Chinese AI model threatens US dominance with a fraction of the cost

DeepSeek causes US tech rout: A China-based AI startup has thrown US dominance in the AI field into doubt after it developed competing technologies on a much smaller budget than its American counterparts, the Financial Times reports. US giant Nvidia’s stock price fell almost 17%, wiping USD 589 bn in market cap in the biggest one-day drop ever recorded for a US company. Tech-heavy indexes took a hit, too: the Nasdaq Composite was down 3%, while S&P fell 1.5%.

A fraction of the cost: DeepSeek spent only USD 5.6 mn training its R1 model, which “performed similarly for around one-fourth of the cost,” Anthony Poo, co-founder of a Silicon Valley-based startup, told the Wall Street Journal. Silicon Valley’s tech giants spend between USD 100 mn and USD 1 bn training each model, according to estimates from Dario Amodei, CEO of US-based AI startup Anthropic.

How DeepSeek achieved it: DeepSeek’s founder, Liang Wenfeng, used proceeds from his hedge fund High-Flyer to finance DeepSeek’s research and development activities, attracting China’s top AI talent with the highest salaries offered in the country. With little more than 2k Nvidia H800 chips — US firms used tens of thousands — they developed the model while sometimes skipping steps US developers thought essential.

DeepSeek claims that the R1 model can outperform US counterparts, and the claim is not without merit. The model seems to outperform rival models on many third party tests, according to CNBC. Some believe OpenAI still has an edge, but several companies are already moving to using DeepSeek as a more cost-efficient option, WSJ reported.

US attempts to curb China’s AI advancements through sanctions seem to have had the opposite effect. China-bound Nvidia chips were capped at half speed capacity before Washington banned the export of Nvidia chips to China altogether last year. The move forced DeepSeek’s engineers to maximize the computing power of existing chips with limited resources. DeepSeek had also started stockpiling Nvidia A100 chips before the US imposed sanctions, prioritizing research and resource-pooling to develop models despite limited resources.

All in the open: DeepSeek opted to release its AI model open-source, in a departure from competition-focused US companies, which makes the rise of more and more competitors to rival US tech giants more likely. This approach might cause Silicon Valley giants to reconsider their high-cost strategies — think Trump’s USD 500 bn megaproject Stargate — in order to keep up with the changing landscape.

The regional angle: DeepSeek upending the AI market could have significant ramifications for the GCC’s AI push. Saudi has been looking to both the US and China for AI technologies, after Saudi officials negotiated with the US government to secure Nvidia H200 chips last year and an Aramco fund invested in Chinese startup Zhipu AI. Meanwhile, the UAE started receiving chips from Nvidia last September, including Nvidia’s high-performing H100 chips.

MARKETS THIS MORNING-

Asian markets are slightly inching down this morning, with Japan’s Nikkei down 0.6% and Shanghai Composite down 0.1%. Meanwhile, Wall Street futures are seeing little change after DeepSeek triggered tech sell-off on Monday.

THE CLOSING BELL-

The DFM fell 0.7 % yesterday on turnover of AED 614.4 mn. The index is up 0.6% YTD.

In the green: Al Salam Sudan (+9.6%), National Industries Group Holding (+4.4%) and GFH Financial Group (+3.5%).

In the red: National International Holding Company (-3.3%), Emaar Development (-13.8%) and BHM Capital Financial Services (-2.1%).

Over on the ADX, the index fell 0.1% on turnover of AED 924.1 mn. Meanwhile, Nasdaq Dubai fell 0.6%.