Amazon invested 50 billion in OpenAI to build a stateful AI development environment. OpenAI and Amazon announced today a multi-year strategic partnership. As part of the cooperation, Amazon will invest US 50 billion in OpenAI (IT Home Note: The current exchange rate is approximately 342.618 billion yuan), with an initial investment of US 15 billion, and an additional US 35 billion in the next few months after certain conditions are met. The two parties will jointly develop a Stateful Runtime Environment based on the OpenAI model, which will be provided through Amazon Bedrock. Stateful developer environments are the next generation of cutting-edge model applications, enabling models to seamlessly access elements such as compute, memory, and identity. Such environments allow developers to retain context, remember previous work, collaborate across software tools and data sources, and call upon computing resources designed to handle ongoing projects and processes. These stateful developer environments will be trained to run optimally on AWS infrastructure and are deeply integrated with Amazon Bedrock AgentCore and infrastructure services to ensure that customers' AI applications and agents can be coordinated with other infrastructure applications running in AWS. The environment is expected to be officially launched in the coming months. According to reports, AWS will become the exclusive third-party cloud distribution provider for the OpenAI Frontier platform, a move that will expand access to OpenAI s most advanced enterprise platform as demand for AI deployment accelerates across industries. The Frontier platform enables organizations to build, deploy and manage teams of AI agents that run within real business systems with shared context, built-in governance and enterprise-grade security features without the need to manage the underlying infrastructure. In addition, OpenAI and AWS have further expanded their cooperation on the basis of the existing 38 billion multi-year agreement, adding an additional 10 billion over the next eight years. The expansion includes OpenAI s commitment to consume approximately 2 gigawatts of Trainium capacity via AWS infrastructure to support the needs of stateful runtime environments, Frontier and other advanced workloads. This protocol reduces the cost and improves efficiency of large-scale production of intelligence. Under this architecture, OpenAI can obtain long-term computing power guarantees, and at the same time cooperate with AWS to deploy specialized chips in its overall computing ecosystem, allowing enterprises to call intelligence on demand without having to manage the underlying infrastructure. This commitment covers Trainium3 and next-generation Trainium4 chips. Amazon said that Trainium4, expected to be delivered in 2027, will bring another major performance improvement, including significantly improved FP4 computing power, greater memory bandwidth, and high-bandwidth memory capacity. Finally, OpenAI will also work with Amazon to develop custom models that Amazon developers can use to power Amazon s consumer-facing applications. Amazon teams can customize OpenAI models for AI products and agents that directly serve customers