DeepSeek’s disruption should open gate for Indian AI initiatives

China's DeepSeek outperforms US models at lower cost, sparking fears of reverse engineering and government censorship
Image used for representational purposes.
Image used for representational purposes.
Updated on
2 min read

A new Chinese artificial intelligence model has rattled American tech giants. The latest release from DeepSeek, a startup based in Hangzhou and controlled by a hedge fund, has bettered or matched American large language models (LLMs) on cost and performance in such a manner that it has upended some long-held assumptions in the tech world. The disruption is so epochal that some investors are calling it China’s ‘Sputnik moment’—referring to the time the Soviet Union beat the US in the space race by launching the world’s first satellite in 1957. It started with the release of DeepSeek’S open-source models R1, suited for advanced reasoning tasks, and V3, for scalable natural language processing. By costing a fraction of OpenAI’s ChatGPT 4o and o1, they challenged the long-held assumption that developing such LLMs requires significant technical and financial resources.

But users are less concerned about cost. It’s the performance of the DeepSeek models that made their demand zoom, with R1 becoming the most downloaded free app in the US within days of launch. The shares of NVIDIA, the company whose chips power ChatGPT, crashed so steeply that it wiped $600 billion off its market value in a day. It impacted other tech giants, too—Microsoft, Oracle, Meta and Broadcom among them—even as it wowed their bosses.

The developments have sparked speculation that DeepSeek resorted to what the Chinese are notorious for—reverse engineering. Especially so because DeepSeek, co-founded by Liang Wenfeng in 2023, developed R1 despite the US’s embargo on China from availing of tech hardware such as NVIDIA’s high-end graphics processing units (GPUs); the Chinese company innovatively used a less powerful, cheaper GPU. There are also concerns that DeepSeek suppresses data embarrassing for the Chinese government—for one, questions about Arunachal Pradesh receive circuitous answers.

Having taken note of China’s strides, the government announced on Thursday that it has selected 18 application-level AI solutions for the first round of funding under the IndiaAI Mission. It has also selected 10 companies to supply GPUs and planned a project to build India’s own LLM. About a year ago, OpenAI founder Sam Altman told a business gathering in India that Indian companies starting the AI journey were unlikely to touch what the US Big Tech had achieved because of the costs and technology. Now that the calculations have changed, India must aim higher in the AI space.

Related Stories

No stories found.

X
The New Indian Express
www.newindianexpress.com