DeepSeek is turning the AI world upside down: Power at 10% the cost?
A Chinese startup just made a big splash in the tech world after unveiling a new AI service that supposedly offers the same power and quality as the AI projects of OpenAI, Meta and Google but at a fraction of the cost.
Hangzhou-based DeepSeek has released R1, an open-source AI that boasts similar capabilities to popular chatbots such as ChatGPT, MetaAI and Gemini. However, unlike its more established counterparts, DeepSeek-R1 doesn’t cost as much to own and operate.
What’s the secret? DeepSeek said it was able to build R1’s large language model (LLM) using reinforcement learning (RL) techniques and without any supervised finetuning.
This allowed the company to cut the cost of its AI development to just 10% compared to those of traditional LLMs.
In fact, DeepSeek claimed they finished the R1 for around US$6 million, which is a far cry from the billions of dollars it usually takes to completely build an AI.
Saving on development costs doesn’t mean DeepSeek sacrificed power, either. The company was quick to point out that R1 has a 97% accuracy when executing its tasks.
DeepSeek also measured R1’s performance using several benchmarks including AIME 2024 (mathematical tasks), AlpacaEval 2.0 (question-and-answer), as well as MMLU (general knowledge), and the AI was on par or even better compared to other chatbots.
Also Read: 25 signs your job is at risk of AI takeover
Turning the AI world upside down
If the R1’s reported capabilities are to be believed, then DeepSeek is poised to disrupt the AI landscape that has practically been dominated by larger tech companies like Alphabet, Microsoft, and Nvidia.
However, its biggest strength lies in its accessibility and potential for open innovation.
By building the R1 using an MIT open-source license, DeepSeek allows more people to freely adopt its LLM. The company has also encouraged others to develop their own chatbots using its technology.
This accessibility, combined with its lower entry price compared to other AIs, makes R1 adoption too enticing to turn down for both individuals and organisations.
DeepSeek’s open-source approach also goes against the usual proprietary ecosystems that many AI companies follow for their services. The company prioritises giving researchers and fellow startups easy access to AI tools.
In this sense, R1 shares similarities with other programs like Mistral AI in France, AI Stack in India, and even Meta’s own Llama 2.
Providing real-world use cases for DeepSeek’s AI
The DeepSeek-R1 has already been deployed in several real-world applications.
Finance. Companies are using DeepSeek’s AI to bolster trust and security for digital transactions by providing better fraud detection.
Health care. DeepSeek’s predictive analytics has helped provide earlier and more accurate diagnoses of illnesses, resulting in better outcomes while also cutting costs for patients.
Logistics. Companies take advantage of DeepSeek’s optimisation algorithms to cut operational costs by streamlining supply chains and reducing delivery times.
Retail. Stores use advanced consumer behavior analysis provided by DeepSeek’s AI to develop hyper-personalized shopping experiences for their customers.
Making a geopolitical statement
The DeepSeek-R1’s release comes at a time when the political and economic tensions between the US and China have started to heat up even further.
The US export ban on advanced chips has made it difficult for many Chinese tech firms to secure the necessary components for their AI developments.
As a Chinese company itself, DeepSeek was not immune to this struggle. Its founder, Liang Wenfeng, reportedly collected around 50,000 Nvidia A100 chips despite the ongoing ban. He then combined these Nvidia chips with cheaper and lower-end ones when he founded DeepSeek in 2023.
The successful launch of DeepSeek’s AI has raised questions about the effectiveness of the US chip ban in curbing artificial intelligence development in China.
Not only was DeepSeek able to develop an AI to rival that of Google and Microsoft’s, the company was even able to produce it at such a very low cost.
Also Read: The most promising AI developments today
And it appears Washington is not the only one rattled by the DeepSeek-R1’s release as the stock market also felt its ripple effects. US tech stocks have reportedly dropped in value, with AI giants like Nvidia, Meta, and Google bearing the brunt.
In Nvidia’s case, the popular chipmaker’s stock fell nearly 17% and lost $588.8 billion in market value, which is the most market value a stock has ever lost in one day.
Fiona Cincotta, a senior market analyst at financial services firm City Index, offered an explanation as to why the market reacted as it did to the DeepSeek launch.
"This idea of a low-cost Chinese version hasn't necessarily been forefront, so it's taken the market a little bit by surprise," Cincotta told the BBC.
"So, if you suddenly get this low-cost AI model, then that's going to raise concerns over the profits of rivals, particularly given the amount that they've already invested in more expensive AI infrastructure."
Despite these, some observers view DeepSeek’s entry into the AI race with some degree of optimism.
In a post on X, leading tech investor and avid Trump supporter, Marc Andreessen, lauded DeepSeek’s achievement, calling the launch “one of the most amazing and impressive breakthroughs” he has ever seen.