Zuckerberg Responds to DeepSeek AI, Commits Billions to AI Development

Market Reaction to AI Model Speculation

U.S. financial markets experienced a downturn on Monday, triggered by concerns regarding the potential impact of DeepSeek’s AI models on the demand for graphics processing units (GPUs). This speculation led to a nearly 20% decrease in Nvidia’s stock value.

Meta's Continued AI Investment

Despite these market anxieties, Meta remains firmly committed to its substantial investments in artificial intelligence. CEO Mark Zuckerberg affirmed the company’s intention to invest “very heavily” in AI technology over the coming years.

Zuckerberg indicated that this investment could reach “hundreds of billions of dollars” during Meta’s first-quarter earnings conference call held on Wednesday.

Capital Expenditure Plans

Previously, Zuckerberg announced that Meta anticipates exceeding $60 billion in capital expenditures for 2025. A significant portion of this funding will be allocated to the development of data centers.

Addressing DeepSeek's Competitive Impact

Responding to inquiries about DeepSeek’s potential influence on Meta’s AI spending, Zuckerberg emphasized that continued investment in AI infrastructure will provide a “strategic advantage” for the company.

Meta acknowledges DeepSeek as a new competitor and is actively analyzing its advancements. However, Zuckerberg stated it is “way too early” to determine whether chip demand will plateau, as these components remain essential for inference tasks, particularly given Meta’s vast user base.

Infrastructure as a Key Advantage

“The capacity to develop this type of infrastructure will likely be a significant benefit for both the quality of our services and our ability to reach the desired scale,” Zuckerberg explained.

Llama 4: Meta's Next Generation Model

Meta’s objective with its forthcoming model, Llama 4, is to establish it as the most competitive AI model globally, even when compared to proprietary models such as ChatGPT.

Zuckerberg anticipates that Llama 4 will incorporate both agentic and multimodal capabilities, mirroring developments undertaken by OpenAI and Anthropic.

“With Llama 3, our aim was to achieve parity with closed models in the open-source domain,” he stated. “For Llama 4, our goal is to take the lead.”

Stay informed with TechCrunch’s AI newsletter! Subscribe here to receive it weekly in your inbox every Wednesday.