In Silicon Valley, and in technology in general, being huge does not guarantee being prepared to win every race. You can have money, talent, data centers, billions of users and machinery capable of integrating anything new into products we use every day. Still, when the board changes, so does the question. Meta has been trying for years to demonstrate that it can not only distribute artificial intelligence at scale, but rather compete at the center of the conversation. The problem is, when we think about chatbots, it still doesn’t seem to be the first name that comes to mind.
14.3 billion dollars. Ea is the figure that Reuters put on the table for a very specific operation. On June 13, 2025, the agency reported that Meta would take a 49% stake in Scale AI for that amount, in a deal that valued the startup at about $29 billion. Scale itself spoke of a significant new investment from Meta, although it did not publish the exact amount of the investment. We are not talking, therefore, about everything that the company has allocated to AI, but rather about an identifiable bet within a much broader bill.
What Meta saw in Scale AI. It was surely not one of those companies that we had on our radar when we were talking about artificial intelligence. It didn’t have the public glare of ChatGPT or the showcase of Gemini, but it did occupy an important place in the machinery that makes it possible to train and evaluate models. His work revolves around data that allows training, evaluating and improving AI systems, including labeled or curated data for training.
The name behind the operation. Meta was not only betting on Scale AI, it was also incorporating Alexandr Wang into its new stage in artificial intelligence. The agency noted that the main driver of the move was to secure Scale’s founder to lead Meta’s superintelligence efforts. Scale itself confirmed that Wang would join Meta to work on its AI projects. So the investment should not be read solely as an entry into the capital of a data company, but as a way to accelerate leadership and talent.
Context. The investment came at a time when Meta needed to strengthen its position in the advanced AI race. It occurred in a context marked by the poor reception of Llama 4, its latest large family of open models, and by competitive pressure against companies such as Google, OpenAI and DeepSeek. It was not just a matter of having more resources or adding a new piece to the organizational chart. What was at stake was to regain momentum in a field where other names were marking a good part of the technical, business and public conversation.
The visible part. The most recognizable result of this new stage is Muse Spark, presented by Meta as the first model of a new family created by Meta Superintelligence Labs. The company assures that it already powers Meta AI in its app and on the web, and that it is being deployed in WhatsApp, Instagram, Facebook, Messenger and its AI glasses. Here, precisely, there is an important point: Meta does not need to convince the user to install another application from scratch: it already has the channels. But converting presence within their own platforms into public relevance within generative AI is another battle.
The limit. Just because the model is within WhatsApp or Instagram does not mean that people use it for many tasks. Muse Spark does not seem to be occupying the place that the GPT or Gemini models do, to name a few examples. Despite this, according to Reuters, Muse Spark has obtained good results in languages and visual compression, although it has fallen behind in coding and abstract reasoning. Meta has managed to be present, but has not yet demonstrated that this presence is enough to change habits.
Strategic turn. Muse Spark does not follow the path that had given Llama so much visibility: The Wall Street Journal described it as a closed model. The company itself talks about an API in private preview for selected partners, not open and general access for any developer. That is to say, Meta has put a new model into circulation, but it has done so in a more controlled way, more integrated into its products and less open than the strategy with which it had tried to differentiate itself in AI.
The crack. Meta can integrate AI into gigantic products, but the generative AI race is also played in another field: that of the names that the user recognizes when they need a chatbot. And there Zuckerberg’s company does not seem to occupy the same place as ChatGPT, Gemini, Claude or Grok. The economic doubt has not disappeared either. And, a no small detail, advertising continues to be the engine of Meta’s income.
Images | Mark Zuckerberg
In | There is a company proving that AI can be the perfect interviewer for companies. His name is Orbio and he is from Madrid
