在人工智慧热潮的最初三年里，许多云端客户因 Microsoft 持有独家销售权而转向 Microsoft，寻求最新的 OpenAI 产品。Amazon 最主要的云端竞争对手在宣布放弃该独家权后隔天，Amazon 表示终于可以向自家客户开放 OpenAI 模型。AWS 将在星期二先提供部分新模型供预览，最强大的 GPT 模型则预计在接下来几周内推出，AWS 执行长 Matt Garman 说客户早就一直要求这件事。
Microsoft 的领先来自于早期对 OpenAI 的大量投入，让 Azure 取得旗舰模型的独家发布权；Amazon 则在其 Bedrock 市场中布局其他来源，如 Anthropic 与 Meta。即使如此，部分长期 AWS 用户，包括前 Grammarly 的 Superhuman，仍转往 Microsoft 取得 AI 服务。Amazon 已投入 50 billion（50×10^9）美元于 OpenAI，而 OpenAI 表示将再追加 100 billion（100×10^9）美元用于 AWS 的运算与晶片支出，比例为 1:2。在旧金山的一场活动中，AWS 与 OpenAI 推出 Bedrock Managed Agents（用于具上下文理解与记忆的自主代理），并同步宣布更大规模推进商务软体。
在承诺压力背景下，此宣布出现于 OpenAI 成长展望受质疑之际：其必须把使用量扩张到足以满足 Amazon、Microsoft 及 Oracle 等基础设施伙伴。华尔街日报报导其未达到多项内部目标，OpenAI 回应称其在“firing on all cylinders”。OpenAI 营收长官 Denise Dresser 指出，企业客户要在熟悉且可靠的基础设施下运作。Garman 补充，AI 需求仍高于可供应算力，OpenAI 将在今年及未来两年持续要更多容量，因此 AWS 正在全球扩充运算力、晶片与记忆体。
During the first three years of the AI boom, many cloud customers moved from AWS to Microsoft to access the latest OpenAI releases because Microsoft held exclusive selling rights. The day after Amazon’s largest rival announced ending that exclusivity, Amazon said it could finally make OpenAI models available to its own customers. AWS will open selected new models for preview on Tuesday, with the most powerful GPT models expected in the next few weeks, after AWS chief executive officer Matt Garman said clients had long requested this.
Microsoft’s lead came from an early, heavy OpenAI commitment that gave Azure exclusive rights to flagship models, while Amazon filled Bedrock with alternatives such as Anthropic and Meta. Even so, some longtime AWS users, including Superhuman (formerly Grammarly), shifted to Microsoft for AI services. Amazon has now invested $50 billion (50×10^9 USD) in OpenAI, and OpenAI said it would add another $100 billion (100×10^9 USD) in AWS compute and chip spending, a 1:2 ratio. At a San Francisco event, AWS and OpenAI introduced Bedrock Managed Agents, designed for context-aware and memory-based autonomous agents, and announced a broader push into business software.
The announcement comes as OpenAI’s growth outlook faces pressure: it must expand usage fast enough to meet commitments to infrastructure partners Amazon, Microsoft, and Oracle. The Wall Street Journal reported that OpenAI missed several internal targets, and OpenAI replied that it was “firing on all cylinders.” OpenAI revenue officer Denise Dresser said enterprise customers want trusted, familiar infrastructure. Garman added that AI demand still exceeds available compute, saying OpenAI wants more capacity this year and in the next two years, so AWS is adding power, chips, and memory globally.