THE WINDS OF CHANGE
Two weeks ago, if you asked somebody who the global leader in AI development and deployment was, what answers would you have expected? Die-hard British patriots would have undoubtedly claimed the crown for the UK, but sadly this is not the case. While Britain has produced notable AI technologies like DeepMind, focused on healthcare and scientific research, and Darktrace, cybersecurity software using AI for threat detection, sadly Blighty doesn’t lead the AI race. A fortnight ago, the United States would have been considered top dog, being home to artificial intelligence giants like Microsoft, Google, OpenAI and Nvidia, with a combined AI market capitalisation of $184 billion in 2024. Then along came China’s DeepSeek, and everything changed overnight. Collectively, Wall Street, the White House and Silicon Valley were in a state of shock, and U.S. Tech stocks lost $1 trillion overnight. Newly-elected American President Donald Trump said:
“This is a wake-up for our industries that we need to be laser-focused on competing against China.”
WHAT IS DEEPSEEK?
DeepSeek, an AI development firm based in Hangzhou, China, was founded in May 2023 by Liang Wenfeng, a Zhejiang University graduate who also co-founded High-Flyer, a China-based quantitative hedge fund that owns DeepSeek. Operating as an independent AI research lab under High-Flyer, DeepSeek focuses on developing open-source LLMs (Large Language Models), releasing its first model in November 2023 and continuously iterating on its core technology. The company gained global recognition in January 2025 with the launch of its R1 reasoning model. While its funding and valuation remain undisclosed, DeepSeek offers various services for its models, including a web interface, mobile application, and API access.
WHY IS DEEPSEEK SIGNIFICANT?
DeepSeek’s rise to the top of the Apple App Store charts and surpassing downloads of ChatGPT, has cemented its place in the public consciousness, challenging the notion that the U.S. would remain the undisputed global leader in AI. While U.S. dominance has largely been driven by massive capital investment, China’s DeepSeek achieved its success at a fraction of the cost of its American competitors. Donald Trump said:
“I view that as a positive, as an asset… you won’t be spending as much, and you’ll get the same result, hopefully.”
DeepSeek stands out for its strong commitment to open-source development, but its most significant technical breakthrough lies in its ability to distil advanced reasoning capabilities from massive models into smaller, more efficient counterparts. One of its models frequently outperforms larger open-source alternatives, setting a new benchmark for compact AI performance. Unlike competitors such as OpenAI, which rely on supervised fine-tuning in the early stages, DeepSeek heavily utilizes reinforcement learning to develop reasoning skills, marking a deliberate shift away from the hybrid training strategies favoured by AI leaders in the United States. Data scientist Drew Breunig commented on the success of DeepSeek:
“If there’s a lesson from DeepSeek’s triumph, it’s this: be wary when the route to progress is simply spending more money. This path fosters no innovation and your poorer competitors will be forced to get creative, work within their constraints, and eventually…they’ll win. Spending is not innovating.”
WHERE DOES DEEPSEEK FALL SHORT?
Benchmark results show that DeepSeek’s models excel in reasoning-intensive tasks, consistently delivering top-tier performance in areas such as mathematics and coding. However, the research also points out some weaknesses, particularly in non-reasoning tasks and factual query accuracy, where DeepSeek lags behind OpenAI’s most advanced models. While no one has independently verified whether DeepSeek is using large compute resources to achieve these benchmark results or has copied OpenAI’s work, U.S. controls on cutting-edge microchips would restrict the resources available to China.
THE NVIDIA STORY
Alex Wang, CEO of Scale AI, which also evaluates AI models, described DeepSeek as being on par with OpenAI in a CNBC interview. He also mentioned that China has acquired approximately 50,000 of Nvidia’s H100 chips despite export controls. A Nvidia spokesperson did not directly address the claim but said that DeepSeek represents a significant AI advancement and serves as an example of Test Time Scaling, a technique that enhances computing power when a model processes data to generate new results. This additional compute power allows the model to explore various options and refine its answers, achieving better results with less training and compute. It enables the model to focus its computational resources more efficiently, much like how exercise initially depletes energy but ultimately builds the body’s capacity to store and use energy more effectively. The spokesperson added that DeepSeek’s approach demonstrates how new models can be created using this technique, leveraging widely available models and compute resources that comply with export controls. He also emphasised that inference requires significant numbers of NVIDIA GPUs and high-performance networking and noted the existence of three scaling laws: pre-training, post-training, and emerging test-time scaling.
