Meta, a social media giant that focuses on open-source big models, has gained a lot from the global AI (artificial intelligence) boom, driving up Zuckerberg's value.
Recently, according to the Bloomberg Billionaires Index, Meta CEO Mark Zuckerberg surpassed Tesla CEO Elon Musk in wealth and entered the top three of Bloomberg's global billionaires list for the first time since November 16, 2020. As of April 9th, Zuckerberg's net worth was $184 billion, only $2 billion behind Musk, and an increase of $56.1 billion so far this year, making him the fourth wealthiest person in the world.
On April 9th, the stock price of Meta (Nasdaq: META) closed at $519.25 per share, down 1.53%. Due to strong financial performance and the attractiveness of AI prospects, Meta's stock price has surged by about 46% this year, making it the fifth best performing company among the S&P 500 index components. Last week, an analyst at investment bank Jefferies stated in a report, "Meta has countless advantages."
Through its unique AI layout, Meta has taken its own path in the increasingly fierce AI competition.
Rush to purchase Nvidia GPUs while deploying custom chips
Although Meta's investment in the metaverse in the past few years did not yield the expected results, Zuckerberg remained steadfast in choosing to bet entirely on AI after the AI craze began to rise. By investing heavily in AI infrastructure, Meta is able to develop and deploy advanced AI services on its own platform, accelerate its advertising business, and deter competitors.
On February 1st of this year, Meta announced its fourth quarter financial performance for the fiscal year ended December 31st, 2023, with revenue of $40.111 billion, a 25% increase from the same period last year and higher than market expectations of $38.9 billion. In the post earnings conference call, Meta announced that it will continue to increase its investment in AI, and expects the company's capital expenditure for the full year of 2024 to be between $30 billion and $37 billion, an increase from the guidance given three months ago of $30 billion to $35 billion.
Meta expects that investment in servers will drive this growth, as well as the company accelerating the construction of new data center architecture sites and deploying new self-developed custom chips within the data center. Zuckerberg also emphasized during the conference call that training and running AI models will "require more computing resources" in the future, and stated that the company "aims to win in AI and will continue to make large-scale investments in this field.".
On January 18th, two weeks before the financial report was released, Zuckerberg high-profile announced that Meta was planning to build its own AGI (General Artificial Intelligence) and promised to open up this technology to the public. For this reason, Meta plans to acquire approximately 350000 H100 GPUs from Nvidia by the end of this year. Including other GPUs, the company's total computing power will be close to the computing power that 600000 H100 can provide. The selling price of the star chip H100 is set between $25000 and $30000, and even if calculated at $25000, Meta will still pay approximately $8.75 billion for the 350000 H100.
In addition to rushing to purchase "AI armaments", Meta also plans to deploy customized chips in its data centers to reduce reliance on NVIDIA chips. In February of this year, according to foreign media reports, Meta plans to launch an AI chip focused on model inference, known internally as Artemis, to further accelerate the company's expansion in the AI field. A Meta spokesperson confirmed the authenticity of the plan, stating that the chip will work in conjunction with hundreds of thousands of GPUs purchased by Meta.
Previously, as early as May last year, Meta stated that it had started building and improving custom chip MTIA to accelerate the development of generative AI models and optimize the inference process. It is reported that the chip will be officially launched in 2025.
Become a representative of open source and join the AI talent competition
In terms of AI software, Meta is adopting a "steady flow" development strategy. Meta spends millions of dollars every quarter on the development of its well-known open-source model, Llama, and obtains massive AI training data from public posts on its social media platforms, Facebook and Instagram.
Last July, Meta released Llama 2, which not only improved performance but also changed the "research only available" condition in the open source protocol to "free for commercial use", directly changing the landscape of the large model industry on its own. According to foreign media reports, Meta may officially release Llama 3 in July this year, and it is highly likely that it will continue to be open source. However, Llama 3 will need to face challenges from the strong performing Google Gemma and a series of emerging models such as Mistral.
In addition, as an industry representative of open source, Meta has begun to consciously build an upstream and downstream ecosystem for Llama 2. Last December, in order to counter the closed source camp represented by OpenAI and Google, Meta teamed up with 57 global technology companies and research institutions such as Oracle, Intel, AMD, and IBM to form the AI Alliance. According to Meta, the main task of the AI alliance is to "support open innovation" and "open science.".
Of course, talent is also an indispensable part of Meta's AI layout. In March of this year, foreign media reported that Meta is taking a series of recruitment measures, including hiring without interviews, in order to maintain its position in the AI field and better attract AI talents. Zuckerberg will also personally send emails to researchers from competitors such as Google DeepMind, inviting them to join Meta. It is reported that Meta's generative AI department will recruit approximately 100 people before the release of Llama 3, with a total department size of around 600 people.
An analysis suggests that in the coming years, investment in AI will directly affect Meta's various revenue streams. As a social media giant with advertising as its core business, in addition to using AI to assist advertisers in creation and promotion, Meta can also use AI to support the chat functions of its major social media platforms, and attempt to integrate AI with the company's once vigorously developed AR (augmented reality) and VR (virtual reality) products to create new revenue streams in the long term.
Compared to other giants such as Google, Microsoft, and Amazon, Meta is not only investing money in researching AI products, but also trying to take a different path towards entertainment. At the Meta Connect Developer Conference in September last year, Meta introduced a series of consumer oriented generative AI products. Among them, the most interesting ones are 28 chatbots made based on well-known figures. Users will be able to have conversations with these celebrity chatbots through text and watch their avatars to make real-time responses based on the chat content. Zuckerberg pointed out at the meeting, "People don't want to interact with a single super intelligent AI, but rather want to interact with various different AIs."