According to The Wall Street Journal, Google is replicating NVIDIA's strategy by transforming its internally-used TPU (Tensor Processing Unit) into a commercial AI chip business for external sales, seeking to capture a share of the accelerator chip market.

This represents another milestone in hyperscale tech companies' commercialization of proprietary chips. For years, NVIDIA has leveraged its GPU and CUDA ecosystem to maintain a near-monopolistic position in the AI accelerator chip market, commanding substantial pricing power. Meanwhile, the external availability of proprietary chips such as Google's TPU and Amazon's Trainium is simultaneously eroding this monopoly from both demand and supply perspectives.

From a medium- to long-term perspective, this development is favorable for computing power costs: more chip options mean fiercer price competition and more abundant supply, expanding the procurement bargaining power of downstream computing providers. The current situation of chip scarcity and elevated prices may ease as alternative solutions mature.

However, the other side of the coin is this: if major customers begin adopting hyperscale tech companies' proprietary chips at scale, their demand structure for third-party computing power will also shift. Some customers may bring their own chips to rent your data center rather than lease pre-configured GPUs. Your business model needs to build in flexibility to accommodate this trend of decoupling computing power from chips.