The AI startup DeepSeek has stirred excitement on social media with news of its affordable and efficient AI model, DeepSeek-R1. However, experts urge caution in drawing conclusions.
Bernstein
Analysts at the brokerage firm Bernstein described the weekend panic on X as overblown. Key takeaways include:
- DeepSeek has not built “OpenAI for $5 million”;
- the models appear impressive but are not miraculous.
DeepSeek is reportedly capable of achieving equivalent performance at a tenth of the cost. Experts suggest considering that the development cost of existing neural networks increases by about the same amount each year. Thus, such innovations are crucial for the further advancement of artificial intelligence.
The freed-up computing power will be absorbed by the growing demand for it, Bernstein added.
Analysts also emphasized that DeepSeek’s innovations have long been known to a large number of top-tier researchers from other labs. They have clearly considered and possibly even used similar strategies themselves.
Morgan Stanley
Morgan Stanley could not verify the development of advanced models with minimal investment. If the information is true, generative AI will operate on increasingly less powerful PCs, and the industry will benefit from the accompanying demand for products.
Goldman Sachs
Analysts at Goldman Sachs noted:
- the potential competition between projects of giant companies and startups;
- a shift to post-training, which requires significantly fewer computing resources;
- the potential for further global expansion of Chinese players.
JPMorgan
JPMorgan confirmed a reduction in AI model training costs. The bank emphasized that large players currently have no clear advantages.
“OpenAI is not a deity; they will not necessarily always be at the forefront,” the firm noted.
Jefferies
The international financial company Jefferies highlighted the positive impact of more compact AI models on the smartphone industry. While DeepSeek offers some hope for the development of this sector, experts are skeptical about launching AI on mobile devices due to a lack of consumer demand for services.
China is the only market where startups strive to enhance the efficiency of large language models due to a chip shortage, Jefferies added. Elon Musk and Donald Trump may understand that the risk of further restrictions lies in the rapid adoption of innovations in China. This could prompt the US president to ease AI dissemination policies.
Citi
Citi expressed doubts about DeepSeek’s success without the use of advanced graphics processors.
“While the dominance of American companies in creating the most advanced AI models may be potentially challenged, we believe that in the face of inevitable tightening restrictions, US access to more advanced chips is an advantage. Thus, we do not expect leading AI companies to abandon more advanced GPU,” analysts noted.
In January, Donald Trump announced $500 billion in private sector investments in AI infrastructure. Experts expressed doubts about the feasibility of such large-scale funding.
China responded with a plan to allocate 1 trillion yuan ($137.5 billion) over five years to financially support AI companies.