Anthropic in Talks with Microsoft to Use Microsoft's AI Chips as New Computing Power Source
This move will further diversify Anthropic's computing power supply chain beyond NVIDIA GPUs, Google TPUs, and Amazon's self-developed chips.
The negotiations are currently in the early stages, aimed at reducing dependence on a single chip supplier and optimizing long-term cost structures.
Source: Public Information
ABAB AI Insight
Anthropic has established a deep cloud collaboration with Microsoft since 2024. This discussion about using Microsoft's self-developed AI chips continues its strategic shift from reliance on a single NVIDIA supplier to a multi-supplier strategy, having previously made large-scale purchases of Google TPUs and deepened cooperation with Amazon.
In terms of capital strategy, Anthropic will shift its training budget and Microsoft Azure resources towards testing self-developed chips. Microsoft, in turn, will utilize idle capacity and provide customized optimization services through its chip projects. Both parties are motivated to diversify supply chain risks and lower the cost per token, while also providing early large customer validation for Microsoft's chip business.
Similar to the long-term negotiations between OpenAI and Microsoft's self-developed chips, as well as Meta and Google's strategies for building TPU/Aquila clusters, leading AI labs are currently transitioning from NVIDIA's monopoly to a diversified chip combination. Early adopters are securing better bargaining power and customization flexibility through multi-supplier lock-in.
Essentially, this represents capital concentration: adopting Microsoft's self-developed chips will shift pricing power from NVIDIA's single supply to a competitive ecosystem of multiple vendors. The mechanism is that the sustained high demand for computing power in AI training forces labs to diversify supplier risks. Microsoft's chips, through deep integration with Anthropic, create customized advantages, accelerating the industry's evolution from hardware monopoly to an open supply chain.
ABAB News · Cognitive Law
The more expensive the computing power, the more a single supplier becomes a strategic risk.
Large model labs first discuss chips, then models; whoever controls the supply chain controls the cost advantage.
Diversification is not about dispersion, but about pulling pricing power back into the hands of buyers from a seller's market.