DeepSeek - a Chinese AI platform - was unveiled to the world on Monday, 27 January, sparking chaos in the AI sector.
US tech firms, which have spent billions on creating the next-gen AI platform such as OpenAI, were sent into a panic after the emergence of the platform which cost a mere $6m of its US counterpart.
Nvidia - a global leader in the manufacturing of semiconductors - saw almost $600bn wiped off it’s market value after the announcement of DeepSeek, which does not need chips to operate.
Similarly, the likes of Meta also lost market value.
OpenAI chief Sam Altman described DeepSeek as “impressive” but vowed that his platform will “obviously deliver much better models”.
He posted on X: "DeepSeek's r1 is an impressive model, particularly around what they're able to deliver for the price.
“We will obviously deliver much better models and also it’s legit invigorating to have a new competitor. We will pull up some releases.
“But mostly we are excited to continue to execute on our research roadmap and believe more compute is more important now than ever before to succeed at our mission.”
Such has been the impact of DeepSeek in the past 24 hours that the newly inaugurated president Donald Trump has waded in, stating that it should be a “wake up call” to the US tech industry.
Follow for live updates
China’s Alibaba releases new AI model that ‘outperforms’ DeepSeek V3
16:51 , Saqib ShahThe Chinese AI models are arriving thick and fast, with Alibaba releasing a new AI that it claims outperforms DeepSeek’s game-changing V3 chatbot on several key benchmarks.
“Qwen 2.5-Max,” as the new software is known, is a large language model (LLM) that was trained and refined in much the same way as ChatGPT, using reinforcement learning and human feedback.
However, its underlying machine learning techniques mimic DeepSeek’s approach. In a nutshell, Qwen is a Mixture of Experts (MoE) model, where the AI is split into different parts, each focusing on a specific aspect of the task.
The burst of DeepSeek V3 has attracted attention from the whole AI community to large-scale MoE models. Concurrently, we have been building Qwen2.5-Max, a large MoE LLM pretrained on massive data and post-trained with curated SFT and RLHF recipes. It achieves competitive… pic.twitter.com/oHVl16vfje
— Qwen (@Alibaba_Qwen) January 28, 2025
This setup is designed to reduce computing costs and work faster, especially when dealing with large-scale models with billions of parameters.
By comparison, regular LLMs rely on all their components for every task, often leading to higher processing demands.
Qwen’s API is available through Alibaba Cloud, which is like a giant virtual computer that businesses can rent and use over the internet. The overarching service is provided by the Alibaba Group, a Chinese online shopping giant (think Amazon, eBay and PayPal all rolled into one).
Short sellers banked $6.6 billion from Nvidia's DeepSeek-induced crash
16:15 , Saqib ShahNvidia’s stock took a dramatic nosedive yesterday, losing nearly 17 per cent of its value and erasing a colossal $600 billion (£482bn) in market cap.
But while the chip giant’s investors were reeling, short sellers were celebrating a massive windfall, pocketing a jaw-dropping $6.7 billion (£5.38bn) in profits.
That’s according to new data from financial analytics platform Ortex.
For the unfamiliar, short selling is a risky bet where investors borrow stocks and sell them in the hope of buying them back later at a lower price. If the stock falls, they profit — which is exactly what happened with Nvidia.
Nvidia came undone on Monday after the sudden growth of DeepSeek, a Chinese AI that has upended the status quo in the tech industry.
Despite allegedly stockpiling Nvidia’s best chips, DeepSeek said it built its most advanced AI using lower-performing Nvidia processors amid US-imposed export controls on sensitive tech.
Its breakthrough suggested that the most expensive hardware might not be needed for cutting edge AI development, derailing the shares of a bunch of minted AI companies in the process.
US tech stocks bounce back after early slump
15:42 , Saqib ShahNvidia is riding a rollercoaster in today’s trading, with its stock flipping from green to red and back again.
The AI chipmaker, which dropped a whopping 17 per cent on Monday, is currently up 2.4 per cent.
OpenAI investor Microsoft is also up 1.56 per cent ahead of its earnings call on Wednesday, during which CEO Satya Nadella will surely field questions about DeepSeek’s impact on the wider AI sector.
