Track Hyper | Meta's Q4 performance skyrockets: aggressive capital expenditures for the whole year

On January 29th, local time in the United States, social media giant Meta released its fourth quarter and full-year financial report for 2024. Meta Platforms, Inc. was registered in July 2004 in the state of Delaware, USA, and is the largest social networking site in the world.

In the fourth quarter, Meta achieved revenue of $48.385 billion, a year-on-year increase of 21%, exceeding expectations of $47.03 billion; net profit was $20.838 billion, a significant year-on-year increase of 49%; earnings per share (EPS) reached $8.02, significantly surpassing analysts' expectations of $6.73 by 19.2%, demonstrating Meta's strong operational execution and growth in key areas (advertising).

The financial report noted that Meta's focus on AI and advertising strategies has yielded results, driving significant growth in the fourth quarter.

For the full year of 2024, Meta achieved revenue of $164.501 billion, a year-on-year increase of 22%, and net profit of $62.36 billion, a year-on-year increase of 59%. Meta's capital expenditures for the fiscal year 2024 reached $39.23 billion, higher than the market expectation of $38.2 billion.

In the fourth quarter of 2024, Meta's Family of Apps series generated revenue of $47.302 billion, accounting for 99.82%; the Reality Labs division (responsible for producing Ray-Ban smart glasses and Quest VR headsets) generated revenue of $1.083 billion; the daily active users (DAP) of the AI chat assistant reached 3.35 billion, a year-on-year increase of 5%; monthly active users surged from 600 million to 700 million within a month.

The Family of Apps includes products such as Facebook, Instagram, Messenger, WhatsApp, and Threads.

These applications have a large user base, providing a vast platform and massive user data for digital advertising businesses. Meta utilizes this data for precise ad placements, offering advertisers efficient marketing channels, thereby achieving growth in advertising revenue.

Digital advertising is undeniably Meta's cash cow, contributing $46.78 billion in revenue this quarter, accounting for 97% of the total, with a growth of over 20%. This achievement is largely attributed to the improvement in advertising placement efficiency due to AI technology.

Meta CEO Mark Zuckerberg emphasized the company's focus on AI: "I expect this year to be a year where highly intelligent and personalized AI assistants reach over 1 billion people."

The "personalized AI assistant" mentioned by Zuckerberg involves Meta's AI initiative: the launch of the AI Engineering Agent. This AI agent's programming and problem-solving capabilities are comparable to those of an excellent mid-level human engineer.

Zuckerberg believes this could be "one of the most important innovations in history," presenting a huge market opportunity for the first company to successfully develop it.

Overall, Zuckerberg stated that over 3.3 billion people use at least one of Meta's applications every day. "We continue to make good progress in the future of artificial intelligence, glasses, and social media. I am excited to see these efforts expand further in 2025." In the earnings call, Mark Zuckerberg believes that 2025 will be a key year for the mass adoption of AI glasses. Zuckerberg also revealed for the first time the sales data for Ray-Ban smart glasses: over 1 million pairs sold in 2024.

Recently, there have been reports that Apple has once again abandoned its AI glasses project, but Zuckerberg believes that smart glasses are the ideal form of AI devices, as they can better integrate with user scenarios to provide personalized services. He also emphasized that 2025 will be a crucial development year for AI glasses as a new product category.

Although Ray-Ban smart glasses are expected to exceed 1 million units shipped in 2024, and Zuckerberg remains optimistic about the future potential of AI glasses, Meta's Reality Labs division is still facing significant losses: while it achieved $1.08 billion in revenue, the losses reached $4.97 billion, a year-on-year increase of 7%.

On January 20, DeepSeek released the DeepSeek R1, which, with its astonishingly low cost and robust performance, has garnered significant attention from Zuckerberg, who seems to have found a strong resonance with Liang Wenfeng from DeepSeek.

During the earnings call, when asked how large language models like DeepSeek would impact Meta's investment plans in AI infrastructure and computing power in the coming years, Zuckerberg candidly stated that the rise of DeepSeek validates the correctness of Meta's open-source approach.

Zuckerberg acknowledged that DeepSeek is worth learning from and hopes to implement some of DeepSeek's technologies in his own systems.

At the same time, Zuckerberg believes that a global open-source standard will inevitably emerge in the future, but the standard must be dominated by the United States. Meta's goal is to ensure that American companies, rather than Chinese competitors, lead the global open-source standards for AI. This is also why Meta maintains such an aggressive stance in AI infrastructure investment.

Meta expects to achieve revenue of $39.5 billion to $41.8 billion in the first quarter of 2025, a year-on-year growth of 8% to 15%; for the entire year of 2025, total expenditures are estimated to be around $114 billion to $119 billion, with capital expenditures of approximately $60 billion to $65 billion, focusing on supporting generative AI and core businesses.

In addition to the massive capital expenditure for AI infrastructure in 2025, Zuckerberg stated that in the coming years, Meta will invest hundreds of billions of dollars in AI infrastructure. "Meta will not reduce AI spending as a result; I still believe that investing heavily in capital expenditures and infrastructure development will become a strategic advantage in the long run."

Meta has been advancing the iteration of Llama, and Zuckerberg indicated that this year is likely to be the year when Llama and open-source become the most advanced and widely used AI models. Llama 4 has made significant progress in training, and Llama 4 mini has completed pre-training, with good performance in inference models