Can China’s low-cost and open-source DeepSeek democratise the AI scene?
The small Chinese start-up is making waves in the tech world with its cheaper model that challenges the dominance of Silicon Valley giants.
A previously little-known Chinese company has shaken up the tech world with its groundbreaking approach to large language model training and accessibility, rivalling US AI giants at just a fraction of the cost.
Hangzhou-based AI company, DeepSeek claims it took just two months and spent under $6 million to build an AI model using Nvidia's less-advanced H800 chips, while OpenAI, Meta, and Google have burned billions on similar AI models.
However, what made DeepSeek stand out was not only its low cost but also its open-source philosophy, a radical move that has set the cat among the pigeons in the fast-growing sector.
The R1 model, released under the MIT licence, allows anyone to download, adapt, and fine-tune it. While it complies with Chinese content moderation rules within its home country, users outside of China enjoy full freedom to modify the model — a feature also adopted by Perplexity AI.
Using DeepSeek's R1 through @perplexity_ai. The beauty of open source models. pic.twitter.com/CiDXDKY8Pg
— perpetual compounder (@perpcompounder) January 27, 2025
Following the R1 model announcement, stocks of American big tech companies, including Nvidia and OpenAI-backed firms, saw sharp declines, with a combined loss of $1 trillion in market value.
US President Donald Trump described DeepSeek as a “positive development” but warned it should serve as “a wake-up call” for American industries to maintain their competitive edge.
While Nvidia described R1 as “an excellent AI advancement”, one of the most influential tech venture capitalists in Silicon Valley, Marc Andreessen described it as “AI’s Sputnik”.
OpenAI chief executive Sam Altman also praised the Chinese rival, saying it was "invigorating to have a new competitor."
Now the Chinese company’s cost-friendly approach is raising an intriguing possibility: could this mark the dawn of a more inclusive and accessible AI era, where innovation is no longer the privilege of only the tech behemoths?
deepseek's r1 is an impressive model, particularly around what they're able to deliver for the price.
— Sam Altman (@sama) January 28, 2025
we will obviously deliver much better models and also it's legit invigorating to have a new competitor! we will pull up some releases.
‘Reinforcement learning’
DeepSeek’s success comes from its reliance on reinforcement learning (RL), an approach that has enabled the company to avoid the resource-heavy methods favoured by its competitors.
RL method allows systems to develop advanced reasoning skills by rewarding correct outcomes without needing pre-labelled datasets.
The company’s R1 paper reveals that its model was trained entirely through this trial-and-error method.
Despite this unconventional method, the model achieved results comparable to OpenAI’s latest releases, excelling in areas like coding challenges, mathematical problem-solving, and general reasoning tasks.
🚀 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
“Open-source models may face slower initial development due to limited resources, but they benefit from a wider range of contributions and perspectives, potentially leading to more robust and adaptable solutions in the long run,” says Cagatay Odabasi, a research engineer at Fraunhofer IPA in Stuttgart, Germany.
He added that large companies cannot match the “massive scale of human resources” that naturally comes from a community-driven approach, while they can also lack the massive funding that large companies possess.
The company's open-source model, starting at just $0.50 per month, has started to challenge the dominance of costly, closed-source AI models, becoming the No. 1 downloaded free app on Apple’s App Store on Monday.
Update: Deepseek is now number 1 on the app store, across all categories.
— Arnaud Bertrand (@RnaudBertrand) January 27, 2025
Quite literally, China has now officially taken the lead in AI. Ironically 3 days after Trump signed an executive order "to sustain and enhance America’s global AI dominance" (https://t.co/kaPCmi1OtG) https://t.co/UKhDN2SjVg pic.twitter.com/g4QkJpBC9r
“This openness undoubtedly pressures AI providers to innovate more aggressively to stay competitive, reduce prices, and offer better terms regarding user privacy,” Odabasi tells TRT World.
“It democratises access to powerful AI, allowing smaller players and researchers to build upon existing work and contribute to the field's advancement.”
This disruption is particularly visible when contrasted with past comments from Sam Altman, the CEO of OpenAI.
In 2023, Altman dismissed the idea that a small team with a $10 million budget could compete in AI, calling it “totally hopeless”.
DeepSeek’s achievements now challenge that assertion, demonstrating that innovation isn’t solely the domain of tech giants.
“I believe that once such a powerful model is made open, the community will always find innovative ways to utilise and even enhance it efficiently,” Odabasi adds.