OpenAI is engaging in a deep collaboration with Amazon that could reshape the AI computing power landscape. According to The Information, OpenAI is negotiating to secure at least $10 billion in funding from Amazon and is considering incorporating Amazon's self-developed AI chip, Trainium, into its training infrastructure, aiming to diversify its computing power supply chain.
If the deal is finalized, this would be the most significant external customer for Amazon's Trainium chips so far. As an AI accelerator independently developed by AWS to break NVIDIA's monopoly, the Trainium series has primarily served internal large model training within Amazon. In early June, AWS just launched Trainium3, which achieved a breakthrough in performance: compared to the previous generation, its computing performance can be up to 4.4 times higher, energy efficiency is 4 times better, and memory bandwidth is nearly doubled.
More importantly, the UltraServer system built on Trainium3 has astonishing scalability—each rack can integrate 144 chips, and the entire cluster can support the deployment of up to 1 million Trainium3 chips, providing unprecedented scale of dedicated computing power for a single AI application. The overall system scale is 10 times that of the previous generation. This ultra-large-scale deployment capability perfectly aligns with the urgent needs of leading AI companies like OpenAI for massive, stable, and low-cost computing power.
Currently, OpenAI's training clusters heavily rely on NVIDIA GPUs, but as model sizes grow exponentially, computing costs and supply risks are becoming increasingly prominent. Introducing Amazon's chips is not only a technical backup but also a strategic hedge—by supporting a second supplier, OpenAI can enhance its bargaining power, ensure delivery security, and promote the evolution of the entire AI infrastructure ecosystem toward a multipolar direction.
For Amazon, this collaboration holds equal significance: if it can include OpenAI in the Trainium ecosystem, it will not only prove that its chips have top-tier AI training capabilities, but it may also attract other large model companies to follow, accelerating its crucial transition from "internal use" to "external commercialization."
In today's AI race, where computing power has become the core of national strength, the potential alliance between OpenAI and Amazon is not just a business decision for two companies, but could become a pivotal turning point in the decentralization of the global AI supply chain—the moat around NVIDIA is being subtly challenged by a second front.