YOUR PERSONAL AI
Perhaps the most significant takeaway from DeepSeek’s announcement isn’t its impact on the U.S.-China competition, but its implications for individuals, public institutions, and anyone concerned about the growing dominance of a small group of tech giants. It’s a positive development for those who want to build their own generative AI tools using data they control, rather than depend on tools from large companies that may not always have their best interests in mind. Berkeley AI PhD student Ritwik Gupta said:
”The internet has historically thrived as a decentralized set of services,” Gupta said. If the goal is to get everyone to have their own ‘personal AI’, then it will be necessary for small models to run on people’s personal devices. I expect companies like Apple, who have a privacy-first model, to continue to push for offline, disconnected algorithms.”
DEEPSEEK – FACT AND FICTION
Whatever people are saying about DeepSeek, mainstream media or otherwise, it’s important to distinguish between reality and fantasy. Acclaimed City of London Councillor and London innovator, Madush Gupta, gives us his lowdown on DeepSeek fact versus fiction:
FACT
- Breakthrough in AI Reasoning – DeepSeek R1 demonstrates advanced problem-solving by “thinking out loud,” similar to OpenAI’s o1.
- Cost-Efficient AI Training – By using reinforcement learning and smaller, more efficient models, DeepSeek has built an AI that competes with industry leaders at a fraction of the cost.
- A Political & Economic Shockwave – DeepSeek’s rise fuels geopolitical debates on AI dominance, U.S.-China tech competition, and the role of export controls on AI chips.
- Open Source, But with Limits – Unlike OpenAI, DeepSeek has made parts of its model open source, but its training data remains undisclosed.
- Investor Frenzy – The company’s success has rocked stock markets, triggering sharp volatility in Nvidia shares and raising questions about AI’s financial future.
FICTION
- DeepSeek is an AI “miracle” – While impressive, its techniques are not secret. U.S. labs like OpenAI, Anthropic, and Google likely have similar capabilities.
- AI regulation in the West is the problem – Some claim DeepSeek’s success proves U.S. regulations stifle innovation, but experts argue that deep capital investment still gives Western AI firms an edge.
- China has leapfrogged OpenAI – DeepSeek’s tech is competitive, but it still relies on Nvidia chips and faces deployment challenges at scale.
- AI will now be cheap for everyone – Lower training costs don’t mean lower operating costs. Running advanced models at scale still requires massive compute power.
“DeepSeek’s rise is a wake-up call for AI strategy, innovation, and investment. The real question is: Will AI be shaped by the biggest spenders like the USA and China or by the smartest innovators such as the UK? My sense is the UK is back in the game. I am filled with optimism”
THE UK ON DEEPSEEK
In the UK, experts have advised caution regarding the swift adoption of DeepSeek, raising concerns about the potential for misinformation and the Chinese government’s possible exploitation of user data. While the government has stated that using the platform is a personal decision for citizens, officials are closely monitoring any national security risks associated with the AI. They emphasized that they would take swift action if any threats to data security arise. Peter Kyle, the UK Technology Secretary, said in a podcast:
“I think people need to make their own choices about this right now because we haven’t had time to fully understand it … this is a Chinese model that … has censorship built into it. So, it doesn’t have the kind of freedoms you would expect from other models at the moment. But of course, people are going to be curious about this.”
Much stronger words of caution came from Ross Burley, a co-founder of the Centre for Information Resilience. He said:
“We’ve seen time and again how Beijing weaponizes its tech dominance for surveillance, control and coercion, both domestically and abroad.”
IS DEEPSEEK A SECURITY AND PRIVACY THREAT?
Amid the growing scrutiny of DeepSeek’s rapid rise, experts from Enkrypt AI, a Boston-based AI security and compliance platform, have shared their initial findings. Their red team uncovered numerous critical security flaws in the model. Enkrypt described the platform as highly biased and prone to generating not only insecure code but also harmful content, including criminal material, hate speech, threats, self-harm content, and explicit material. Further research has revealed that the model is vulnerable to manipulation, or “jailbreaking,” which could allow it to aid in the development of chemical, biological, and cyber weapons. Enkrypt warned that this posed significant global security risks. On top of this, an article from Forbes emphasises how DeepSeek is taking AI privacy risks to a whole new level. The AI tells users of the app:
“The personal information we collect from you may be stored on a server located outside of the country where you live. We store the information we collect in secure servers located in the People’s Republic of China. We automatically collect certain information from you,” DeepSeek says, “including internet or other network activity information such as your IP address, unique device identifiers, and cookies.”
The platform also gathers specific device and network connection details, such as the device model, operating system, keystroke patterns or rhythms, IP address, and system language. As London’s #1 cybersecurity company for SMEs, we at Zhero would recommend using DeepSeek with caution. As Michael Wooldridge, an AI Professor at Oxford University said:
“I think it’s fine to download it and ask it about the performance of Liverpool football club or chat about the history of the Roman empire, but would I recommend putting anything sensitive or personal or private on them? “Absolutely not … Because you don’t know where the data goes.”
And protecting your data is our business. Reach out today and find out how we do it.