In the AI chip sector, a major rivalry aimed at "moving away from Nvidia" is unfolding. Social media giant Meta has recently reached a significant multi-year, multibillion-dollar agreement with Google to rent Google's self-developed Tensor Processing Units (TPUs) for developing its next-generation AI models.

This move directly challenges Nvidia's dominance in the AI chip market. For a long time, Nvidia has been the preferred supplier for Meta when training models. Just days ago, Meta even announced plans to purchase millions of GPUs from Nvidia and AMD. However, Meta's decision to rent Google's TPUs not only aims to ease computing power anxiety but also to explore alternatives to GPUs within its self-built data centers. It is reported that Meta is even considering purchasing TPUs directly starting next year.

Google's Strategy: Both a Customer and a Competitor

The logic behind this deal is quite subtle. Google Cloud executives have set a goal to capture about 10% of Nvidia's annual revenue (approximately $20 billion) by selling TPUs. To achieve this, Google not only collaborates with investment institutions to lease TPUs externally but also tries to attract big clients like OpenAI and Meta through differentiated competition.

Interestingly, due to strong demand for GPUs in the cloud, Google itself remains one of Nvidia's largest customers. It must spend heavily to purchase Nvidia's latest chips to maintain competitiveness in the cloud market while promoting its own TPUs to erode Nvidia's market share.

Market Chain Reaction: Pressuring Chip Prices Down

This "turf war" in the AI chip market is clearly a good thing for downstream developers. According to industry reports, because of the existence of alternatives like TPUs, OpenAI successfully reduced its procurement price by 30% during negotiations with Nvidia.

As giants like Meta begin to shift toward a diversified computing power strategy, Nvidia's dominant position is facing unprecedented pressure. This "arms race" for computing infrastructure is evolving from a simple capacity competition into a comprehensive battle of architecture and ecosystem.