OpenAI, a Longtime Nvidia Customer, Plans to Buy Google’s AI Chips

The cost to rent AI compute from major cloud providers will likely fall.

AI lab Anthropic disclosed in its S-1 filing that it manages its AI compute needs as a unified pool, drawing from hardware supplied by Amazon, Google, Nvidia, and AMD. This multi-vendor strategy allows Anthropic to direct workloads to the most productive chips, reducing dependency on any single supplier. The company states this flexibility was developed over years with significant investment. This approach is highlighted as a key advantage for investors, offering greater negotiating leverage and mitigating supply chain risks. While the flexibility is challenging to implement, particularly for training AI models, it is seen as a significant advantage for large AI companies with the resources to adopt it. Smaller players are less likely to pursue such a strategy due to the investment required.

Anthropic's multi-chip strategy signals a growing trend among major AI companies to diversify their hardware suppliers, reducing their reliance on Nvidia. This diversification provides AI labs with increased negotiating power and resilience against supply chain disruptions, potentially eroding Nvidia's market dominance. The ability to treat compute as a fungible resource across different hardware platforms is a complex undertaking, requiring substantial investment and sophisticated software, which only the largest AI players can realistically achieve.

This development suggests a future where AI compute could become more commoditized, especially if supply eventually outstrips demand. Companies like Anthropic are actively building the infrastructure for such a future, pushing the boundaries of hardware flexibility. The long-term implication is a potential shift in pricing power from chip manufacturers to large AI consumers, a scenario that could significantly alter market dynamics and profitability for hardware providers.