DeepSeek has quickly become a sensation after moving ahead of ChatGPT on Apple’s app marketplace. Built in China, it has attracted global admiration due to its open-source nature.
Other large-scale language models typically consume extensive data and resources, but DeepSeek takes a different path. This method has made advanced AI more accessible for smaller enterprises and academic labs, thanks to reduced expenses.
Commentators claim it has ignited new debates on possible directions for advanced algorithmic progress. With less need for huge budgets, DeepSeek indicates a different path for quicker and lower-cost innovation.
How Could It Affect UK Startups?
Seasoned professionals at Kreston Reeves, a firm known for financial advice, predict that DeepSeek will transform the investment environment for British AI ventures. They reference its success on limited capital, showing that notable accomplishments do not always require enormous outlays.
Some investors may slow down their spending if comparable achievements are possible in other regions at a fraction of the cost. This prompts questions about how companies might alter their proposals and budget requirements.
Less-established teams may feel motivated to try alternative systems rather than scrambling to gather huge sums.
Is The UK Government Ready?
Officials have introduced a series of proposals named the “AI Action Plan,” by VC Matt Clifford. These recommendations revolve around stronger infrastructure, talent expansion, and data sharing for both public bodies and private companies.
One draft envisions a dedicated “Sovereign AI” entity that would link state agencies with the commercial sphere. Supporters claim it could share crucial datasets, promote joint ventures, and steer resources toward promising pursuits without excessive hurdles.
Skeptics warn that a focus on quick innovation could leave insufficient time for thorough oversight. They raise worries that this rapid style may compromise safety or ethics in the rush to accelerate AI growth.
Experts Predict The Impact On AI Startups
Experts have varied views on what the AI startup’s rise will do to the industry as a whole, this is what they think…
Our Experts:
- Abraham Yousef, Senior Insights Analyst, Sensor Tower
- Richard Robinson, CEO, Robin AI
- Claire Woodcock, VP of Science and ML, Climate X
- Peter van der Putten, Director, AI Lab at Pegasystems, Assistant Professor, AI, Leiden University
- James Bent, VP Solutions Engineering, Virtuoso QA
- Peter Garraghan, CEO and co-founder, Mindgard
- Roch Glowacki, AI Law Specialist and Managing Associate, Lewis Silkin
- Jan Kunkler, Principal Data Scientist, Lobster
- Mark Pearson, Founder, Fuel Ventures
- Jing Jing Xu, Managing Director, Fuel Ventures Asia
- Mark Standen, Director of Technology, Acorn by Synergie
- Megha Kumar, Chief Product Officer and Head of Global Geopolitical Risk, CyXcel
- Chris Beer, Senior Data Journalist, GWI
Abraham Yousef, Senior Insights Analyst, Sensor Tower
“DeepSeek’s swift rise to prominence internationally and within the United States, surpassing 3 million total app downloads since launch, can likely be attributed to increased consumer interest in AI, coupled with its comparisons to well-known US-developed AI chatbots and assistants.
“Prominent US tech companies have notably increased their investment in developing AI models over the past year. However, DeepSeek claims to outperform several leading AI models on various key metrics, reportedly achieving these results at a fraction of the cost.”
Richard Robinson, CEO, Robin AI
“DeepSeek’s R1 release has highlighted a massive shift in the AI landscape, with the cost of intelligence falling faster than many anticipated. This is excellent news for companies like ours that are focused on applying these models to solve real-world business challenges.
“As the cost of foundational intelligence decreases, it unlocks opportunities for the application layer of AI to innovate and create value, which is where we believe the future lies.
“Europe has always had impressive AI talent, but historically the challenges have been around limited access to capital and the lack of a large home market to scale in.
“The beauty of advancements like this is that they lower the barriers to entry for startups, enabling talent from any corner of the globe to compete on a level playing field. AI also helps overcome market-specific challenges like language, making it easier for European startups to make a global impact. I’m optimistic that as the field matures, capital will increasingly flow to where the talent is.
“As for DeepSeek, it’s difficult to assess their achievement fully from the outside. While the costs and effort behind this kind of breakthrough may be underplayed, what really matters is the competition it spurs and the way it accelerates innovation across the industry.
“For us at Robin AI, legal use cases are a bit distinct, as they rely heavily on proprietary datasets often hidden in private databases or on individual lawyer desktops. This means foundational AI breakthroughs, while important, need to be adapted and enhanced to unlock value in specialised fields like ours.”
Claire Woodcock, VP of Science and ML, Climate X
“The DeepSeek announcement demonstrates how fast innovation can move when the academic community are sharing their latest techniques and approaches.
