Listen here on your chosen podcast platform.
Chinese AI startup DeepSeek is limiting new registrations for its AI service, citing “large-scale malicious attacks” on its platform.
The issues, which began at around 1.30pm UK time, are slowing down the website and playing havoc with the company’s API (the tech that lets other apps talk to DeepSeek’s AI).
Earlier on Monday, DeepSeek said it was restricting sign-ups to those with Chinese mobile phone numbers. The company’s website and app appears to be working for those who previously created accounts, though users have noticed that the AI is taking longer to process queries.
News of the disruption comes after DeepSeek overtook ChatGPT to become the top app on Apple’s charts.
As more Western users have flocked to DeepSeek, concerns about Chinese censorship have also surfaced. For instance, the bot refuses to respond or abruptly ends conversations about topics like the Tiananmen Square Massacre.
Here’s what you need to know about the AI that has Silicon Valley in panic mode.
Stock market shock
The emergence of DeepSeek, a Chinese AI that can allegedly go toe-to-toe with US giant ChatGPT, has rattled global markets.
On Monday, US stock indices took a nosedive as jittery investors dumped tech stocks, spooked by fears that AI development costs had spiralled out of control. The sell-off sparked a trillion-dollar wipeout, according to Bloomberg, hitting US and European chipmakers, AI companies, and energy firms the hardest.
AI powerhouse Nvidia saw around $400bn (£321bn) wiped from its market value, while tech titans Meta, Google, and Microsoft also took significant hits.
Obviously
— Elon Musk (@elonmusk) January 27, 2025
The downturn was triggered by the release of DeepSeek’s latest AI model, which it claims operates at a fraction of the cost of OpenAI’s ChatGPT, the current poster child for modern AI with more than 300 million active users.
The buzz around the Chinese bot has hit a fever pitch, with tech heavyweights weighing in. On Monday, Elon Musk poured cold water on DeepSeek’s claims of building its advanced models using far fewer, less powerful AI chips than its US competitors.
What is DeepSeek?
DeepSeek’s AI looks and functions much like ChatGPT and other large-language models. It can answer questions, generate poetry and prose, and write complex code (the programming language used to build everything from apps to websites).
🚀 DeepSeek-R1 is here!
— DeepSeek (@deepseek_ai) January 20, 2025
⚡ Performance on par with OpenAI-o1
📖 Fully open-source model & technical report
🏆 MIT licensed: Distill & commercialize freely!
🌐 Website & API are live now! Try DeepThink at https://t.co/v1TFy7LHNy today!
🐋 1/n pic.twitter.com/7BlpWAPu6y
What sets it apart, however, is that DeepSeek claims to offer all of these functions (along with a few others) at a level comparable to its larger rivals, and all at a significantly lower cost.
In a research paper, DeepSeek’s engineering team revealed they only needed about $6m (£4.8m) in raw computing power to develop their new system – roughly 10 times less than Meta's investment in its latest AI technology, according to The New York Times.
Notably, DeepSeek achieved all this under the constraints of strict US export controls on advanced computing tech in China. As restrictions from the Biden administration started to bite, the Chinese firm was forced to get resourceful, building its models with fewer and far less powerful Nvidia AI chips.
What impact has DeepSeek had?
While the Chinese-US tech race is marked by increasing protectionism, DeepSeek has taken a different approach. Following in the footsteps of companies like Meta, it has decided to open-source its latest AI system.
By sharing the underlying code with the wider tech community, the company is allowing other businesses, developers, and researchers to access and build upon it. It means that anyone with the right expertise can now use DeepSeek’s models to create their own products or conduct research.
The company’s breakthroughs have sent shockwaves through the tech industry. Meta, NVIDIA, and Google’s stock prices have all taken a beating as investors question their mammoth investments in AI in the wake of DeepSeek’s models. The fear is that DeepSeek will turn out to be the new TikTok, a Chinese giant that encroaches on the market share of US tech giants.
Industry titans have described it as a wake-up call for the West. Venture capitalist Marc Andreessen hailed the success of DeepSeek as a "Sputnik moment”, suggesting it will inject a new level of competition and innovation into the AI landscape.
Deepseek R1 is AI's Sputnik moment.
— Marc Andreessen 🇺🇸 (@pmarca) January 26, 2025
Alexandr Wang, CEO of Scale AI, who became the world’s youngest self-made billionaire in 2022, warned that the gap between US and Chinese AI is narrowing. Speaking to CNBC, the entrepreneur called DeepSeek’s latest AI model an “earth-shattering” release.
However, Mr Wang expressed doubts about DeepSeek’s claims of using fewer resources to build its models, speculating the company may have access to a large number of chips.
Who owns DeepSeek?
DeepSeek is an AI start-up that was spun off from a Chinese hedge fund called High Flyer-Quant by its manager, Liang Wenfeng, according to local media.
Mr Liang has credited the company’s success to its fresh-faced team of engineers and researchers.
Rather than focusing on years of experience, the company prioritises raw talent, with many of its developers being recent graduates or newcomers to the AI field. This approach, according to its founder, has been key to the company’s growth and innovation.
How do you use DeepSeek?
DeepSeek’s AI can be used for free on both the web and through a mobile app. Its interface is virtually identical to ChatGPT’s: Users can interact with the bot by typing text or uploading files or images.
There’s also the option to enable DeepSeek’s R1 reasoning system. Unlike AI that identifies patterns in data to generate content, like images or text, reasoning systems focus on complex decision-making and logic-based tasks. They excel at problem-solving, answering open-ended questions, and handling situations that require a step-by-step chain of thought, making them better suited for trickier tasks like solving maths problems.
DeepSeek claims R1 achieves similar or slightly lower performance as OpenAI’s o1 reasoning model on various tests.
In addition, users can ask the AI to search the web as part of its responses, which is useful for finding recent events or verifying information.
Although it is a multimodal AI that can recognise various types of media, DeepSeek’s consumer-facing bot does not currently generate images or process audio. In comparison, ChatGPT offers speech recognition to all users, and an advanced voice mode for more natural conversations is available to paying members. OpenAI has also introduced a video-generating AI called Sora in the US for subscribers.
