Quality Matters. Zuckerberg Compares His AI to Much Cheaper Models from DeepSeek

Mark Zuckerberg is convinced that the significant costs of developing artificial intelligence and infrastructure for it were justified and will become a strategic advantage for Meta.

During a conversation with Wall Street analysts, Zuckerberg was asked several questions about DeepSeek's artificial intelligence models and the implications of the Chinese startup's success for Meta's artificial intelligence strategy. He said that what DeepSeek was able to achieve for relatively little money “only reinforced the belief that this is what we need to focus on,” The Verge writes.

Zuckerberg tried to dispel concerns that his company wasted billions of dollars on graphics processors during the development of AI. According to him, the success of the low-cost DeepSeek does not mean that less computing is needed.

«[We need] to apply more computing, … to create higher levels of intelligence and a higher quality of service. … I continue to think that very significant investments in capital expenditures and infrastructure will over time become a strategic advantage», — he said.

Zuckerberg also noted that there are «a number of new things in the viral Chinese AI models that are still digesting [Meta]». The company aims to implement the best of DeepSeek in its Llama AI models. Meta is currently preparing to release Llama 4 with multimodal and agent-based capabilities. Release is expected in the coming months.