You are currently viewing 2024: Meta plans $37bn for AI data centers, digital infrastructure, will deploy own chips

2024: Meta plans $37bn for AI data centers, digital infrastructure, will deploy own chips

Social media giant Meta has announced an increased capital allocation of up to $37 billion for new digital infrastructure in the current year, marking a $2 billion boost from its initial estimate. The decision comes on the heels of Meta’s robust financial performance in Q4 2023, with revenues hitting $40 billion, a notable 25% YoY increase, bringing the company’s total earnings for 2023 to $134.9 billion, up by 16% compared to the previous year.

Susan Li, Meta’s CFO, provided insights into the expanded capital expenditure (CapEx), emphasizing the necessity of ongoing investment in hardware critical for training and running artificial intelligence (AI) systems. The increased spending, projected between $30 billion and $37 billion over the next 12 months, will primarily focus on server investments, covering both AI and non-AI hardware, along with data centers.

Li explained, “Our updated outlook reflects our evolving understanding of our AI capacity demands as we anticipate what we may need for the next generations of foundational research and product development.”

She highlighted Meta’s efforts in constructing new data centers with updated AI-centric architecture, incorporating advancements in cooling and power management systems. The company claims to have achieved a 31% reduction in construction costs and halved construction times.

While not providing guidance beyond 2024, Li emphasized Meta’s commitment to ambitious long-term AI research and product development, hinting that infrastructure investments could extend beyond the current year. She acknowledged the dynamic nature of the planning process, considering factors such as AI services demand and the availability of critical components like GPUs, stressing Meta’s vigilance regarding the supply chain.

Meta’s data center strategy includes redesigning for the AI era, with new facilities in Kuna, Idaho, and Temple, Texas, featuring innovative cooling and power management solutions. Additionally, Meta is exploring in-house silicon to incorporate an updated iteration of its custom AI-focused chips into data centers, reducing dependence on Nvidia’s GPUs.

During discussions with analysts, CEO Mark Zuckerberg emphasized Meta’s commitment to investing in AI and expanding data center capabilities to meet escalating compute requirements. Zuckerberg acknowledged the challenge of predicting compute needs for training AI models and running AI inference engines, highlighting the trend of large language models requiring approximately ten times the compute resources each year.

Zuckerberg articulated Meta’s ambition to develop premier AI products and services, envisioning a future where users benefit from sophisticated AI assistants, creators engage with AI-driven communities, businesses leverage AI for customer interactions and support, and developers access cutting-edge open-source models for building applications.

In an earlier interview, Zuckerberg outlined Meta’s plans to operate compute power equivalent to “600,000 Nvidia H100s” by the end of 2024, indicating the company’s focus on surpassing any other individual company in terms of computational capabilities.

Asides its AI data centers, other priority areas for Meta include its connectivity projects across the world. Last year, it invested $1 billion in 2Africa, a 45,000km subsea project connecting 46 cable landing stations in 33 countries in Africa, Asia and Europe.