Yuan CEO Mark Zuckerberg announces Overhaul Meta’s AI efforts in a memorandum of staff on Monday include a large number of employees at rival AI companies.
The restructuring comes after Meta’s efforts to stay at the forefront of AI in recent months. Zuckerberg has fixed the future of Meta to a large extent in its rapidly growing technology, saying the company will spend between $64 billion and $72 billion to build data centers to handle the AI workload this year. This is only $28 billion in annual capital expenditure in 2023.
Zuckerberg’s announcement marks a major strategic shift and active investment in AI, with CEOs reportedly investing billions of dollars to secure key talents for AI talent. Here are the key points of the basic principles behind Zuckerberg’s memo and meta-action:
- Creation of Metas Super Intelligent Laboratory (MSL):
Meta is integrating all its efforts involved in building large AI models, including its work team, working on Llama models, product teams, basic AI research (FAIR) teams, and a completely new unit focused on developing the next generation of cutting-reduction AI (under new departments). MSL is co-led by former CEO and co-founder Alexandr Wang, who will become Meta’s “Chief AI Officer”, former Github CEO and AI investor Nat Friedman. - Strategic Objectives:
Zuckerberg said in the memo that the clear purpose is to build “a personal superintelligence for everyone.” Superintelligence will be AI that can surpass humans in most cognitive tasks. - Positive talent acquisition:
Yuan is poaching top AI talents with competitors like Openai. Googleand humans. To this end, according to comments from Openai CEO Sam Altman, he has provided unprecedented signing bonuses, reaching $100 million. Zuckerberg announced in a memo that 11 highly respected AI researchers have been recruited from these other AI labs. Meta also invested $14.3 billion in AI as part of the deal, bringing Wang to Meta and reportedly investing billions of dollars into Friedman’s AI-focused venture capital firm to ensure he moved to Meta. - Failed acquisition attempt:
The hiring hype comes after an attempt to refuse to buy critical AI startups, including former OpenAI chief scientist Ilya Sutskever’s company’s security super-intelligence and confusing AI. - Llama struggle and competitive pressure:
Meta’s latest family of Llama AI models is called Llama 4 and performs poorly. The company faces criticism that it has published misleading benchmark numbers for the Llama 4, aiming to make the model look more competitive than it actually is. The release of Llama 4 Behemoth, the largest (according to Meta, the most powerful model), has been repeatedly postponed. Meta has not yet made its debut with a model with “reasoning” capabilities, but has lost competition for competitors like Openai, Anthropic, Google, DeepSeek and Alibaba’s Qwen. This led to an internal debate on Meta’s AI direction and increased the urgency to revitalize its AI portfolio. - Retention and Reputation Challenges:
Meta suffered losses from major camel researchers losing competitors, further increasing the need to attract and retain world-class AI talents. The company also conducts ongoing legal and ethical review of data practices, - Huge capital commitment:
Meta is investing tens of billions of dollars in infrastructure, data centers and custom hardware to try to ensure leadership in the AI era.
Next is the full text, such as Zuckerberg’s memo, which is from wealth reporter Salon Goldman:
As the speed of advancement of artificial intelligence accelerates, the developing super intelligence is seeing. I believe this will be the beginning of a new era of humanity and I am completely committed to the efforts to lead the Meta. Today, I want to share some details about how we organize our AI efforts to build our vision: everyone’s personal super intelligence.
We will call our overall organization Meta Super Smart Lab (MSL). This includes all our foundations, product and equity teams, and new labs dedicated to developing next-generation models.
Alexandr Wang joined Meta as our Chief AI Officer and lead MSL. Alex worked with me for a few years and I think he is the most impressive founder of his generation. He has a clear understanding of the historical importance of superintelligence and, as co-founder and CEO, he built ScaleAi in a rapidly growing company involving nearly all leading models throughout the industry.
Nat Friedman also joined Meta in collaboration with Alex to lead MSL, leading our work in AI products and applied research. NAT will work with Connor to define his role. He is Microsoftrecently runs one of the leading AI investment companies. NAT served on our Meta Advisory Group last year, so he already has a good understanding of our roadmap and what we need to do.
We also have several strong new team members joining today or joining in the past few weeks I’m also happy to share:
- Trapit Bansal – Created RL on the idea chain and co-creator of Openai’s O series model.
- Shuchao Bi – Co-creator of GPT-4O Voice Mode and O4-Mini. Previously trained after leadership of OpenAI.
- Huiwen Chang – Co-creator of GPT-4O image generation and invented Maskgit and Muse text-to-image architecture on Google Research
- Ji Lin – Helps build O3/O4-Mini, GPT-4O, GPT-4.1, GPT-4.5, 4o-Imagegen and operator inference stacks.
- Joel Pobar – Inferences of anthropomorphism. Previously, META in HHVM, HACK, FLOW, REDEX, performance tools and machine learning lasted for 11 years.
- Jack Ray – Gemini’s former technical leader and Gemini 2.5 reasoning. LED Gopher and Chinchilla Early LLM efforts Deep state.
- Hongyu Ren – Co-creators of GPT-4O, 4o-Mini, O1-Mini, O3-Mini, O3 and O4-Mini. Previously led a group that trained in Openai.
- Johan Schalkwyk – Former Google Fellow, an early contributor to Sesame, and a technology leader at Maya.
- Wear the sun – Google DeepMind’s Gemini training, coding and reasoning. The last two generations of Waymo’s perception model were created before.
- Jiahui Yu – Co-creators of O3, O4-Mini, GPT-4.1 and GPT-4O. Previously led Openai’s Perception Team and led multimodal in Gemini.
- Shengjia Zhao – Chatgpt co-creator of GPT-4, all mini models, 4.1 and O3. Previously synthesised data in OpenAI.
I’m excited about our plan to make progress on Llama 4.1 and 4.2. These model power meta AI, over 1 billion actives per month in our applications and increasingly meta-agent use, can help improve our products and technologies. We are committed to continuing to build these models.
Meanwhile, we will begin researching the next generation model to reach the border around next year. Over the past few months, I’ve spent months meeting at Meta, other AI labs and promising startups to bring the founding team together to achieve this small, talent-intensive effort. We are still forming this group and we will also ask a few people from AI Org to join the lab.
Meta is providing the world with a unique position of super intelligent. We have a strong business that supports larger computing than smaller labs. We have deeper construction and growing products that can attract billions of people. We are pioneering and leading the AI glasses and wearables category, which are growing very fast. Our company structure enables us to act with greater faith and boldness. I’m optimistic that this new approach to talent and parallel approaches to model development will allow us to provide everyone with a commitment to personal superintelligence.
We will have more great people on every level joining this effort in the coming weeks, so stay tuned. I’m so glad to sneak in and go to work.
***
Disclaimer: For the sake of this story, Fortune uses generated AI to aid in the first draft. The editor verified the accuracy of the information before publishing.
