(MENAFN- PRovoke)
If you woke up this morning and checked the stock markets, you would have seen that it has been thrown into utter chaos with US Stocks plummeting as traders left the tech sector and reportedly erased over US$1 trillion in market cap.
The culprit? A new AI model by Chinese startup DeepSeek which claims to rival OpenAI and has become the number one downloaded free app on Apple's Iphone store.
What Is DeepSeek?
DeepSeek is an AI startup that was founded in 2023 in Hangzhou, China. It released its first AI large language model late in 2023. About a month ago, DeepSeek began getting more significant attention after it released a new AI model, DeepSeek-V3, that it claimed was on par with OpenAI and that was more cost-effective in its use of Nvidia chips to train the systems.
However, DeepSeek only started getting significant traction last week when it released a follow-up research paper that detailed another DeepSeek AI model called R1 which showed more advanced reasoning skills and that was significantly cheaper to o1 which is a model sold by OpenAI.
“The key feature that shocked experienced AI users, I think, is the DeepThink R1 model, which shows the entire thinking process to users. At the moment, no alternatives are showing the thinking process in such detail, not even OpenAI's o1 Pro that I've been using for over a month,” said Tianyu Xu, an AI educator.
He added that in the thinking process, the model has demonstrated a good level of emotional intelligence.“It feels like you are reading the thoughts of another human instead of robotic voices or procedures,” he said, noting that one can also use R1 with web search, browsing up to 50 websites, and composing a solid answer.“This is already on par with or ahead of Google's Gemini Deep Research.”
Xu added that finally, another feature of the model is simply speed.“While it takes time to think, it generates output very quickly. For Chinese users, in particular, the speed is on par with generating English text. However, on ChatGPT, processing Chinese queries and generating Chinese responses usually takes much longer.”
Adding to his point, Milind, an AI scientist, said that the significance of R1 lies in its ability to handle complex planning tasks, a capability that has traditionally been challenging for standard Large Language Models (LLMs).
“What makes this particularly revolutionary is that DeepSeek has made this technology openly accessible through their open-weights approach and MIT license, offering multiple model sizes to suit different needs. This democratisation of advanced AI reasoning capabilities represents a significant shift in the accessibility of cutting-edge AI technology,” he said.
Why Is It Causing A Buzz Now?
Now we have certainly seen a multitude of AI platforms pop up over the last few years, many with very similar capabilities. However, what is turning heads in the industry is that the developers of DeepSeek are claiming that they have built it on 1/30th the cost of similar API models at US$6 million.
“This amount dwarfs in comparison to what the mega tech has been spending. It could be a game-changer, suggesting you don't need a Silicon Valley-sized fortune and massive computing power to create a powerful AI engine,” said Nishant Kaushal, founder and CEO of ADNA Research.
He added that it is a“classic David and Goliath story” that has caught everyone's attention - a small Chinese startup has dared to leapfrog Silicon Valley giants and redefine innovation rules and it seems to have the potential to win, he said.
“We've had a flurry of AI releases, but DeepSeek is resonating because it has awakened us to the potential for disruption and democratisation,” explained Kaushal.“Plus, the timing is right. We're at a point where everyone is realising how transformative AI could be and DeepSeek feels like it could open the doors for more players to get involved.”
He added that it also enables smaller companies to build innovative AI products, that was until a month ago, largely within the domain of the Silicon Valley giants.
“We're already hearing about SMEs flocking to DeepSeek, retraining it as their foundational model,” he said, noting that for companies, this could mean new opportunities for innovation. DeepSeek's cost-effectiveness lowers the barrier to entry for using advanced AI in all sorts of applications, from customer service to data analysis.
“Investors should be looking at the companies building on top of DeepSeek. It feels like the early days of the internet – the real value wasn't just in the technology itself, but in the innovative services and businesses that it enabled,” he said.
Meanwhile, Matt Collette, CEO of Sequencr AI, an AI consultancy highlighted the billions poured into Generative AI over the last two years. Goldman Sachs' 2024 report questioned whether the projected US$1 trillion in capital and operating expenditures by tech companies such as Meta, Google, and others could ever be recouped, he posited. DeepSeek's success lends weight to those concerns, as it demonstrates that comparable performance can be achieved at a fraction of the cost.
“DeepSeek's innovation directly challenges the pricing structures of OpenAI, Microsoft, Google, and others. If you can train a model for US$5 million, there's no need to charge US$20 to US$30 per user per month. DeepSeek's significantly lower API costs are likely to put downward pressure on industry pricing, which is a win for businesses looking to adopt Gen AI,” he said.
Why Could DeepSeek Become A Political Issue?
Beyond pricing though, DeepSeek could potentially create political issues particular as the United States becomes more cautious about sharing information with China.
Collette highlights this by saying that before switching from ChatGPT to DeepSeek, businesses must consider whether they're comfortable with a Chinese company“potentially using their data to train models” as well as the outputs they will get from said models.
He added that the data privacy, bias, and ethics concerns surrounding DeepSeek echo the issues raised with ByteDance and TikTok but with even higher stakes.
“China-based AI model outputs may be skewed, especially on topics sensitive to China. Furthermore, DeepSeek's open-source models lack the guardrails employed by competitors such as OpenAI and Anthropic, raising concerns about misuse for disinformation or malicious purposes,” he said.
Colette added that US and Western countries may respond with tighter restrictions on Chinese AI companies and related technologies.“The AI race has entered a new chapter, where cost-efficient, highly performant models from China are poised to reshape the competitive landscape,” he said.
He added that part of the controversy, as it related to DeepSeek, is that it appears their V3 model was trained using ChatGPT. “In testing, it has a tendency to itself as ChatGPT, which has led many to believe that the model was trained on datasets containing outputs from OpenAI. This has raised concerns about data quality and how much DeepSeek relied on existing models to develop their own,” he explained.
Rob Van Alphe, AI advisor at Folgate Advisors, added that users will need to properly read terms and conditions about data capturing and ownership, as cyber-attacks could happen before using DeepSeek and watch how it develops pointed out that since it's popularity skyrocketed, deepSeek has already since suffered a cyberattack, causing it to temporarily limit registrations.
MENAFN28012025000219011063ID1109136478