Anthropic is relying more heavily on NVIDIA GPUs than competing AI processors, with newly disclosed figures showing 4.7 gigawatts of contracted computing capacity based on NVIDIA hardware.
The AI company has also secured 4.5 GW of Google's Tensor Processing Units (TPUs), alongside additional capacity from Amazon Trainium and AMD GPUs.
The figures suggest that NVIDIA continues to attract substantial investment despite competing processors potentially offering lower operating and acquisition costs.
One important factor appears to be financing, as lenders reportedly place greater confidence in the long term resale value of NVIDIA GPUs compared with specialized AI chips.
Anthropic's Computing Capacity Includes NVIDIA, Google, Amazon, and AMD
According to figures attributed to Anthropic's Form S-1 filing, the company's contracted computing capacity is distributed across four major hardware platforms.
| AI hardware supplier | Contracted capacity |
|---|---|
| NVIDIA GPUs | 4.7 GW |
| Google TPUs | 4.5 GW |
| Amazon Trainium | 1.2 GW |
| AMD GPUs | 0.8 GW |
| Total | 11.2 GW |
NVIDIA accounts for approximately 42% of the reported capacity, narrowly exceeding Google's share of around 40%.
These figures represent contracted infrastructure capacity rather than the number of processors deployed or their computing performance.
The distribution also indicates that Anthropic is avoiding dependence on a single hardware supplier.
Broadcom Reportedly Secures $63 Billion Financing Deal for Google TPUs
The financing challenges associated with specialized AI processors became apparent during negotiations involving Broadcom and Anthropic.
Under the reported arrangement, a special purpose vehicle would borrow $40 billion from a banking consortium.
Additional financing from Blackstone would increase the total to approximately $63 billion.
The funds would support purchases of TPUs developed by Google and Broadcom, which would subsequently be leased to Anthropic.
However, negotiations reportedly encountered difficulties as lenders became more cautious about financing specialized computing infrastructure.
Broadcom's CEO allegedly intervened by offering higher interest payments and additional financial guarantees, helping secure the agreement.
NVIDIA GPU Resale Value Could Influence AI Infrastructure Financing
NVIDIA GPUs offer broad compatibility across AI training, inference, and other computing workloads.
Their established software ecosystem and wider customer base may also make them easier to resell or redeploy.
By comparison, specialized processors such as TPUs can introduce additional uncertainty when lenders estimate their future value.
This distinction could influence financing costs even when alternative hardware is cheaper.
Anthropic's reported infrastructure commitments demonstrate that AI hardware procurement depends on several considerations beyond processor pricing.
Availability, financing terms, software compatibility, and expected long term value can all influence purchasing decisions.
NVIDIA's position in Anthropic's contracted capacity suggests these financial considerations remain significant as AI infrastructure investment continues expanding.
Discussion (0)
Be the first to comment.