Recently, there has been a big stir in the chip industry.

The Information reports that Google has begun to actively reach out to major companies like Meta, hoping to have them try out their TPU chips. This is not just a simple business promotion—Google's goal is clear: to replace the currently used NVIDIA graphics cards in data centers.

The logic behind this matter is actually quite straightforward. Currently, AI training is basically monopolized by NVIDIA's GPUs, with both price and supply in their control. As one of the earliest companies to work on AI chips, Google has mainly used TPUs for its own purposes over the years, and now it suddenly wants to sell them to the public, which is an interesting timing choice.

If giants like Meta really start using TPU, the impact could be significant. First of all, in terms of computing power costs, having an additional option will generally drive prices down; secondly, although NVIDIA's CUDA ecosystem is deeply entrenched, if large clients begin to diversify, it will somewhat shake the foundation.

This matter is worth paying attention to in the realm of cryptocurrency. Projects like RNDR and Akash, which focus on decentralized computing power, have always been telling the story of breaking the monopoly of centralized computing power. Now, even traditional tech giants are starting to 'de-NVIDIA' internally, and the narrative logic of decentralized computing networks may need to be reassessed.

From another perspective, the competition for AI infrastructure has completely extended from the software layer to the hardware. This battle for computing power has just begun, and we will see how it evolves in the future.