the cost of building advanced AI may now have fallen, meaning these firms will have to spend less to remain competitive. That, for them, could be a good thing.

But there is now doubt as to whether these companies can successfully monetise their AI programmes.

US stocks make up a historically right now, and technology companies make up a historically large percentage of the . Losses in this industry might force investors to sell off other investments to cover their losses in tech, leading to a whole-market downturn.

And it shouldn’t have come as a surprise. In 2023, a warned that the AI industry was exposed to outsider disruption. The memo argued that AI companies “had no moat” – no protection – against rival models. DeepSeek’s success may be the proof that this is true.