“As with any decision, there are pros and cons to both open and closed source models. The danger with open source is that you rely on the community to honour and adhere to ethical principles, and whilst this argument can also be made for closed source commercial models, these institutions are somewhat easier to hold to account.
“The main benefit of open source is the transparency afforded, which allows researchers to really understand what’s going on under the hood. Another thing I love about open source, and in particular DeepSeek, is that it’s possible to run the model locally, thereby making it easier to keep your personal data private.
“Given the amazing performance of DeepSeek, this could certainly tip the scales towards open source. However, what we consistently see with open source offerings is that companies build products and services above and below the OS element in the value chain, thereby encouraging and locking customers into a particular provider or ecosystem. It will be interesting to see if this rule holds true in the age of LLMs!”
Peter van der Putten, Director, AI Lab at Pegasystems, Assistant Professor, AI, Leiden University, Netherlands
“I am always optimistic about what this means for our local AI startup community, and it is a good reminder that the rules of the game are never set in stone.
“That said, to provide some nuance, DeepSeek was not a small effort. The DeepSeek paper mentions close to 200 (196) collaborators on this work alone, other sources mention the order of 100 FTE’s. The DeepSeek team has published 10+ papers, and whilst it was founded in April 2023, it was a spin off of an AI-based hedge fund trading company.
“So this company is not coming out of nowhere, and whilst claims were made it only cost $5.5M to train, the model wasn’t calculated on the back of a napkin, but considerable kit was used for this.
“As research and market matures and commodifies further, additional efficiency will further drive the down the cost of the base foundation model market. So there will be more opportunity for new entrants, but the more interesting market for startups will be companies that add true value on top of these building blocks.
“DeepSeek’s new reasoning model R1 is not the ground-shaking step change that some have said. It does widen the market for base foundation large language models which is healthy for competition on quality and the cost and level of openness. Nonetheless, DeepSeek is not perfect and there is always room for improvements.
“On the claims about performance and that DeepSeek claims it is competitive in quality with OpenAI’s reasoning model o1. This is impressive but DeepSeek’s model has not been benchmarked yet extensively against o3, OpenAI’s latest flagship model. Nor do questions or problems always require reasoning, regular models such as Goole Gemini and ChatGPT-4o still beat reasoning models such as R1 and o1.
“Reasoning models such as R1, o1 snd o3 challenge the assumption that we have run out of data to create better models as ‘we have used the whole internet’. They make clever use of compute at test time, i.e. when you ask the question, rather than just at training time, by aiming to reason towards finding a solution.
“The most interesting model actually is not the top scoring model, but DeepSeek-R1-Zero model. It is significant because while pretrained on human output, it uses its own reasoning output to further refine the model without human feedback. In theory this could be an avenue towards scaling model performance further.
“This obviously would require balancing efficiency on the other hand, through using parameters at lower precision (quantisation), teaching simpler models with more expensive and capable models (distillation), breaking up the network architecture (mixture of experts) etc.
“So DeepSeek broadens the market but we are far from done yet. Once models get deployed in the real world, through online services, in apps or even inside cars or robots, there is endlessly more data that could be tapped to further scale these models. But of course, we as a society need to ask s where could or should we use these models to drive responsible impact in our economies and people’s lives.”
James Bent, VP Solutions Engineering, Virtuoso QA
“The advanced reasoning capabilities of DeepSeek may affect start-ups positively in terms of increasing accuracy and trust, two of the core acceptance factors in AI.
“The fact that it is Chinese makes data sovereignty and security an issue, in that you can’t opt out of data sharing for model training. This can put pressure on how viable solutions are built with DeepSeek, particularly in Enterprise markets.
“What it does do is open up the market and force other providers, across chips, storage, camera sensors, cloud, smart phone offerings, to up their innovation game.”
Peter Garraghan, CEO and Co-Founder, Mindgard
“DeepSeek is a strong example of the power of open-sourcing AI models. By leveraging insights and contributions from other researchers and open-source releases, they’ve demonstrated how collaboration can drive rapid innovation. What’s remarkable is that their team has achieved a highly advanced release at a fraction of the cost compared to industry giants.
“There’s a narrative circulating about the $6M cost of DeepSeek’s model, but it’s important to clarify that this reported figure represents the estimated compute cost. It does not account for associated staff, discarded training runs, or the thousands of GPU cards used in the process. In any case, this disrupts the prevailing narrative that only companies with the largest budgets can deliver transformative AI advancements.
