Google has been promoting its Tensor Processing Unit (TPU) chips to external AI infrastructure providers, but key neocloud buyers have not shown strong adoption. Multiple industry comments indicate that the open market remains overwhelmingly dominated by Nvidia GPUs.
What industry participants are saying
- Nebius reports that demand is still 99% for Nvidia GPUs.
- CoreWeave noted that capacity goes where customers want it, implying buyer preferences determine which hardware is deployed.
- Lambda’s CFO said the company “bleeds green,” a remark signaling that current demand and business realities heavily favor Nvidia.
These statements point to limited traction for TPUs outside Google’s own operations.
Why Google’s promotion hasn’t changed the market
While TPUs may be effective within Google’s infrastructure, the broader open market operates under different constraints: the entrenched CUDA ecosystem, existing Nvidia cluster deployments, customer trust, and resale logic. According to reporting, the only significant external purchaser mentioned is Anthropic, and that relationship has been connected in coverage to a reported roughly $40 billion Google investment commitment.
Conclusion
Based on current signals, TPUs do not yet appear to be a market winner in open AI infrastructure. Instead, it looks like hardware Google is investing in and supporting while customers with demonstrable demand continue to prefer Nvidia GPUs.
