Chinese AI firm DeepSeek has emerged as a potential challenger to US AI leaders, demonstrating breakthrough models that claim to offer performance comparable to leading chatbots at a fraction of the cost. The company's mobile app, released in early January, has also topped iPhone charts across major markets including the US, UK, and China.

Founded in 2023 by Liang Wenfeng, former chief of AI-driven quant hedge fund High-Flyer, DeepSeek makes its models open-source and incorporates a reasoning feature that articulates its thinking before providing responses. This approach has sparked mixed reactions from Wall Street analysts, with some seeing potential negative implications for the builders of large-scale AI models, while others believe it could reshape competition between established tech giants and startups by lowering barriers to entry.

Jefferies analysts warn that DeepSeek's efficient approach "punctures some of the capex euphoria" following recent spending commitments from Meta and Microsoft, each exceeding $60 billion this year. They suggest that pressure on AI players to justify ever-increasing capex plans could ultimately lead to a lower trajectory for data center revenue and profit growth. On the other hand, Citi questions whether DeepSeek's results were achieved without the use of advanced GPUs, which could be a crucial factor in the company's cost savings.

Goldman Sachs sees broader implications, suggesting that the development could reshape competition between established tech giants and startups by lowering barriers to entry. They note that the latest developments could lead to potential competition between capital-rich internet giants versus startups, given the lowering barriers to entry, especially with recent new models developed at a fraction of the cost of existing ones.

Bernstein analysts take a more nuanced view, acknowledging that DeepSeek may have reduced costs of achieving equivalent model performance by 10x, but noting that current model cost trajectories are increasing by about that much every year anyway. They believe that innovations like this are necessary for AI to continue progressing and that any new compute capacity unlocked is far more likely to get absorbed due to usage and demand increase.

Morgan Stanley analysts suggest that if advanced LLMs can indeed be developed for a fraction of previous investment, we could see generative AI run eventually on smaller and smaller computers, benefiting the SPE industry. They also note that the potential for further global expansion for Chinese players, given their performance and cost/price competitiveness, could be a significant factor in the future of AI development.

J.P.Morgan analysts highlight DeepSeek's research papers and models' efficiency, noting that they're dramatically reducing costs, such as inference costs for their V2 model, which are claimed to be 1/7 that of GPT-4 Turbo. They quote Liang Wenfeng, who believes that "more investments do not equal more innovation" and that big firms have existing customers, but their cash-flow businesses are also their burden, making them vulnerable to disruption at any time.

The implications of DeepSeek's breakthrough models are far-reaching, and as the tech industry continues to evolve, it will be interesting to see how established players and startups alike respond to this new challenge. Will DeepSeek's innovative approach disrupt the status quo, or will US AI leaders find a way to adapt and maintain their dominance? Only time will tell.