Less than a week ago, Meta confirmed the acquisition of a stake valued at $15 billion in Scale AI and the signing of its founder, Alexandr Wang, to run a new superintelligence laboratory.
What until then was a rumor has materialized into one of the most ambitious operations in the history of AI, revealing Mark Zuckerberg’s determination to turn the company into a benchmark in the development of the Meta AI environment.
We are going to analyze this multi-million dollar move in detail, along with the rest of the investments, products and challenges that complete Meta’s strategy.
Strategic vision of Goal and future architecture
Meta conceives AI as a lever for innovation to enhance all its platforms (from social networks to advertising) and explore new markets such as wearable devices with integrated AI. Under this vision, it has designed a three-pronged strategy:
- Long-term research: creation of a superintelligence laboratory led by Alexandr Wang, with the aim of advancing towards AGI.
- Practical development: integration of AI tools into current products, such as programming and video editing assistants.
- Infrastructure and data: massive deployment of data centers and assurance of high-quality data flows.
The superintelligence laboratory and the “Fantastic 50” team
A few days ago Meta announced the launch of a new AI laboratory focused on superintelligence. At the helm is Alexandr Wang, co-founder and CEO of Scale AI; Around him, a team of up to 50 elite researchers, the so-called “Fantastic 50”, recruited with million-dollar hiring packages.
These include names like Jack Rae (ex-DeepMind) and Johan Schalkwyk (ex-Sesame AI). This team will seek to publish cutting-edge research and train large-scale models that compete with reference labs in AGI.
Investments and strategic alliances
To give muscle to its ambitions, Meta has mobilized financial resources and key alliances.
Scale AI participation and capital expenditure
- Investment in Scale AI: Meta subscribed to an issue of shares in the startup worth $15 billion, acquiring nearly 49% of its capital. The goal is to ensure a stable supply of quality labeled data to train your AI models.
- Infrastructure spending: In 2025, Meta plans to invest between $60 billion and $65 billion in capital expenditure (CapEx), primarily in new data centers and specialized hardware (GPUs and ASIC servers) designed for intensive AI workloads.
Collaborations with other model manufacturers
Despite developing its own models internally (Metamate), Meta recognizes the advantage of combining its technologies with external solutions.
For this reason, it integrated the Devmate assistant, launched last March, which combines internal and third-party models (such as Anthropic’s Claude 2) to accelerate the work of its engineers, reducing the time spent on routine tasks by 50%.
Recent products and innovations
Meta is already deploying several initiatives to show concrete progress and improve its position in the market.
Llama 4 models and their evolution
At the beginning of April 2025, Meta presented Llama 4, a multimodal family with several versions:
- Scout (17 B parameters): optimized for conversational tasks.
- Maverick (17 B with 128 experts): multipurpose specialization.
- Behemoth (approx. 288 B parameters): still in training phase and scheduled for publication later this year.
Although Llama 4 has improved performance compared to previous generations, some benchmarks indicate that there is still a gap compared to competitors in complex reasoning and context understanding.
Video editing with AI and Meta AI App
Meta launched an AI-powered video editing feature within the Meta AI app and on third-party platforms. This tool allows users to modify costumes, scenery and lighting with simple commands.
It is currently available in the US and 15 other countries, with a roadmap to incorporate natural language instructions and advanced styles before the end of the year.
In parallel, it is preparing an independent Meta AI application to compete directly with widely spread chatbots. Its goal is to offer a conversational assistant with multimodal capability (text, voice, image) and advanced personalization.
Devices and accessibility: Ray-Ban Meta and WhatsAI
In the hardware field, Meta renewed its Ray-Ban Meta glasses incorporating AI and multimodal recognition functions in April 2024.
In addition, it is working on WhatsAI, a prototype announced in May 2025 that uses generative AI to offer real-time descriptions to people with visual disabilities, and can be integrated into glasses or mobile applications.
Video generation: Movie Gen
Another product in development is Movie Gen, a model capable of generating video and audio sequences from scripts and styles defined by the user. Introduced in October 2024, it is aimed at content and advertising creators.
Challenges for Meta and future perspectives
Talent retention problem
Although the recruitment of experts has been impressive, Meta lost 4.3% of its AI workforce in 2024, with signings such as Mike Krieger (co-founder of Instagram) leaving for Anthropic. Talent rotation poses the challenge of maintaining cohesion and culture in long-term projects.
Fierce competition and speed of innovation
In the short term, OpenAI is preparing GPT-5, while Google accelerates acquisitions such as Character.AI. The perception that Meta’s models are one step behind could penalize the adoption of its products if it does not accelerate launches and improve benchmarks.
Meta’s AI strategy combines monumental investments, partnerships with data startups, hiring top talent, and aggressive product and hardware deployment.
Although he faces retention challenges and needs to close the technical gap on rivals, his recent moves (from Llama 4 to Ray-Ban Meta) show he means business.
The coming months will be decisive to see if Meta is able to transform its ambitious plans into tangible results and regain a leadership position in the era of artificial intelligence.
This post is also available in: