In a revealing statement during Meta’s second-quarter earnings call, CEO Mark Zuckerberg has shed light on the exponential growth in computing power required for the company’s ambitious AI projects. Zuckerberg announced that training the next iteration of Meta’s large language model, Llama 4, will demand a staggering ten times more computing power than its predecessor, Llama 3.
This revelation comes on the heels of Meta’s recent release of Llama 3.1 405B, the company’s largest open-source model to date, boasting 405 billion parameters. The rapid scaling of these models underscores the intensifying race in AI development among tech giants.
“At this point, I’d rather risk building capacity before it is needed rather than too late, given the long lead times for spinning up new inference projects,” Zuckerberg stated, highlighting Meta’s proactive approach to infrastructure investment.
The financial implications of this AI-driven strategy are significant. Meta’s capital expenditures surged nearly 33% to $8.5 billion in Q2 2024, primarily attributed to investments in servers, data centers, and network infrastructure. CFO Susan Li indicated that these investments are likely to further increase capital expenditures in 2025.
Interestingly, while Meta is pouring resources into AI development, the company doesn’t anticipate its generative AI products to contribute significantly to revenue in the near term. However, Meta AI, the company’s consumer-facing chatbot, has found its largest market in India, leveraging the massive user base of WhatsApp in the country.
As Meta continues to push the boundaries of AI capabilities, the company faces the dual challenge of managing escalating costs and maintaining its competitive edge in the rapidly evolving AI landscape. The success of these investments could reshape Meta’s position in the tech industry and potentially redefine the future of social media and digital interactions.
