
The Chinese AI startup DeepSeek has stirred up social media with reports of a cheap and efficient AI model, DeepSeek-R1. Analysts from several major brokerage and investment firms caution that concerns about its immediate impact may be overblown.
Bernstein
- DeepSeek has not built an “OpenAI for $5 million.”
- While the models look impressive, they are not miraculous.
Bernstein analysts note that DeepSeek’s potential tenfold reduction in training costs is critical, as the expense of creating advanced neural networks typically grows by roughly the same amount each year. However, the freed computing resources will likely be consumed by the rising demand for AI.
The experts also point out that many high-caliber researchers from other labs are familiar with the innovations DeepSeek claims, and they may already have employed similar techniques.
Morgan Stanley
Morgan Stanley could not confirm the accuracy of DeepSeek’s claims about building advanced models with limited funding. Should the data prove correct, generative AI could run on progressively less powerful hardware, benefiting the sector by boosting product demand.
Goldman Sachs
Goldman Sachs analysts highlighted:
- Potential competition among projects from big-tech firms and startups.
- Transition to “post-training” methods requiring far fewer compute resources.
- Prospects for further global expansion of Chinese AI players.
JPMorgan
JPMorgan confirmed a general trend of falling expenses for AI model training. The bank noted that major companies no longer hold undeniable advantages:
“OpenAI is not a god; they’re not necessarily always going to be at the forefront.”
Jefferies
Jefferies stressed the positive effect that more compact AI models could have on the smartphone sector. Despite the hope DeepSeek provides, the firm is skeptical of running AI on mobile devices due to insufficient consumer demand for on-device AI services.
The company emphasized that China is the only market in which startups are motivated to increase model efficiency, in large part due to a scarcity of advanced chips. Observers say Elon Musk and Donald Trump might realize that more innovation in China could force the U.S. president to ease restrictions on AI technology.
Citi
At Citi, analysts remain doubtful of DeepSeek’s progress without cutting-edge GPUs:
“Although American dominance in building the most advanced AI models can potentially be challenged, we believe the inevitable tightening of restrictions favors U.S. access to more advanced chips. We do not expect leading AI firms to abandon high-end GPUs.”
Context
- In January, U.S. President Donald Trump announced $500 billion in private-sector investments in AI infrastructure, but experts voiced skepticism over such large-scale funding.
- In response, China disclosed plans to invest 1 trillion yuan ($137.5 billion) over five years to support AI companies financially.