Special Reports
A special report is content that is edited and produced by the special reports unit within The Irish Times Content Studio. It is supported by advertisers who may contribute to the report but do not have editorial control.

What’s all the fuss about DeepSeek?

Built at a fraction of the cost of its US peers, the AI software upended the industry

DeepSeek’s launch left investors clutching for the smelling salts. Illustration: iStock
DeepSeek’s launch left investors clutching for the smelling salts. Illustration: iStock

The launch of the DeepSeek R1 reasoning model in January, on the day of Donald Trump’s second inauguration as US president, rattled the tech industry and its investors, wiping billions off stocks.

High performing but cheaper to develop than blockbuster rivals such as OpenAI’s, DeepSeek’s development was a case study in the law of unintended consequences. After all, it was US moves to limit China’s access to high performing chips that forced the latter to innovate.

Developed at a fraction of the cost to build and train than US peers, and requiring significantly less power to operate, DeepSeek upended previous assumptions.

These were not just in relation to the cost of developing large language models but the expensive data centre infrastructure required to support them. DeepSeek reportedly uses just 10 per cent of the power required by its US rivals.

READ MORE

With Ireland’s data centres predicted, by 2027, to use more electricity than all households in the State combined, it’s easy to see why DeepSeek’s launch left investors clutching for the smelling salts.

Nvidia, maker of the GPUs by now synonymous with AI, suddenly looked less unassailable. Though its share price recovered, nearly $600 billion (€550 billion) was wiped off its market cap in a day, a record. On the other hand, things were looking rosier for traditional chip behemoths such as Leixlip’s stalwart Intel.

The fact that DeepSeek’s model is open source, providing developers around the world with access to the code, was only going to fuel further developments, changing the game everywhere.

Maryrose Lyons, founder of Ireland’s AI Institute
Maryrose Lyons, founder of Ireland’s AI Institute

“The major impact has been that the Chinese have developed a new way to create a frontier model, disproving the notion that you must invest a whole lot of money in computing power,” says Maryrose Lyons, founder of Ireland’s AI Institute. “That they have created DeepSeek-R1-Zero and made it available open source has a significant impact because it reduces the gap in quality between open source and paid models. Having such a quality model available for all for free has massive potential for reducing inequality, globally.”

This can make open-source options more viable for enterprises, especially for running operations on confidential data.

“The fact that they were able to develop it in a different way is actually massively game-changing, because there are probably other research houses building by using the DeepSeek method now. They don’t have to have massive billions of dollars for computing power,” she says.

It’s also great to use and, she adds, can be used via Perplexity AI. “Perplexity has a version of the model running on an American server, so none of your data is going to China,” says Lyons, who believes the launch of DeepSeek has forced others, including ChatGPT, which recently brought out its new 4.5 model, to up their game.

For businesses, one of the primary concerns about using any generative artificial intelligence products remains security.

Sam Glynn, founder of Code in Motion
Sam Glynn, founder of Code in Motion

“There is always a question, when you are using any of these tools, whether the information you are providing is secure and private,” says Sam Glynn of CodeInMotion, a consultancy that provides IT security advice, assurance and oversight services to firms operating in regulated sectors.

That includes the questions or prompts you give it. Users may feel it is of little value but in fact, companies can infer a lot from it.

Between prompts and the metadata which have long been collected in relation to your location and IP address, and the kind of devices you use, it is possible to build up enormously detailed pictures of an individual.

With search engines such information is typically used for the purpose of serving up advertisements, but it is a form of surveillance and can have an enormous impact not just on what we buy but even how we vote, he adds.

He points to the Cambridge Analytica scandal of the 2010s, which saw Facebook user data used to influence the outcome of US elections and the Brexit referendum. The rise of GenAI only adds to the risk.

Currently, Irish users rely on EU regulation to protect our privacy but, given that the US tech industry is pushing back against such regulation, and indeed it appears, on the EU itself, the stakes could yet get so much higher.

“We’re on a collision course between the EU’s perspective, which is that all this starts with fundamental human rights and part of that is our right to privacy, fair treatment and transparency,” says Glynn. “Whereas the US starts with capitalism and what works for the tech bros.”

Their much-feted desire to “move fast and break [things]” is all well and good, “until it causes a catastrophic issue, such as bringing down a power grid or a massive data breach”, he says.

He believes the EU should stick to its guns, and its AI Act, and retain its focus on privacy and security, as new GenAI platforms emerge, and the services they offer expands.

Europe’s privacy laws, from GDPR to its AI Act, come from an understanding of “how countries or nations turn against their people, and are trying to make sure that people are protected. If you look at why we have data protection laws, it’s because of what happened with the likes of the Stasi or Nazis years ago,” he says.

“Fundamentally the basis of these things is about trying to protect the individual, whereas all the narrative around AI is that Europe is being left behind because AI is actually all about growth and the economy and capitalism. Not everything Europe is doing right now is exactly right but on this one we should be closer to Berlin than we are to Boston,” he says.

It is why employers urgently need to set policies and protocols around GenAI usage. That includes ensuring there is no “shadow usage”, wherein people use approved AI applications at work but unapproved ones for work while on their laptop or phone at home, he points out.

The need to ensure proper usage is all the more important as use cases more rapidly emerge. These go far beyond personal productivity to the rise of agentic AI, the ability for AI tools to not just turn data into knowledge, but to then turn that knowledge into action, kicking off a chain of process events without human intervention or guidance.

Marc Hanlon, technology consulting director at PwC
Marc Hanlon, technology consulting director at PwC

Like blockchain and non-fungible tokens before it, this year has seen GenAI come down from the top of the hype cycle, the peak point of inflated expectations, even while the practical applications for it are growing fast, points out Marc Hanlon, technology consulting director at PwC.

Its research reveals a surge in innovation and activity to enable AI adoption, with 67 per cent of survey respondents either testing or at partial implementation stages of AI adoption. That’s up from 47 per cent in June 2024.

The survey also found that while businesses are reporting greater confidence about GenAI processes and controls, there is more work to do on trust and governance, to ensure safe and secure outcomes.

Businesses surveyed welcome the EU AI act as an important mitigant to prevent the potential negative impact of AI, but the cyber security risk arising from GenAI remains a major concern, it found.

In another PwC survey, 44 per cent of leaders said they were already seeing productivity gains because of AI, while 24 per cent said they were already seeing profitability gains.

For example, one client uses a chatbot to answer the 50 per cent of queries to its HR department that relate to information that already exists in their HR policies.

“But we are now starting to move into more advanced use cases within process automation,” he says. “For example, a client is using agentic AI for their onboarding process, which involves interactions with lots of different departments. As such we are starting to see enterprise productivity gains.”

In another instance a bank client is using agentic AI to manage customer account queries, leaving personnel to focus on sales calls, reducing customer service costs by 10 per cent and increasing revenues by 5 per cent.

Without referring to any one GenAI product in particular, Hanlon reckons any moves that reduce the cost of either the computing or energy power required to develop or run GenAI tools will only encourage more such use cases. “Any technology that can decrease the cost of consumption is going to help with adoption,” he says.

Sandra O'Connell

Sandra O'Connell

Sandra O'Connell is a contributor to The Irish Times