DeepSeek breakthrough, Nvidia stock rout raise concerns over need for huge AI spending

DeepSeek, a little-known Chinese start-up whose affordable technology matches the likes of the most formidable American tech titans such as OpenAI and Meta Platforms, has raised investor concerns over the necessity of pouring billions of dollars into research and development. The Chinese artificial intelligence (AI) company released its open-source reasoning model, R1, earlier this month. The model’s capabilities roughly match those of advanced models from OpenAI, Anthropic, and Google while having significantly lower training costs. “The impressive performance/cost dynamics have raised investor concerns about the necessity of the billions of dollars of capital expenditures by large US tech companies and the billions more they are planning to spend on generative AI in the coming years,” said Malik Ahmed Khan, an equity analyst from Morningstar. Large US tech companies may follow suit and replicate some of the AI training techniques that DeepSeek leveraged to drive the cost of R1 down, he said in a research note on Tuesday. When DeepSeek launched, it quickly became the most downloaded app in Apple’s US AppStore. The better-than-expected performance of DeepSeek led to a sell-off of Nvidia and tech peers on Wall Street on Monday. Its shares plunged 17 per cent to US$118.58, erasing almost US$600 billion of market value. The narrative on Monday was that the eye-watering sums spent on AI by mega-cap tech companies could be somewhat obsolete if a cheaper solution exists, analysts at Montreal-based BCA Research said. In essence, the moat protecting the “Magnificent Seven” US tech giants may not be as wide as previously thought, it said. “This was bound to happen, and it could have come from anywhere,” said Alicia Garcia-Herrero, chief economist for Asia-Pacific at French investment bank Natixis. “It came from China, all the more so, in a way, to see a correction.” Wei Sun, principal analyst for AI at Counterpoint Research, said the Nvidia sell-off reflects shifting perceptions in AI development. “DeepSeek’s success challenges the belief that larger models and more computing power drive better performance, posing a threat to Nvidia’s GPU-driven growth strategy,” Sun said. “It also questions the need for massive investments like the US$500 billion Stargate Project. By achieving cutting-edge results with fewer resources, DeepSeek highlights the potential of efficient innovation over sheer scale, marking a turning point in the AI race.” While the correction may have been an initial reaction to the hype around DeepSeek, Garcia-Herrero said she does not believe it would be a permanent correction. “The US is going to double down on investment in AI,” she said. “There will be more money in the US for the AI industry, if not more than before. Because [previously] it was kind of the understanding that China was falling behind, especially in Gen AI and now we see that was not the case.” US President Donald Trump has called DeepSeek’s strong showing a “wake-up call” and “positive” for America’s tech sector. He also warned that he would slap tariffs on foreign-made semiconductors, including chips made by Taiwan Semiconductor Manufacturing Company or TSMC, if they do not start producing them stateside. Despite DeepSeek not having access to leading AI accelerators, such as Nvidia’s best Hopper and Blackwell products, its large-language model (LLM) is believed to be on par with some of the best models released by OpenAI and others. “DeepSeek raises concerns that future LLMs will be developed with fewer AI GPUs from Nvidia, lower capital expenditure from hyperscalers, and perhaps less of a need for a massive energy buildout in the US,” said Brian Colello, strategist at Morningstar. Nvidia on Monday responded to the hype generated by the Chinese AI firm and said its advances show the usefulness of its chips for the Chinese market and that more of its chips would be needed in the future to meet demand for DeepSeek’s services. DeepSeek claims to have trained its model on only 2,048 Nvidia H800 GPUs over a two-month period, which at US$2 per hour explains the US$5 million headline number announced. This is driving a reassessment of AI investment strategies, focusing attention on the sustainability of AI capital expenditures, the AI competitive landscape, and the monetisation of AI, according to Richard Clode, a portfolio manager at Janus Henderson. DeepSeek is charging significantly less than OpenAI to use its models – about 20-40 times lower- according to Janus Henderson. This adds to concern about AI monetisation, given the extraordinary amounts of capital outlay deployed in the West. Last year, OpenAI said it closed a long-awaited funding round at a valuation of US$157 billion, including the US$6.6 billion the company raised from an extensive roster of investment firms and big tech companies, including Microsoft and Nvidia. Analysts say that DeepSeek shows a Chinese company can compete with the best-of-breed AI companies in the US, despite the current restrictions on Chinese access to advanced US semiconductor technology. “The breakthrough could be that DeepSeek can achieve a state-of-the-art model without state-of-the-art equipment,” according to analysts at ING. According to ING, the market needs to price in a slower adoption rate of high-performance computing and revenues of companies selling AI models.

DeepSeek breakthrough, Nvidia stock rout raise concerns over need for huge AI spending

.css-1c6uqr6{color:inherit;font-weight:inherit;font-size:inherit;font-family:inherit;line-height:inherit;overflow-wrap:break-word;}Large US tech companies may follow suit and replicate some of the training techniques that DeepSeek leveraged to drive the cost of R1 down

Large US tech companies may follow suit and replicate some of the training techniques that DeepSeek leveraged to drive the cost of R1 down