Google Follows Nvidia's Playbook to Expand TPU Chip Business Externally
Claire Weston
Google is converting its in-house AI chip, the TPU, into a commercial product for outside customers — bundling hardware, software tools, and developer support in a direct echo of Nvidia's platform model and a formal entry into GPU competition.
What is Google actually doing with TPU?
The TPU — Google's custom-designed AI chip — was built to power its own search, advertising, and cloud services internally.
Google is now repackaging it as a full AI platform: hardware, software tools, and developer support sold together to outside enterprises, not as a standalone chip.
This means → Google is copying Nvidia's core strategy: lock customers in through the ecosystem, not through chip specs alone.
Where is the $3.2 billion going?
The Wall Street Journal reported that Google provided $3.2 billion in financial guarantees for "Lake Mariner", an AI data-center project in western New York State.
The facility will lease computing power from thousands of Google TPU chips to AI startup Anthropic.
In plain terms = Google is not selling chips directly — it is using its cloud platform as the middleman. Customers rent TPU compute through Google Cloud; Google earns cloud-service revenue.
Is Google the only tech giant building its own chips?
Amazon's AI chief has reportedly begun talks with potential customers about selling its in-house AI chips externally.
Microsoft and Meta are also investing heavily in custom chip development to cut costs and secure compute supply.
This means → Nvidia's dominance in the AI data-center market faces structural pressure from multiple directions — not one or two rivals, but an entire customer base moving to reduce dependence simultaneously.
What is Google's real test?
The key question is singular: will outside customers pay for TPU's software ecosystem?
Nvidia's moat is not the chip itself — it is CUDA, Nvidia's programming toolkit that the vast majority of AI developers already use to write code. Switching costs are steep.
This reflects a competition not over hardware performance but over ecosystem stickiness: whether Google can convince developers that migrating is worth the effort.
Content is for reference only, not financial advice.