Everyone’s Talking About DeepSeek: Industry Takes You Might Have Missed

Unless you’ve been completely off the grid this past week, you’ve probably seen the headlines about DeepSeek. The Chinese AI app has shaken up the tech world, grabbing everyone’s attention, shooting to the top of Apple Store downloads and rattling markets in the process.

DeepSeek’s ability to rival the power of leading AI models like OpenAI’s – yet reportedly at a fraction of the cost – caused a stir among investors, who were shocked to learn that it was developed using fewer advanced chips, undercutting assumptions about the cost and resources required to power top-tier AI.

The release of DeepSeek’s latest version on 20 January, R1, saw nearly $600billion wiped off Nvidia’s market value – the largest single-day loss in US history – after DeepSeek’s use of fewer advanced chips raised concerns about reduced demand for Nvidia’s high-end hardware.

The surge in interest surrounding DeepSeek has not only been evident in stock market movements but also in global search trends. DeepSeek’s Google Trends score skyrocketed 50-fold, outpacing even ChatGPT in the US and taking top spots in regions like Hong Kong and Singapore.

With so much buzz around DeepSeek, we’ve heard from plenty of industry experts this week. Here’s our pick of the initial insights you might have missed.

‘DeepSeek won’t replace US giants, but will drive efficiency’

Anton Chashchin, founder & CEO of N7 Capital, a private fintech group, believes DeepSeek’s success challenges AI costs but won’t replace US leaders like Nvidia and Microsoft. Instead, it will push them to improve, making AI smarter and more efficient.

“DeepSeek’s success shows that AI breakthroughs can come from anywhere, not just the US,” he said. “Its lower-cost model is a challenge, but American companies like OpenAI still have big advantages – strong customer bases, cloud services, and deep industry ties. They will likely adjust by making AI more efficient and affordable.”

‘DeepSeek signals a new era of AI innovation and accessibility’

According to Krishna Venkatraman, chief data officer at LatAm consumer lender Kueski, DeepSeek is changing the game by showing that the high costs of AI training and inference don’t have to be the norm. He points out that through smart optimisations and sparse model architectures, DeepSeek has reduced inference costs while maintaining the power of advanced reasoning models.

“Fasten your seat belts – we may be entering a new phase of accelerated innovation,” he said. “The lower compute demands of inference and the open source model weights for DeepSeek’s R1 model means that companies can get the power of a state-of-the-art reasoning model at a relatively trivial cost. This will open the door to far more companies looking to explore the value that these models can bring to their business. I expect this to be true for fintech as well.

“It is time for companies who have stayed on the sidelines to jump in and get started in evaluating GenAI capabilities in earnest for their individual use cases.”

‘DeepSeek proves that strategy and agility can surpass raw computing power’

Dr Kjell Carlsson, head of AI strategy at Domino Data Lab, a provider of an enterprise AI platform, sees DeepSeek’s success as a powerful reminder that AI’s true value comes from leveraging it effectively, not from the size of infrastructure or the exclusivity of models.

“For companies seeking to maximise value from AI, the lesson is clear: success hinges on flexibility and capability, not exclusive partnerships or infrastructure scale,” said Dr Carlsson. “Rather than locking into specific LLM providers or focusing solely on hardware access, organisations should prioritise building the end-to-end capabilities to source AI innovations, design solutions tailored to their unique needs, and operationalise them effectively.”

‘DeepSeek shifts the focus to efficient, scalable AI solutions’

DeepSeek’s breakthroughs underline a shift in the AI landscape, pushing opportunities beyond the narrow focus on semiconductors and infrastructure, according to Rahul Bhushan, managing director of investment management company ARK Invest Europe.

He explains that with its V3 model matching GPT-4’s performance using just five per cent of the GPU compute and R-1 delivering at a fraction of GPT-1’s cost, DeepSeek proves that AI’s future lies not in throwing more GPUs at the problem, but in creating efficient, scalable solutions.

“Moreover, the fact that DeepSeek’s innovations are open source cannot be overstated,” said Bhushan. “This move opens the door to widespread adoption and decentralisation, a trend that could democratise AI access and accelerate progress far beyond traditional players in the West.

“It also hints at China’s growing strategic ingenuity in shaping the AI landscape under constrained circumstances. We strongly urge investors to re-evaluate their AI funds and positions. Focusing solely on semiconductors risks being materially underexposed to where the real opportunities are emerging: scalable, efficient AI solutions and the open-source ecosystems enabling them.”

‘DeepSeek’s ‘more with less” strategy opens doors for broader AI adoption’

Giuseppe Sette, president of Reflexivity, a market research firm, says DeepSeek’s breakthrough has set the stage for broader AI adoption. He points out how the model saves both money and computation power by activating only the most relevant portions for each query.

“This shows that with AI the surprises will keep on coming in the next few years. And even though that might be a bit of a shocker today, it’s extremely bullish in the long-term – because it opens the way for deeper and broader adoption of AI at all scales.”

‘DeepSeek presents both risks and opportunities for AI and tech’

DeepSeek’s achievements are significant but should come with some caveats, suggests John Belton, portfolio manager at asset management firm Gabelli Funds.  While the model showcases cheaper training and inference methods, he questions the transparency and accuracy of the data and potential licensing issue.

“What DeepSeek has achieved is ‘real’ to an extent,” comments Belton. “Some of the headline figures are a bit misleading (for example, no, you cannot recreate DeepSeek with $6million) and the extent to which they distilled existing models (took shortcuts potentially without licence) is an unknown. But there are a few clear breakthroughs which show the industry a path to cheaper means of both training and inference.”