“DeepSeek’s success should be viewed as both a warning and an opportunity for European AI startups. On one hand, it illustrates that groundbreaking innovation is possible by building on shared knowledge and optimizing resource use. This opens up a path for Europe to develop competitive AI solutions, showcasing the potential of talent-driven, rather than purely capital-driven, innovation.
“However, Europe must address the broader challenge of scaling investments. For instance, individual U.S. companies are spending more on AI infrastructure than entire national budgets in Europe. While the UK’s recent initiatives, such as AI Growth Zones and AI Action plan, are commendable steps in the right direction, much depends on the government’s ability to bridge the gap between vision and execution. We have the ingredients to become an AI powerhouse, but achieving parity in this global race will require more than just ambition.”
Roch Glowacki, AI Law Specialist and Managing Associate, Lewis Silkin
“Some have likened DeepSeek’s rise to a Sputnik moment, which sparked the space race during the Cold War. However, this time round it is not merely a wake-up call but likely a reality check. China has already surpassed the United States in several AI domains, and natural language processing was, arguably, the last stronghold. Regardless of how this story unfolds, China’s asserting its competitiveness in the field of AI. And if DeepSeek proves to be as cost-effective and capable as advertised, even the last AI stronghold may be slipping from the US tech giants.
“This development could be exciting news for the UK and Europe. If cost is no longer a barrier to entering the AI race, our talented entrepreneurs and computer scientists could provide us with a competitive advantage in the long run.
“If DeepSeek’s achievements in algorithmic efficiency prove to be true, they could open doors for new players and fuel a new cycle of innovation. Paradoxically, DeepSeek’s success today may well lay the groundwork for its own displacement tomorrow.”
“OpenAI is currently entangled in several lawsuits for alleged IP infringements. One of the accusations is that DeepSeek has used OpenAI models to train their own model which breaches the terms of use of ChatGPT. It is slightly ironic, considering that OpenAI itself has likely utilised content from numerous sources, potentially infringing on IP rights and the terms of the websites from which such content was obtained. But as ironic as I find it, I am not sure we should say much about these allegations until we know more.”
Jan Kunkler, Principal Data Scientist, Lobster
“The release of Deepseek is signaling a fundamental shift in how we approach AI innovation.
“What’s particularly interesting about DeepSeek’s release is how it challenges the current cost structure of AI services, potentially leading to a more competitive and accessible market. It is notable that it has achieved this on a relatively modest budget, challenging the assumption that breakthrough AI development necessarily requires the massive resources of Silicon Valley giants.
“DeepSeek’s release serves as a wake-up call in an industry that has increasingly focused on accumulating massive capital for compute resources. Its success demonstrates that innovative approaches and efficient resource utilisation can compete with well-funded competitors.”
Mark Pearson, Founder, Fuel Ventures
“For European entrepreneurs, this is a wake-up call that there is a global demand for AI innovation, not just from traditional markets like the US but from markets that might not have been prioritised before. At Fuel Ventures, we are constantly encouraging founders to think globally from day one.
“As for our LP base, there’s a growing awareness of the opportunity and challenge posed by companies like DeepSeek. Many of our investors are increasingly interested in funding European AI startups that focus on areas where we can excel – whether it’s AI ethics, applied AI for vertical-specific industries, or leveraging Europe’s deep academic research talent.
“DeepSeek’s advancements are a reminder that the AI race is truly global, and for Europe, this is both a challenge and an opportunity. With the right level of ambition and investment, we can ensure that European AI startups remain competitive on the world stage.”
Jing Jing Xu, Managing Director, Fuel Ventures Asia
“DeepSeek’s success highlights the power of collaboration in resource-constrained environments, such as China’s limited access to advanced US chip technology. European startups, already experienced in navigating similar constraints, could adopt this approach, leveraging DeepSeek’s breakthroughs to explore new applications and business models. This could foster partnerships within Europe, creating a more resilient and competitive ecosystem.
“Additionally, the buzz surrounding DeepSeek may invigorate European investors, encouraging greater financial support for AI startups. However, this surge in Chinese AI innovation also intensifies pressure on European startups to innovate cost-efficiently, as they must now compete in a market increasingly defined by affordability and scalability. Balancing these opportunities and challenges will be critical for European players aiming to thrive in this rapidly evolving landscape.”
More from Artificial Intelligence
Mark Standen, Director of Technology, Acorn by Synergie
“DeepSeek’s innovative strategy and cost effectiveness are set to disrupt and reshape AI strategies across the board. It makes AI more accessible for everyone, globally.