Apple, meanwhile, has sailed through the market storm unscathed, with its stock climbing for two consecutive days. While its performance may stem from being left out of the AI frenzy — especially with the underwhelming debut of Apple Intelligence — it also took a hit earlier this month from sinking iPhone sales in China.
OpenAI releases new AI chatbot for US government
15:31 , Saqib ShahUS government agencies can now run ChatGPT on their own secure servers, OpenAI announced on Tuesday as it revealed its new “ChatGPT Gov” chatbot.
The custom-made bot features all the latest AI perks, while giving the US government more control over the bot’s security and privacy, especially as it relates to sensitive data.
ChatGPT Gov runs on the top-of-the-line GPT-4 model, can save chats, analyse uploaded files, and comes with the ability to create custom AI tools.
OpenAI’s ChatGPT Enterprise, which powers the upcoming ChatGPT Gov, is still waiting for approval to handle sensitive government data. OpenAI says it’s a slow process, but it’s optimistic that efforts to modernise government tech under President Trump could help speed things up, reports CNBC.
Despite the delays, OpenAI is confident that ChatGPT Gov will be available for testing soon — possibly within a month. The company expects the tool to make a big impact for agencies handling sensitive data, like those in defense, law enforcement, and healthcare.
There are concerns, however, about the government’s use of AI, especially in law enforcement. Critics worry that AI tools could lead to privacy violations or biased decision-making, particularly in sensitive areas like criminal justice.
The launch of ChatGPT Gov comes in the wake of DeepSeek's cost-efficient AI, which has rattled global markets and sent tech stocks plummeting. Its disruptive potential has investors spooked, with concerns over how cheaper, more accessible AI could impact major players in the tech sector, especially OpenAI.
Nvidia and other AI-related stocks drop for a second day
15:05 , Saqib ShahTech stocks that have benefited handsomely from the AI gold rush continue to be battered by the emergence of DeepSeek.
The likes of Nvidia, Broadcom and Super Micro were down by 1.5 per cent, 1 per cent and 5.3 per cent, respectively, in early trading on Tuesday.
After a promising start, the trio quickly shed their initial gains to drop into the red for a second day in a row.
The release of DeepSeek’s latest AI model, which it claims can go toe-to-toe with OpenAI’s best AI at a fraction of the price, sent global markets into a tailspin on Monday.
The news came on the back of a massive $500 billion investment into AI infrastructure in the US by OpenAI, SoftBank and Oracle - as announced by President Trump.
Along with chipmakers, AI-adjacent sectors including energy also took a hammering in the US and Europe on Monday.
US stock markets open tentatively following AI selloff
14:48 , Saqib ShahStocks appeared unchanged on Tuesday after a major rout sparked by the release of DeepSeek’s “cheap” AI model and its impact on tech investments in the US.
Nvidia was initially up 3 per cent, but it quickly dropped those gains, winding up in the red.
Investors appeared cautious after the AI bellwether plummeted nearly 17 percent on Monday, shedding roughly $600 billion in market value - the biggest ever one-day drop for a US firm.
To put that into perspective, Nvidia lost the equivalent of Sweden’s entire GDP in a single day.
The tech-heavy NASDAQ was also flat to start the day.
DeepSeek vs OpenAI - what are the differences?
14:30 , Saqib ShahAs a major new entrant in the AI sector, DeepSeek is inevitably being compared to the industry’s big gun, OpenAI.
Indeed, the two companies are being judged on their duelling AI models, which appear to have matching capabilities. But, there are some major differences between the firms.
In terms of their origins, DeepSeek was reportedly spun out from a Chinese hedge fund, while OpenAI is currently transitioning away from its nonprofit roots and moving towards becoming a fully for-profit entity.
DeepSeek and OpenAI have also diverged significantly in their approaches to open-sourcing their AI models. DeepSeek has fully embraced open source with its DeepSeek-R1 model, granting developers free access to modify and build upon it. Users can even download and run the model locally, fine-tuning it to address any privacy concerns.
OpenAI, on the other hand, has moved away from its initial open research ethos. Models like GPT-4 are not open source, with the company granting use through commercial partnerships and paid APIs. This proprietary approach allows OpenAI to maintain tighter control over its models.
An API (Application Programming Interface) is a set of rules and specifications that enables other software programs to interact with DeepSeek's or OpenAI's AI models and utilise their capabilities. Think of it like a white-label service for AI, allowing other companies to integrate the model's functionality into their own products, such as apps.
