Zuckerberg says Meta won’t slow down AI spend despite DeepSeek’s breakthrough.

Meta CEO Mark Zuckerberg stated on Wednesday that it is still too early to determine how DeepSeek’s breakthrough AI model will impact Meta and the broader tech industry. Despite this, he dismissed the idea that Meta’s AI spending would soon decline, even though DeepSeek’s advancements suggest cutting-edge AI might require significantly fewer financial and computing resources than previously assumed.“It’s probably too soon to draw firm conclusions about how this affects infrastructure and capital expenditures,” Zuckerberg said. “There are multiple trends unfolding simultaneously.” The tech industry was shaken this week by DeepSeek’s AI model, developed by a Hangzhou-based AI lab affiliated with the Chinese hedge fund High-Flyer. DeepSeek claimed it built its R1 large language model in just two months for under $6 million. This revelation caused Nvidia’s stock to drop 17% on Monday, as it raised concerns that companies might need fewer of its GPUs for AI development.

During Meta’s fourth-quarter earnings call on Wednesday, Zuckerberg addressed DeepSeek’s innovation. Meta reported a 21% increase in revenue year over year, reaching $48.39 billion, surpassing Wall Street projections. Meta’s stock rose nearly 2% on Monday, indicating that some investors believe DeepSeek’s cost-efficiency could allow Meta to develop advanced AI systems at a lower cost. The company recently announced plans to invest between $60 billion and $65 billion in AI this year and projected total expenses for 2025 between $114 billion and $119 billion. Zuckerberg mentioned that Meta is still analyzing DeepSeek’s achievements and aims to incorporate some of its advancements into its own AI projects. “This is how innovation works, regardless of whether the competitor is Chinese or not,” he said, highlighting how technological breakthroughs often benefit multiple companies. While AI models may eventually require fewer computing resources for training, Zuckerberg noted that substantial server power is still essential when these models are actively running, as it contributes to better intelligence and service quality. “It will be costly for us to serve all these users because we cater to a massive audience,” said Zuckerberg, noting that Meta had 3.35 billion daily active users in the fourth quarter. Meta plans to continue learning from technologies like DeepSeek’s while advancing its own open-source Llama AI models to push the industry forward. “I still believe that significant investments in capital expenditures and infrastructure will be a long-term strategic advantage,” he said. “We may learn otherwise in the future, but for now, I strongly believe that developing this kind of infrastructure will be a key competitive edge.” Zuckerberg also suggested that DeepSeek’s success reinforces Meta’s commitment to open-source AI development. “A global open-source standard will emerge,” he said. “For national interests, it’s crucial that this standard is American.”