DeepSeek’s AI breakthrough: How it may change the AI landscape, the US-China rivalry, and more

The emergence of DeepSeek, a Chinese AI company, has sent ripples through the global artificial intelligence landscape. For a while, it looked like the US tech giants will become a dominant force with China playing catch-up. DeepSeek’s recent release of its open-source reasoning model, R1, which reportedly outperforms leading models from US firms like OpenAI, has not only showcased China’s growing capabilities in AI but also reignited the debate over the future of AI innovation, the role of geopolitics, and the sustainability of the current AI arms race.

The cost factor

Reports have suggested that DeepSeek’s R1 model has been developed at a cost of less than $6 million. It is in stark contrast to the billions of dollars spent by Silicon Valley companies to build their AI models — and are continuing to do so.

DeepSeek’s success lies in its use of advanced techniques like Test Time Scaling, which optimises model performance during inference, and its strategic utilization of Nvidia’s export-compliant GPUs tailored for the Chinese market.

Nvidia, the US-based chipmaker, has publicly praised DeepSeek’s architecture, calling it an “excellent advancement” and highlighting its reliance on export-compliant hardware.

Jim Fan, senior research manager, Nvidia in a post on X said tech companies shouldn’t be worried about how DeepSeek managed to achieve a breakthrough at lower costs. “Many tech folks are panicking about how much DeepSeek is able to show with so little compute budget. I see it differently - with a huge smile on my face. Why are we not happy to see *improvements* in the scaling law?”

He also said that “DeepSeek is unequivocal proof that one can produce unit intelligence gain at 10x less cost, which means we shall get 10x more powerful AI with the compute we have today and are building tomorrow. Simple math! The AI timeline just got compressed,” he said.

A new arms race in the making?

The rise of DeepSeek could become a big part of the broader US-China rivalry in AI, a domain increasingly seen as critical to economic and military power. The US stole the march with OpenAI, Google, and Microsoft, releasing models such as GPT-4 and Gemini. However, China’s rapid advancements, fueled by government support, a vast talent pool, and access to cutting-edge hardware, are set to challenge this dominance.

DeepSeek’s success also highlights the limitations of US export controls aimed at curbing China’s access to advanced AI technologies. Despite restrictions on high-performance chips, Chinese firms like DeepSeek have found ways to innovate using export-compliant hardware, raising questions about the effectiveness of such measures.

An editorial published in the Beijing Daily on January 27 noted that DeekSeek’s early success “delivers a resounding slap in the face to those attempting to maintain a technological advantage through isolation”. No names were taken but it was a pointed comment towards the US.

What it means for the future of AI?

DeepSeek’s rise calls into question some of the fundamental assumptions underpinning AI development. For the last few years now, the AI industry has operated on the belief that bigger is better—that larger models trained on massive datasets with exorbitant computational resources are the key to achieving superior performance. DeepSeek’s R1 model, however, demonstrates that efficiency and innovation can rival sheer scale.

This shift challenges the sustainability of the current AI arms race, which has seen companies pour billions into ever-larger models, often at the expense of environmental and economic costs. DeepSeek’s cost-effective approach suggests that the future of AI may lie in optimising existing resources rather than perpetually scaling up.

Moreover, DeepSeek’s open-source strategy stands in contrast to the proprietary models favored by many US firms. The only exception is Meta’s Llama models.

Can China surpass the US?

It’s a bit early to say because the emergence of DeepSeek will see the likes of OpenAI and Google up the ante. OpenAI CEO Sam Altman has already indicated that their models will be “pulled up” for earlier release. OpenAI’s o3 model, which was unveiled earlier this week, is tipped to be much better than the existing models. Altman had said that the model has been designed to be better at coding, solving math and science problems.

Going forward, it is expected that Google too will flex its AI muscles and Microsoft too will throw its might. DeepSeek’s rise does set the stage for an intriguing battle in the AI landscape.

Ethan Mollick, professor at University of Pennsylvania’s Wharton school, and an AI specialist, broke down capabilities of LLM models: A comparison