Finally, the two companies differ in their revenue models. DeepSeek primarily generates income through its API offerings, which are notably cheaper than those of competitors, while giving away its AI for free. OpenAI, on the other hand, offers both an API to businesses, along with subscription plans that grant users access to its most advanced AI models, along with other perks.
DeepSeek could be eyeing Huawei as an alternative to Nvidia
14:09 , Saqib ShahWhile DeepSeek has previously utilised Nvidia products, it is also focusing on adapting existing models for Huawei's hardware, particularly for "inference" tasks – the computation required for AI applications like chatbots.
Huawei is actively courting Chinese tech giants to adopt its Ascend chips, backed by strong government support, according to recent reports.
Chinese officials are encouraging local companies to move away from Nvidia and embrace domestic alternatives.
Huawei's strategy focuses on inference, when a trained AI model uses its smarts to respond to queries, betting that this area will see increased demand as AI applications proliferate.
Despite the push, Huawei's Ascend chips reportedly face performance challenges and slower connectivity compared to Nvidia's offerings.
Is DeepSeek safe? Cyber-security experts raise alarm amid data-sharing fears
13:57 , Saqib ShahWith DeepSeek currently the most downloaded app on iPhone, attention has turned to the company’s data-sharing practices.
According to an investigation by Wired, DeepSeek’s privacy policy explicitly states that it sends data to China.
That’s hardly surprising for a Chinese company, but it also has three sweeping categories of data collection that are more concerning.
First, there’s the information you directly share, such as text inputs, audio files, prompts, uploaded content, and feedback. Second, there’s data collected automatically — likely including device information and location data. Lastly, DeepSeek reserves the right to obtain details about you from third-party sources.
While DeepSeek users can delete their chat history, it’s unclear if this action fully erases the data from the company’s servers.
There was another worrying twist in the DeepSeek saga on Monday. Following its viral rise on the app store charts, the company began limiting new users from signing up for its AI service, citing a large-scale cyberattack.
While the attack's cause remains unclear, experts like Toby Lewis of Darktrace suggest it could be due to overwhelming user demand following the platform's viral success, or, more worryingly, the exploitation of vulnerabilities in its systems.
Meanwhile, Dan Schiappa of Arctic Wolf highlighted additional anxieties, including data privacy risks associated with Chinese foundational models and the potential for intellectual property theft.
“Considering DeepSeek is already limiting its registrations due to a cyber attack, you have to wonder whether they have the appropriate security and policies in place to maintain your privacy,” said Schiappa.
DeepSeek releases new image-generating models that can rival OpenAI's Dall-E
13:00 , Saqib ShahCapitalising on the breakout success of its AI service, Chinese tech giant DeepSeek has released a new lineup of AI models that can analyse and generate images — and it’s making bold claims about their capabilities.
The biggest buzz is around Janus Pro 7B, the heavyweight of the new models, which DeepSeek says beats OpenAI’s DALL-E 3 and Stability AI’s Stable Diffusion XL on key performance tests. Those are big names in the AI world, known for producing lifelike images from text prompts.
AI models like Janus Pro 7B are measured in “parameters,” which indicate their problem-solving prowess — the more parameters, the better the performance. With 7 billion parameters, Janus Pro 7B is designed to create visuals, answer image-based questions, and craft visual stories.
Furthermore, DeepSeek is making these models freely available for download on AI development platform Hugging Face under an MIT license. This means developers and businesses can use them for commercial projects without jumping through legal hoops or paying hefty fees. Essentially, DeepSeek is throwing open the doors to its tech.
The models also hint at future updates for DeepSeek's AI platform, which, for now, remains unable to generate images.
DeepSeek's mysterious CEO
12:00 , Andrew WilliamsAn X account purporting to be DeepSeek founder Liang Wenfeng has been posting “funny” memes online, but the platform’s community notes feature suggests it’s a fake account. But who is Liang Wenfeng?
Before becoming a tech leader, Liang Wenfeng founded a hedge fund called High-Flyer, in 2015.
According to the FT, he branched out in 2021, buying thousands of Nvidia graphics cards to kick-start an AI project. He did so just before US restrictions limited the export availability of certain high-tech products in China.
He is reportedly 40 years old at present.
