AWS will offer OpenAI models following the end of Microsoft's exclusivity, as OpenAI falls short of revenue expectations and confronts $100 billion in infrastructure obligations.

      TL;DRAWS plans to sell OpenAI models following Microsoft's termination of its exclusive reselling rights, marking the completion of a restructuring initiated by Amazon's $50 billion investment in OpenAI's $110 billion funding round. The two companies have collaborated to develop a Stateful Runtime Environment for agentic AI on Bedrock. However, this development comes amid reports from the WSJ indicating that OpenAI has not met its revenue and user targets, facing a projected cash burn of $25 billion against a $30 billion revenue expectation, alongside significant infrastructure commitments to AWS, Azure, and Oracle that rely on growth OpenAI has yet to demonstrate.

      Amazon Web Services (AWS) announced on Tuesday that it will start selling OpenAI's models to its cloud customers, one day after Microsoft ended its exclusive reselling agreement that had granted Azure unique access to OpenAI's technology for the initial three years of the generative AI era. Matt Garman, AWS's CEO, shared with Bloomberg Television that this move addresses a long-standing customer demand. Starting Tuesday, select new OpenAI models will be available for preview on AWS, with the most robust GPT models expected within weeks. This announcement finalizes a restructuring of a significant partnership in the AI sector that began in February when Amazon pledged up to $50 billion as part of OpenAI's funding round, which valued the ChatGPT creator at $852 billion and represented Amazon's largest investment in a single company. Additionally, OpenAI committed to spending $100 billion on AWS computing resources and Trainium chips over eight years, utilizing two gigawatts of capacity. The central inquiry of this deal is not whether OpenAI's models are capable of being sold on competing clouds, but rather if OpenAI can sell enough of them anywhere to justify its promised expenditure.

      The restructuring involved Microsoft and OpenAI modifying their partnership agreement, which dismantles the exclusive arrangement central to the initial phase of the AI boom. Microsoft’s license to OpenAI's intellectual property for models and products has transitioned from exclusive to non-exclusive, extending through 2032. Microsoft retains a 20 percent share of OpenAI's revenue until 2030, now capped. OpenAI is still required to deploy its models on Azure first unless Microsoft cannot or opts not to support the necessary capabilities, maintaining Azure’s early access benefits while allowing AWS and Google Cloud customers to utilize OpenAI’s technology via their existing cloud services. Additionally, the restructuring eliminates the AGI clause, which could have revoked Microsoft’s commercial rights had OpenAI’s board decided the company achieved artificial general intelligence. This clause had become more of a governance hindrance than a practical safeguard, and its removal indicates both companies recognize the nature of their relationship as commercial rather than existential.

      Practically, this means OpenAI can now serve customers across any cloud platform. For AWS, which spent the formative years of the AI boom curating superior alternatives to its Bedrock model marketplace—including models from Anthropic, Meta, and Mistral—gaining access to OpenAI's models mitigates what was Azure’s most significant competitive edge. Amazon's additional investment of up to $25 billion in Anthropic, announced alongside the OpenAI deal, ensures AWS now holds exclusive or preferred access to the two most commercially relevant AI model families in the enterprise sector. The productivity company Superhuman had sought AI services from Microsoft specifically because Azure was the sole cloud providing OpenAI's models. That scenario has now changed, allowing AWS customers to utilize OpenAI within their existing infrastructure, retaining their current security configurations and compliance certifications.

      The two companies have jointly created a Stateful Runtime Environment powered by OpenAI’s models, offered through Amazon Bedrock, which facilitates persistent orchestration and memory for AI agents executing complex, multi-step workflows. Rather than relying on disjointed API calls, the runtime retains context throughout various steps, maintaining memory, tool states, environment configurations, and identity boundaries. Amazon introduced this as Amazon Bedrock Managed Agents during an event in San Francisco, where AWS also revealed a more extensive push into business applications. OpenAI’s Chief Revenue Officer, Denise Dresser, remarked that business clients desire these models within a familiar, reliable environment. The Stateful Runtime Environment operates within AWS infrastructure and is integrated with Amazon Bedrock AgentCore, meaning its agentic capabilities are not merely an extra feature grafted onto existing infrastructure but a co-developed layer that embeds OpenAI's models within AWS operationally.

      This joint product illustrates a broader transition in the sale of AI models. The first wave of the AI boom focused on model access: which cloud provided the best models. The second wave emphasizes model integration: which cloud can make those models most functional within existing enterprise workflows. Microsoft has begun introducing its internal AI models through a team led by Mustafa Suleyman, indicating a parallel strategy aimed at positioning its offerings amidst OpenAI's growing presence. Meanwhile, Google Cloud has responded with a $750 million partner fund for agentic AI deployments at Cloud Next 2026, investing through consulting partners to secure enterprise