Shri. Shubhradeep Nandi, Chief Data Scientist, Data Analytics Unit, Government of Andhra Pradesh, praises DeepSeek for innovation and a carefully planned approach at the Convergence India Expo 2025 at Pragati Maidan, Bharat Mandapam. He admires the Chinese innovators who found a way to beat America’s sanctions with limited-capacity GPUs to shock the world with their world-class foundational AI model. He says, “As it came, it took the world by storm, causing huge slumps in the markets worldwide, especially that of the US.” 

Mr. Nandi says that “ US regulation mandated Nvidia not to supply its high-end GPUs to China or the Chinese companies.” This is aimed at disabling China from developing foundational AI models. He goes on to explain the contributing situation to the development of DeepSeek into 3 segments. It includes governance and policy, AI innovators, and, thirdly, the market.

Innovation Making Its Way

Owing to US regulations, Nvidia developed H-800 chips to satisfy the chip demand from China. H-800 comes with less computational power than the H-100 and H-200 Graphics Processing Units (GPUs). It is a crippled version of the high-end GPUs. The Chinese AI innovators took these and played the slayest game.

To read what Jaya Jagadish, Country Head and Senior VP of Silicon Design Engineering at AMD India, said about AI at CI 2025, click here!  

The Chinese innovators took 2048 of the H-800 GPUs and clustered them to generate a handsome amount of computational power. They also took inspiration from the original transformer architecture to tweak this cluster GPU for performance. Finally, they came up with their foundational model at nearly 1/10th of the cost incurred by their Western counterparts.

Super Efficient Approach 

While DeepSeek is built with low computational capacity, it doesn't lag behind its Western counterparts in responses. It is because, unlike its competitors, it doesn't use all of its computational power at once. As Shri Nandi says, “DeepSeek doesn’t activate the complete computational block; it takes fractions out of it and works on them.” This approach enables it to deliver a highly efficient performance.

Effect on Market 

The fact that you don't need high-end GPUs to make a foundational AI model took the markets around the world by surprise. Many investors lost their confidence in the market, and major economies took a huge plunge into depression and devaluation. Although the DeepSeek revolution has returned to the pavilion among continuous launches from Western AI companies, the lessons dispersed remain very important to lead the way forward in AI development. 

To read about OpenAI's all-new GPT 4.1 and how it will benefit the developers, click here!

Convergence India, 2025 

The 32nd Convergence India Expo, held alongside the Mobile India Expo at Bharat Mandapam in New Delhi, serves as a key platform for driving India's technology-led future. With more than 45 conference sessions, the event brought together industry leaders, innovators, and government officials to explore emerging technologies and evolving market trends.