According to a media report, Meta Platforms is considering acquiring a 49% stake in Scale AI for $15 billion – a move that underscores the strategic shift in the AI arms race: In addition to computing power and data, human intelligence is now moving to the center of events.

While Meta's capital expenditure of up to $72 billion this year is primarily flowing into data centers and servers, another resource is becoming increasingly scarce: high-quality training data. OpenAI co-founder Ilya Sutskever has already warned of reaching a "peak data" – a saturation point comparable to the concept of "peak oil".

Scale AI has specialized in this exact bottleneck. The company cleans, categorizes, and labels data to make it usable for machine learning models. CEO Alexandr Wang emphasizes that freely available data sets are depleted and future data would be more difficult to process.

Particularly for Meta CEO Mark Zuckerberg, whose models like LLaMA rely on extensive datasets, this is becoming increasingly challenging. Privacy laws like those in Europe restrict the usability of publicly shared content. At the same time, Reddit is suing AI developer Anthropic for unauthorized data scraping – a signal of growing regulatory and legal risks.

One solution is called "Reinforcement Learning from Human Feedback" (RLHF).

The acquisition of Scale AI would not only secure Meta access to a key service provider but could also pave the way for a new "superintelligence" division—with Wang at the helm. Goal: to systematically integrate not only knowledge but also human thinking into models.

Despite a rumored revenue of $2 billion for 2024, the $15 billion deal appears ambitious at first glance. However, considering Meta's expected stock buybacks and dividend payments amounting to $44 billion, the price remains relative – especially when considering it concerns strategic supremacy in a trillion-dollar market.

The AI race has so far revolved around chips, electricity, and data. Now it is clear: Human neurons are also a scarce resource.