OpenAI Rival DeepSeek Shakes Up AI Universe: What Exactly Did It Do?
DeepSeek-R1, designed to run on less-advanced chips, challenges Nvidia, Meta, and Microsoft by bypassing US tech restrictions, sparking concerns over AI market dynamics.
In a move reminiscent of a tech underdog stealing the show, DeepSeek—a Chinese AI startup founded by quant fund chief Liang Wenfeng—has flipped the global AI narrative. Its latest model, DeepSeek-R1, offers comparable performance to Western AI giants at a fraction of the cost, raising questions about America's technological dominance.
Released last week, DeepSeek-R1 is an open-source AI model that emphasises efficiency and transparency. The model, widely seen as competitive with those of OpenAI and Meta Platforms Inc., is designed to run on less-advanced chips, bypassing the US-imposed restrictions on cutting-edge semiconductor exports to China. Its app has surged to the top of Apple’s App Store rankings.
The launch of DeepSeek-R1 sent shockwaves through global markets. Investors panicked over the startup’s ability to deliver high-quality AI at a fraction of the cost. The launch triggered a selloff in global tech stocks, with shares of major US AI-related companies, including Nvidia, Meta, and Microsoft, dropping in response to DeepSeek's success.
Nvidia Corp., synonymous with AI chip innovation, saw its shares close nearly 17% lower, while ASML Holding NV ended 7% down. The tech-heavy Nasdaq 100 dipped nearly 3%, while the S&P 500 dropped 1.5%. Markets across New York, London, and Tokyo stock exchanges felt the shocks.
Nvidia has called the Chinese model and "excellent AI advancement", compliant with US technology export controls around high-end AI hardware. Notably, running the AI model needs a lot of high-performance GPUs and networking, which made the US-based company's worth trillions of dollars.
DeepSeek's Recipe To Success?
But the bigger question is — what is the secret sauce behind DeepSeek's success? DeepSeek-R1 delivers impressive performance compared to OpenAI's and Meta's models but runs on less-advanced chips. Its efficiency stems from innovative multi-stage training and large-scale reinforcement learning, bypassing the need for expensive high-tech infrastructure. The breakthrough makes AI accessible to users and companies on a tighter budget.
Its rise highlights how Chinese engineers are navigating US sanctions, proving that innovation does not always require cutting-edge technology. By developing models that rely on fewer resources, DeepSeek challenges the notion that China’s AI lags behind its Western counterparts.
Despite US bans on advanced semiconductor exports, DeepSeek has sidestepped these restrictions by optimising efficiency with accessible open-source technology.
DeepSeek’s strategy of offering free and unrestricted AI access poses a direct challenge to the costly infrastructure investments of giants like Nvidia, Microsoft, and Meta. While Meta plans to increase AI spending to $65 billion, OpenAI, SoftBank, and Oracle recently announced a $100 billion venture to expand AI infrastructure in the US.
As global investors monitor DeepSeek’s next moves, its market disruption highlights a shifting balance in technological power.
