As Chinese AI startup DeepSeek shakes up Silicon Valley, there could be implications for India, others in the fray

Triggers for the Market Rout

There could be three broad triggers for the stock market selloff in the West. The big one is the realisation that DeepSeek has managed to train a foundational model to get results similar to those achieved by American rivals such as ChatGPT and Meta’s Llama, but at much lower costs by using far fewer chips. This realisation set off concerns that the red-hot demand for specialised hardware needed to train AI models will now taper off. That is clearly a negative for companies such as Nvidia, which have seen their fortunes turn on demand projections from the global AI boom.

The other big concern is that America’s lead over China in AI looks smaller than at any time since ChatGPT was launched. China’s catch up is startling because it was seen as really far behind. That is disconcerting for many in the West, given that the Biden administration had tried its best to slow down the Chinese AI advance by blocking high-tech chip exports and the high-end machines used to manufacture these chips, to that country.

Then there are worries that the success of cheaper Chinese models threatens to upend the economics of AI evolution. China’s LLMs are decidedly cheaper to develop, with QwQ, a model developed by e-commerce giant Alibaba that was released around the same time as DeepSeek’s R1 ‘reasoning model’ model, emerging as another credible contender in the AI race. The Chinese gains are therefore not a fluke. More importantly, DeepSeek’s model was apparently trained by using 2,000 second rate Nvidia chips versus some 15,000 first class chips for models such as Meta’s Llama, according to the Economist. That changes the conventional perception of how investment intensive the “training” process of a LLM is, before the next stage — “inference” — can be achieved. Inference refers to the process by which AI models generate responses to queries after it has gone through training on vast volumes of data scraped off the internet.

DeepSeek was started by hedge fund manager Liang Wenfeng in 2021, when he started buying thousands of Nvidia graphic processing units for what was then just an AI side project while running his trading fund High-Flyer. DeepSeek’s data crunching was originally aimed at leveraging AI to identify patterns that could affect stock prices. It was much later that the data crunching project turned into a standalone AI venture.

Wake-up call!

Reacting to the advances by DeepSeek, US President Donald Trump termed it a “wake-up call” for US industries. Trump said in Miami on Monday that if DeepSeek’s claims of developing a model that is at par with American models but took far less to build were true, he viewed it as a “positive”. “The release of DeepSeek AI from a Chinese company should be a wake-up call for our industries that we need to be laser-focused on competing to win,” the American president said.

Nvidia, in a reaction after Monday’s stock market rout, said that DeepSeek’s work had been achieved in compliance with US chip export controls, and that the chipmaker continued to see strong demand for AI inference. “DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely available models and compute that is fully export control compliant,” Nvidia said.

Impact on other countries, India

The implications of this for countries such as India is that if foundational AI models can be trained relatively cheaply, then it will dramatically lower the entry barrier for nations keen to build models of their own. The success of DeepSeek and Alibaba models has shown that the fixed cost of building models can actually be brought down, a vital factor for countries that hope to get into this race but are constrained by resources such as GPU availability or the funding needed for setting up a foundational model from scratch and scrounging for the data that it needs to crunch.

This also comes when there is a debate in India over whether to build a foundational model from scratch or rely on some already available open source LLMs to build wrappers on top of them. Infosys co-founder Nandan Nilekani has said that India should not focus on building large language models while others in the AI industry, including Aravind Srinivas, the founder of Perplexity AI, stating publicly that Nilekani’s comment on India not needing to build its own AI models “is wrong.” “…he’s (Nilekani’s) wrong on pushing Indians to ignore model training skills and just focus on building on top of existing models. Essential to do both,” Srinivas said in a post on X. Speaking about the breakthrough made by DeepSeek, Srinivas added in another post: “I hope India changes its stance from wanting to reuse models from open-source and instead trying to build muscle to train their models that are not just good for Indic languages but are globally competitive on all benchmarks.”