Source: OpenAI, a major Nvidia customer, plans to buy Google's TPUs; a job listing shows it seeks an engineer to "lead the optimization" of its models on TPUs

The cost to rent AI compute power falls as supply outpaces demand.

Anthropic is diversifying its AI hardware by utilizing chips from multiple vendors, including Amazon's Trainium, Google's TPUs, and AMD GPUs, alongside Nvidia's GPUs. This strategy, detailed in its S-1 filing, aims to create a fungible compute pool, allowing Anthropic to optimize workloads across different hardware based on productivity. The company stated this flexibility took years and significant investment to build. Analysts suggest this approach could diminish Nvidia's market dominance and pricing power over time, as major AI companies gain more negotiating leverage and reduce dependency on a single supplier. While switching between chip architectures is complex, particularly for AI model training, the largest AI firms possess the resources to implement such multi-chip strategies. Anthropic continues to be a substantial customer of Nvidia, spending billions on their compute hardware.

Anthropic's explicit articulation of a multi-chip strategy, now presented as a core advantage to investors, signals a growing trend among major AI players to mitigate risks associated with single-vendor dependency. This approach leverages software to dynamically allocate AI workloads across diverse hardware, including Google TPUs and Amazon Trainium, thereby challenging Nvidia's entrenched position. The ability to shift compute demands across platforms not only bolsters negotiating leverage but also provides a buffer against potential supply chain disruptions and price hikes, fundamentally altering the economics of AI infrastructure.

While the technical hurdles of cross-platform AI model training remain substantial, the strategic implications for Nvidia are significant as larger entities explore hardware fungibility. This diversification strategy, though resource-intensive, represents a gradual shift towards AI compute becoming a more commoditized service, a scenario where Nvidia's current pricing power could be eroded. The long-term consequence for the market is a potential recalibration of AI hardware procurement and a more competitive landscape where specialized hardware gains traction alongside general-purpose GPUs.