'DeepSeek is amazing news for startups': How the much-hyped Al model could help impact companies

When Wall Street started digesting the implications of DeepSeek’s latest AI model, it quickly led to a stock market sell-off that wiped €1 trillion off the biggest U.S. tech giants like Nvidia.

But for smaller companies like impact startups, however, DeepSeek's chatbot could provide plenty of reasons to celebrate.

Some experts say the new open-source AI model could open up massive new opportunities for smaller companies and research labs by drastically reducing the cost and energy-requirements to train their own agents.

To put it another way: the moat around cutting-edge AI models is disappearing and the technology is being democratised.

"Amazing news for startups"

“Deepseek is amazing news for startups. It breaks with the idea that you need billions to even compete in AI,” says Philipp Moehring, co-founder of Berlin-based VC firm Tiny Supercomputer Investment Company, which specialises in backing small European tech startups, writing on X. “It reignites the idea that people are more important than money.”

Training large-scale AI models like OpenAI’s ChatGPT or Meta’s Llama has previously cost hundreds of millions of euros and required access to thousands of Nvidia’s most advanced chips.

Not to mention massive amounts of electricity.

It’s part of the reason why the Trump administration recently announced a $500 billion investment in infrastructure tied to AI, including the construction of more data centres.

But the news that DeepSeek supposedly trained its latest model on hardware costing around €6 million while using less advanced chips has the potential to be much cheaper – especially for companies in Europe and Asia.

Cheaper costs

“DeepSeek has just inverted in an instant some of the most fundamental assumptions about the U.S. lead in technology, the global economy, and the future of energy,” Rory Stewart, a former British government minister and current Yale professor, writes on X.

Nicholas Thompson, the CEO of the Atlantic, also says in a video posted on LinkedIn that DeepSeek could be "a really positive thing" in terms of making AI more accessible.

“It took DeepSeek $6 million to train their last model, there are lots of companies that can afford doing that. Many more than can afford $100 million,” Thompson says.

The cost and energy-requirements for using DeepSeek are also lower, in part because of its method of using a so-called Mixture of Experts design. While other major LLM models have to activate the entire model – often using hundreds of billions of parameters – to answer each individual query, DeepSeek has designed its network so that certain parts are experts in specific areas. That means only a fraction of the model’s parameters are used for each query, drastically reducing the computing power needed to find the best answer.

Its approach to reinforcement learning also means that less human input is needed to train a model to solve logical problems within a certain area.

That’s especially beneficial for smaller companies and researchers who only need their AI model to be an expert on solving certain types of problems.

Scepticism continues

There are still reasons to be sceptical about DeepSeek’s impact on the AI landscape. Some experts are questioning the Chinese firm’s claims about how much it cost to train its chatbot, while others have expressed concerns about whether the model can freely generate responses that violate Chinese government restrictions.

It’s also not clear what impact it could have on the environment. Some experts argue that DeepSeek could reduce the need for data centres that use up massive amounts of electricity.

However, Vic Shao, the founder of three cleantech companies including AMPLY Power, Inc. and Green Charge Networks, says DeepSeek could actually create a larger electricity demand as more and more companies and organisations use its AI capabilities.

“(DeepSeek) is a net positive for AI, as efficient use of power and energy is much needed. However I don’t want those of us in the energy sector to breath a sigh of relief and think demand will go down - in fact I think just the opposite,” Shao writes on LinkedIn. “As LLMs are proving to be useful in business and everyday life, and costs are driven down over time, utilization will only grow exponentially.”