Looking at the capital flow of tech giants in 2025, we can understand the true gameplay of the current AI competition.

A core logic runs through it: whoever controls high-quality data, whoever holds the future of large models.

The most straightforward example is the $14.3 billion investment—one tech company used cash to acquire a 49% stake and also recruited top AI talent. This investment was not primarily about algorithms or server scale, but about data. High-quality, multimodal, real-world data.

By 2025, large models have entered a saturation competition phase. Differences at the algorithm level are minimal, and training frameworks are becoming increasingly transparent. The real barrier is who can continuously acquire the most premium training datasets—that determines the ceiling of the model.

Looking at the acquisition of Scale AI, the approach of big tech companies has shifted from "self-sufficiency" to "buying out high-quality supply chains." Instead of spending time and effort building data annotation systems from scratch, it’s more efficient to acquire the industry’s strongest data infrastructure directly. This approach is both fast and ruthless.

In the AI competition of 2025, the final showdown is still about capital intensity—who has more money, who dares to spend, and who spends accurately, will be able to run faster.