Amazon and the top artificial intelligence startup Anthropic have reached a milestone in their collaboration on the 20th local time. The two companies announced they will sign a long-term agreement lasting ten years, through funding and resource exchanges of hundreds of billions of dollars, to jointly consolidate their leading positions in the field of generative AI.
According to the latest agreement, Anthropic has committed to purchasing computing resources from Amazon Web Services (AWS) worth over $100 billion over the next decade. This staggering amount not only sets a new record for orders in the cloud computing industry, but also marks a deep integration between the two companies at the infrastructure level.

As part of the resource exchange and support, Amazon plans to increase its investment in Anthropic to a maximum of $2.5 billion. Previously, Amazon had invested $800 million, and in this investment plan, $500 million will be immediately available, while the remaining $2 billion will be released gradually based on specific business milestones that Anthropic achieves in the future.
In addition to the large-scale transactions of funds and computing power, the cooperation between the two companies has extended into the core R&D of semiconductor chips. It is reported that Anthropic will deeply participate in the R&D work of Amazon's custom chip team Annapurna Labs, jointly defining and optimizing future generations of Trainium series AI chips. In return, AWS will open up multi-generation Trainium computing clusters with a total power level of 5GW to Anthropic, providing a solid hardware foundation for the continuous iteration of its large models.
This deep alliance is seen as a typical example of a tech giant and a star large model company "complementing each other": Amazon secures its largest customer for cloud services and a core chip testing ground, while Anthropic gains almost unlimited funds and underlying computing power, which can support its continued high-intensity output in the intense global AI competition.