Behind the numbers
11:39 , Andrew WilliamsOne of the key figures behind the current DeepSeek palaver is the claim its latest AI model cost $6 million to develop. And given some of the figures you see thrown about in the tech world, that's not a huge amount of money at all.
It is derived from a paper published by the company that explains how the DeepSeek-V3 model was made.
The $6 million figure relates to the number of compute hours of AI training DeepSeek-V3 involved, based on rental access to an Nvidia H800 GPU costing $2 an hour.
There were 2,664,000 computing hours of “pre-training”, 119,000 hours of “Context Extension” and five thousand hours of “Post-Training.”
Multiply those by the $2-an-hour figure and you end up with $5.576M. In other words it’s a partial view of the company’s real costs. And those calculations will no doubt be subject to further scrutiny in the coming days and weeks.
Nvidia speaks out
11:21 , Andrew WilliamsNvidia is one of the companies most directly and clearly affected by the hype surrounding DeepSeek, with hundreds of billions of dollars sheared off its market value in the last few days. It too has positive words to say about DeepSeek’s AI models, though.
“DeepSeek is an excellent AI advancement and a perfect example of Test Time Scaling,” an Nvidia spokesperson writes in a statement released to several outlets including Bloomberg.
“DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely-available models and compute that is fully export control compliant.”
Those export controls to which the company refers are the US government’s limits on the cutting-edge computing hardware Chinese firms are able to acquire.
DeepSeek’s AI was trained using Nvidia’s own hardware, including its H800, an Enterprise-grade graphics chipset specialised for use in AI data centres. Nvidia produces tailored versions of its hardware for the Chinese market in order to adhere with US restrictions.
Open for business
11:06 , Andrew WilliamsFollowing a large-scale cyberattack, DeepSeek warns new users they may have trouble registering to use the service. Its chatbot is available through web browsers and an app available for both iOS and Android.
“Due to large-scale malicious attacks on DeepSeek's services, registration may be busy. Please wait and try again. Registered users can log in normally. Thank you for your understanding and support,” the DeepSeek website reads.
However, despite user registrations being limited less than 24 hours ago, new users can currently get on-board based on our own testing.
The DeepSeek chatbot has so far told us it has both a knowledge cutoff of October 2023 and July 2024. As with all LLM chatbots, approach with caution.
Microsoft CEO Satya Nadella's take
10:53 , Andrew WilliamsMicrosoft CEO Satya Nadella commented on DeepSeek’s AI models during the World Economic Forum at Davos.
He said DeepSeek is “super impressive in terms of both how they have really effectively done an open-source model that does this inference-time compute, and is super compute efficient,” as reported by CNBC.
“We should take the developments out of China very, very seriously.”
Microsoft’s share value dropped by almost five per cent over the weekend following the emergence of DeepSeek’s AI models. It has rallied slightly since, but still remains down on the figures from January 24.
Nadella also referred to the Jevons Paradox in a recent X post, suggesting this latest development is part of AI becoming more cost-effective and efficient, paving the way for a further increase in its popularity.
Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can't get enough of. https://t.co/omEcOPhdIz
— Satya Nadella (@satyanadella) January 27, 2025
Circumventing DeepSeek's censorship
10:42 , Andrew WilliamsDeepSeek’s chatbot is unwilling to respond to topics that fall under the Chinese government’s censorship laws but some users are already finding a way around them.
Taiwan, the Hong Kong riots, Tiananmen Square and comparisons between Xi Jinping and Winnie the Pooh are all off limits for Chinese chatbots.
However, as reported by The Guardian, DeepSeek's chatbot can be coaxed into describing the famous Tiananmen Square photos by asking it to replace certain characters with numbers. And referring to “Tank Man” rather than words that might fire up DeepSeek’s censorship programming.
Signs of recovery for Nvidia
10:29After a record-breaking loss for Nvidia, there are early signs of recovery.
Shares in the tech firm were up by 4.8% in pre-market trading.
Nvidia has remained tight-lipped on its historic crash, but markets analyst Jesse Cohen believes the great rush could be over.
Nvidia $NVDA shares are up 5% in overnight trading to almost $125.
— Jesse Cohen (@JesseCohenInv) January 28, 2025
Is the DeepSeek selloff over? pic.twitter.com/8sF5wSsLQE
