Editorial: DeepSeek tremors

• Barely a week into Donald Trump’s presidency, American arrogance has been delivered a stunning blow by a little-known team of engineers from Hangzhou in southeastern China. Even as the new US President was in full flow with his supremacist agenda, threatening to ‘clean up’ Gaza, buy Greenland, and ‘take back’ the Panama Canal, a firm called DeepSeek has released artificial intelligence (AI) models that are staggeringly superior to and more cost-efficient than the best American offerings developed with significant investments, using enormous computing power and infrastructure. With absolutely no fanfare, the young startup has brought home to the tech-bros around Trump that global dominance of high technology is not a one-horse race.

In just one week since its unveiling, DeepSeek’s latest version V3 has demonstrated an ability to handle complex and nuanced queries better than OpenAI’s ChatGPT. It excels in tasks requiring high contextual accuracy and offers creative and dynamic responses comparable or better in granularity to the best American models out there. It has seamless multilingual capabilities with accurate translations, which is something all the apps put out by the big American companies have struggled with.

DeepSeek’s achievement is astounding on several counts. It is reported to have spent just six million dollars on development compared to billions spent by OpenAI alone. It did so without access to the top-end Nvidia A100 chips as they are sanctioned against sale to China, and so relied on cheaper, lower-end ones. Not least, DeepSeek is open-source unlike OpenAI, which means developers anywhere can build further upon it.

DeepSeek’s impact has been immediate and wide-ranging. The entire universe of AI stocks in the US went into a tailspin as the Chinese app shot straight to the top of the charts on Apple’s App Store, beating OpenAI’s ChatGPT to second place. The chatbot has been universally acclaimed by experts for the quality of its responses.

Western critics of course harp on DeepSeek’s reticence on the 1984 Tiananmen Square rebellion but skirt clear of OpenAI’s deference to authoritarian regimes everywhere and ignore entirely the question whether any of the world’s self-claimed custodians of democracy remain truly democratic at all—including US and India.

DeepSeek’s success contains lessons for India. This is what Aatma Nirbharata looks like. China’s ability to stand up to American sanctions and tariffs and still establish pre-eminence in electric mobility, AI, rare-earth metallurgy and climate-smart technology is the result of far-sighted policymaking followed by concrete implementation.

India in comparison has done little to attract, retain and leverage its talent pool in IT, preferring instead to indulge in incantations. While Indian and Chinese students make a beeline for frontline US technology institutes, the latter are lured back and encouraged to innovate, build and manage their country’s new economy while ours stay back to serve our remittance economy.

At a time when India is bending over backwards to Trump hoping to save the H-1B programme, China’s DeepSeek illustrates our lost opportunities. When OpenAI’s Sam Altman visited India in 2023, he was asked by an Indian CEO whether we could, with “about $10 million and three bright IITians” build a large language model (LLM) on the lines of ChatGPT. “You could try,” Altman said. “But it’s hopeless to try to compete with OpenAI on training foundation models.” It sounded insulting, but he meant that India was not deploying the resources necessary to compete. China did, and here we are.