“It’s cost effective and can be developed at a fraction of the cost of western counterparts, allowing smaller companies to leverage advanced AI without significant financial burdens
“Also, its higher performance modelling at reduced cost enables greater results faster.
“It will have a market dominance and means less reliance on the bigger and more established players like Google and OpenAI, and will create greater competition.
“DeepSeek will also enable all sectors to take advantage of Open Source Advanced AI which, if managed well, will be game changing within sectors that may have missed their opportunity to flex within an ever changing marketplace.
“Using Deepseek at work can be quiet beneficial when you need advanced AI capabilities, such as data analysis, coding assistance, general information retrieval. For example,
Customer Support – Put Deepseek within a chatbot to provide instant support to customers – deploying at a much more competitive cost than counter products
Research – Retrieve and manage information from multiple sources
Coding – Allow Deepseek to look for bugs, generate code snippets and help manage that process
Data – Use Deepseek to process and manage large data sets”
Megha Kumar, Chief Product Officer and Head of Global Geopolitical Risk, CyXcel
“DeepSeek released its flagship DeepSeek-R1 reasoning AI model on January 20, and has since become of the most downloaded AI app on Western and Chinese app stores, outpacing OpenAI’s ChatGPT.”
“It is an open-source model, which means that anyone can use, customise or improve the AI model since key information such as training data, documentation on data collection and processing, the model’s source code and parameters are publicly available.
“Depending on how much of this information is made public, a model can be more or less open. In principle, open-source AI models can be safer and more transparent, since anyone can use and audit the model, and can enable smaller enterprises and individual developers to develop customisable solutions more cheaply. Closed source models are the opposite: all the key information is held as proprietary data, for example like OpenAI’s GPT series; companies do that for the business advantage.
“DeepSeek was developed as open source out of compulsion due to the tightening of US restrictions on the export of advanced semiconductors to China. So reportedly the model uses fewer Nvidia chips and less computational power than models of equivalent power such as OpenAI GPT series.”
“It does underline China’s commitment to AI development and the quality of the country’s resident AI talent. The announcement of DeepSeek was probably deliberately timed to coincide with Trump’s inauguration and the announcement of the US Stargate AI plan.
“However, DeepSeek-R1 reasoning AI model’s sophistication is probably an outlier, and all Chinese AI models, include DeepSeek, appear to comply with Chinese censorship laws. Nonetheless, the success will worry both Washington and US AI firms.
“Their response will be greater innovation and tighter restrictions on China’s access to Western AI technology, particularly semiconductors. However, these restrictions are more likely to slow or temporarily impede China’s AI innovation, rather than to halt them.”
“The DeepSeek AI lab is backed by a Chinese financial trading firm and is bound to have received some state support from Beijing. The lab’s breakthrough will boost venture investor confidence in the AI start-up ecosystem worldwide and their ability to innovate and compete with large, highly-resourced big tech players.
“In the past, large tech firms have tended to acquire top start-ups (for example, Facebook bought Instagram and Google bought YouTube and DeepMind). Now in the coming years such acquisitions will take place in the AI ecosystem too, but not as easily or frequently since competition watchdogs are more determined to prevent monopolies.
“Partnerships between big tech and AI start-ups are more likely (for example, the one between Microsoft and OpenAI). A more favourable regulatory environment would help promising startups to go it alone and scale up.”
Chris Beer, Senior Data Journalist, GWI
“The emergence of DeepSeek as a challenger to ChatGPT is important in many ways – but one overlooked element is that it points to “reasoning” models as a new paradigm in AI. Up to now, we’ve mostly seen AI models that give answers quickly, based on pattern recognition and statistical correlations, but reasoning models think through a prompt more logically before arriving at an answer.
“This represents a significant shift in AI capabilities and performance – but few commentators are reflecting on what this means for the user experience. While quality (49%) and accuracy (46%) of responses are the most important factors in adoption of AI, these are in a trade-off with the speed of responses, which 25% of AI users say is important.
“We also know from our research that speed – along with cost – is one of the biggest drivers of tech adoption in general. Reasoning models might be more thoughtful but can have slower response times, which might impact user perception. An AI chatbot with reasoning capabilities might produce the best answers, but if they take longer, even by a few seconds – or feel like they take longer – some users might stick to the models that produce faster results.”
“Until now, much of the conversation around the best AI models has focused on their ability to hit certain technical benchmarks. But consumer behaviour is much more nuanced. For the makers of AI, they’ll likely need to weigh up whether intelligence, speed or personality is their best play.”