For much of human history, written knowledge was the privilege of the few. While the earliest recorded texts date back to around 3100 BC, when the Sumerians inscribed administrative and economic data onto clay tablets, access to the written word remained scarce for thousands of years. Only a select elite could read and write, let alone produce texts. This exclusivity persisted until Johannes Gutenberg’s printing press in the 15th century, a breakthrough that unleashed the single greatest democratization of knowledge in human history. Since then, literacy and access to information have become nearly universal, fueling technological and scientific progress on an unprecedented scale.

Now, a similar transformation may be unfolding—this time in artificial intelligence. On December 26, 2024, a relatively obscure Chinese company, DeepSeek, shocked the world by unveiling DeepSeek-V3, an AI model that not only competes with industry titans (OpenAI’s ChatGPT, Meta’s Llama, and Alphabet’s Gemini) but also challenges the fundamental economic and technological assumptions of AI development. The model’s release—accompanied by the full publication of its research paper—marks a watershed moment, raising questions about accessibility, cost, and the future of technological supremacy.

The AI race has long been dominated by a handful of American firms. OpenAI’s ChatGPT, Google’s Gemini, and Meta’s Llama have set the standard, with each successive iteration demanding billions in investment and cutting-edge infrastructure. DeepSeek-V3, however, defies this logic. Based on international AI benchmarking tests, it matches or even outperforms these Western models in several domains, despite being trained under vastly different conditions.

First, DeepSeek-V3 is open-source. Unlike its closed, proprietary counterparts, this model allows anyone to access, study, modify, and improve its code. This single decision shatters existing barriers, enabling governments, universities, startups, and even individuals to develop their own AI systems—without relying on American tech monopolies.

Second, the model was trained on older hardware—specifically, Nvidia’s H800 chips—due to US export restrictions that have barred China from acquiring more advanced semiconductor technology since 2022. These restrictions were intended to slow China’s AI progress, but DeepSeek’s success suggests that innovation can thrive even under constraints.

Third, training costs were dramatically lower than those of its American counterparts. While companies like OpenAI and Google sink billions into their models, DeepSeek trained its AI on an estimated $5.6 million, a fraction of what Western firms typically spend. Moreover, it did so without relying on Nvidia’s CUDA platform, long considered the gold standard for large-scale AI development. This signals a potential shift away from Nvidia’s dominance in AI hardware—a development that could reshape the industry.

Finally, DeepSeek-V3 is not just inexpensive to develop; it is also cost-efficient to operate. The computational power it offers is sold at costs 30 times lower than those of equivalent American models, making high-performance AI accessible to a far wider audience of developers.

The impact of DeepSeek-V3 is already reverberating beyond China. In Washington and Silicon Valley, the news has rattled executives and policymakers alike. The consensus assumption—that the US held a three-year AI lead over China—has been abruptly upended. Europe, too, lags behind, as do most other nations that have struggled to keep pace with American AI hegemony.

But it is in smaller nations and mid-sized enterprises where the implications are perhaps most profound. Consider Cyprus, a country that—until recently—would have needed to invest two years’ worth of GDP to develop a national AI model competitive with those of OpenAI or Google. Today, that cost has plummeted to a fraction of what the renovation of Eleftheria Square had cost. Likewise, mid-sized companies that previously had no realistic path into AI development can now build and deploy their own models, sidestepping dependence on Silicon Valley. And this is just the beginning.

DeepSeek has demonstrated that AI innovation is not solely a function of financial resources but of ingenuity, adaptability, and strategic vision.

History has seen such moments before. In the mid-15th century, Gutenberg’s printing press broke the monopoly of scribes and scholars, wresting control of the written word from the hands of the elite and making it available to the masses. The implications were profound: knowledge spread, scientific revolutions followed, and economies transformed.

Now, history may be repeating itself. DeepSeek’s CEO, Liang Wenfeng, has—perhaps unwittingly—become a modern-day Prometheus, taking the fire of artificial intelligence from the few and making it accessible to the many.

There are inflection points in history where technological progress accelerates and power structures shift. December 27, 2024, the day DeepSeek-V3’s technical report was published, may well be remembered as one of them.

By Demetris Herakleous, Associate Economic Affairs Officer at United Nations