Chinese AI startup DeepSeek has unveiled AI models with comparable features and functionality of some of its most popular US rivals at a fraction of the development cost.
The claim has undermined the prevailing idea that AI development requires exponential levels of funding, compute power and energy, calling vast US AI investments into question.
Go deeper with GlobalData
Markets reacted sharply today (27 January) as tech stocks tumbled at speculation that DeepSeek may have just burst the AI bubble. DeepSeek’s claim to have achieved parity with its US rivals at a fraction of the cost was described by veteran technology investor Marc Andreessen as AI’s Sputnik moment.
Market reaction follows the release of DeepSeek’s new model series DeepSeek R1 on 20 January which has seen its mobile app since surge to the top of Apple’s iPhone download charts.
DeepSeek was founded in 2023 by China’s own Sam Altman character, Liang Wenfeng, who also founded the startup’s financial backer Chinese hedge fund High-Flyer. Wenfeng is said to have stockpiled NVIDIA A100 chips now under Biden era US export restrictions.
DeepSeek claims to beat any competition among open-source models and “rival the most advance closed-source models globally.” The company claims its DeepSeek R1 has a “performance on par with OpenAI-o1.” Most critically, the startup claims to have achieved this at a cost of $5.6m, a fraction of the vast sums spent by its US rivals, as well as using far fewer AI chips. For comparison, OpenAI is said to have spent around $7bn in 2024 on training its AI models.
How well do you really know your competitors?
Access the most comprehensive Company Profiles on the market, powered by GlobalData. Save hours of research. Gain competitive edge.
Thank you!
Your download email will arrive shortly
Not ready to buy yet? Download a free sample
We are confident about the unique quality of our Company Profiles. However, we want you to make the most beneficial decision for your business, so we offer a free sample that you can download by submitting the below form
By GlobalDataGlobalData principal analyst Gavin Sneddon noted that there may questions around the accuracy of DeepSeek’s published costings and exactly how open source it is, but if the claims are within an order of magnitude, significant implications still arise.
“As more information emerges about the real development costs incurred by Deepseek, it can be expected that both the government and Big Tech will begin to look more closely at how to measure the return on investment of AI infrastructure going forward,” said Sneddon.
The assumption has been that the restrictions on chip, technology and research which western countries, particularly the US have imposed on China would slow down AI development in the country.
“However, while there are conflicting narratives around this, the possibility must now at least be considered that the impact of these measures has been more similar to that of an athlete ‘training at altitude’ and has, in fact, enhanced the creativity and efficiency of Chinese AI teams who have been forced to deliver against considerable resource constraints,” said Sneddon.
DeepSeek shows enterprise AI should be cheaper
According to London School of Economics fellow and director of the European Centre for International Political Economy, Hosuk Lee Makiyama, the focus should not be on how DeepSeek achieved parity with its US rivals, but rather, that OpenAI and other Silicon Valley GenAI services have not yet invented a commercial use-case that justifies the capital raised from investors.
“DeepSeek has maintained a lower pricing thanks to lower costs from training smaller models and cheaper hardware infrastructure – which is largely thanks to US export controls. Ironically, the limitations have made R1 better adapted for the commercial reality of GenAI monetisation and enterprise integration,” said Makiyama.
Increasing investor unease at the vast amounts spent on developing AI models in the US has prompted valid questions about return on investment. The cost of building large language models has been passed on to enterprise end users with vendors such as Microsoft charging $30 per seat for its Copilot GenAI feature, for example.
GlobalData senior analyst Beatriz Valle notes that GenAI propositions are actually becoming more expensive as OpenAI’s ChatGPT Pro business product is $200 per seat per month. “This may not be sustainable over the longer term. DeepSeek and other startups will be exerting downward pressure on prices for Western counterparts,” said Valle.
According to macroeconomic forecasting firm TS Lombard’s China expert, Rory Green, the pricing issue is hard to judge at this stage. “DeepSeek may not trigger immediate price cuts but as the technology is open source it significantly lowers barriers to entry and therefore competition within the GenAI space will increase bringing with it cost reductions for end users,” said Green.
The revelation of DeepSeek’s capabilities is more broadly the latest example of “China Shock 2.0”, The People’s Republic of China’s exceptionally rapid movement up the tech value chain, noted Green.
“It raises a question we have been talking about for some time – Can anyone beat China? The nation has a historically unique set of characteristics that when combined are making Chinese companies competitive across almost all tech areas,” he said.
The announcement of the new Trump administration’s $500bn AI infrastructure Stargate project, last week, demonstrates the strategic global importance of AI. And DeepSeek may signal a significant win for President Xi Jinping in the AI arms race between China and the US.
China has 30% share of global manufacturing, a developed market level technical ability and emerging market cost. “It’s still cheap,” noted Green. China also has the benefit of enormous state support in areas of finance, politics, human resources, among many others, he added.
“We think the combination will prove very difficult to beat. Last year it was an EV shock, this year AI, next year could be robotics and aviation,” said Green.
President Xi Jinping has long been aware of the importance of AI and has outlined his ambition for China to become the world leader in AI by 2030.
Broad recognition that AI needed to be led by private sector entrepreneurs rather than government prompted President Xi to assign national AI champions Baidu, Alibaba, Tencent, iFlytek and SenseTime to lead the charge.
Ultimately, Chinese competitiveness in areas traditionally dominated by advanced nations will lead to greater trade conflict and national security concerns, warned Green. “However, the rest of the world, will benefit from cheaper advanced manufactured products, EVs and solar are good existing examples,” he said.
