OpenAI models now available on Amazon Web Services OpenAI模型现已在亚马逊网络服务上可用

OpenAI just turned its cloud infrastructure rival into its global distribution partner, and the implications of that tectonic shift are far more interesting than the press release lets on. The immediate story is transactional: OpenAI’s most powerful models—GPT-5.5, the reasoning-focused GPT-5.4, and its code-generation workhorse Codex—are now natively available on Amazon Web Services via the Bedrock platform. Prices match OpenAI’s own API, existing AWS contracts apply, and it all runs in both commercial and government-standard regions. But let’s skip the bullet points and talk about what this actually means.

This isn’t about customer convenience; it’s about a fundamental realignment of power in the AI value chain. For years, Microsoft was the crown jewel in OpenAI’s hardware and distribution strategy. Azure wasn’t just a cloud provider; it was the exclusive launchpad for OpenAI’s commercial ambitions. That exclusivity is now officially over. By placing its crown jewels on AWS—the world’s largest cloud platform by market share—OpenAI is hedging its bets, diversifying its revenue streams, and sending a very clear message to Microsoft: our partnership is not a marriage. It’s a strategic alliance, and we have other options.

From Amazon’s perspective, this is a masterstroke of competitive arbitrage. For too long, AWS has been playing catch-up in the generative AI race, not in raw capability—its own Bedrock-hosted models from Anthropic and others are formidable—but in perceived cultural relevance and developer mindshare. They couldn’t build a "GPT killer" fast enough. So, they did the next best thing: they bought the right to host the original. They’ve essentially said, "If you can’t beat the model, commoditize the cloud it runs on." This immediately neutralizes Azure’s key differentiator for enterprise clients. Why make a complex migration to Azure just for GPT access when your existing AWS contract now covers it? It’s a brilliant defensive move that protects AWS’s core compute business from erosion.

But here’s the critical, under-discussed angle: the US-only limitation. This isn’t just a technical or regulatory footnote; it’s the most revealing part of the deal. It screams geopolitical caution. OpenAI is tiptoeing around international data sovereignty and export control sensitivities, especially concerning government region access. By restricting this to US-based AWS regions for now, they’re creating a clean, legally defensible firewall. It allows them to service the lucrative US federal and state government contracts—a market where Azure’s partnership with OpenAI had an early foothold—while avoiding the labyrinth of global compliance nightmares for the moment. It’s a calculated, conservative move that prioritizes scaling in the largest, most legally predictable market before wrestling with the world.

This also exposes a fascinating tension in the AI industry’s "ecosystem" play. We’ve been sold a story of integrated stacks: your AI, your cloud, your hardware, all in a synergistic loop. This deal blows that up. It proves that frontier AI models are becoming portable commodities. The value is shifting from the model itself to the orchestration layer—the platforms, tools, and contracts that manage its deployment. Bedrock is winning not because it has the best homegrown model, but because it’s becoming the neutral marketplace. It’s the Amazon of AI models, and that’s a far more formidable long-term position than simply having one exclusive, high-priced tenant.

What this really signals is the beginning of the end of the "exclusive partnership" era in frontier AI. The models are too valuable and the cloud markets too vast for any single pairing to hold. Expect Google Cloud to be aggressively courting a similar deal next. The future isn’t a handful of walled gardens; it’s a fierce, brokered competition where model providers play cloud giants against each other for distribution, and enterprises can pick their AI from a menu without switching their infrastructure. OpenAI just accelerated that future, securing its independence while making AWS the new center of gravity. The cloud wars just became AI wars, and the most valuable weapon isn’t the algorithm—it’s the distribution channel.

当OpenAI宣布GPT-5.5、Codex这些顶尖模型能直接在AWS Bedrock上使用，并且定价还跟自己家一模一样时，这感觉就像一个原本在市中心开了家独家旗舰店的奢侈品牌，突然宣布在隔壁最大的百货商场也开了专柜，价格一分不差。表面看是渠道扩张，底下涌动的全是算计和妥协。

这事儿的核心从来不是“用户多了一个选择”，而是“OpenAI的算盘打得比你我想象的更响”。和微软的Azure深度绑定，一度让OpenAI看起来像是微软的“独家军火商”。现在，转身就在AWS这个“老对手”的平台上架，像极了商业世界里那句老话：没有永远的朋友，只有永远的利益。OpenAI显然不满足于只做微软云生态的附属品，它需要更大的舞台，需要接触AWS上那庞大的、从初创公司到政府机构的客户群。所谓的“相同定价”，与其说是对AWS用户的优待，不如说是OpenAI为了快速抢占新渠道，暂时牺牲了渠道利润的妥协。它赌的是规模效应，是模型使用量的绝对增长，是将自己的标准更深地植入企业基础设施。

对AWS来说，这简直是一场尴尬又必要的胜利。长久以来，在高端AI模型服务上，Azure凭借与OpenAI的合作，多少压了AWS一头。如今，AWS终于把最强对手的武器拿到了自己柜台上，这不仅能立刻提升其AI服务的吸引力，更是一种战略止血。客户不需要为了用最好的模型而迁移到Azure，他们可以在熟悉的AWS环境里继续消费。AWS提供的“商业和政府区域”更是直指痛点——那些对数据合规、本地化运行要求严苛的企业和机构，现在似乎有了一个“既用顶尖模型，又守数据规矩”的路径。当然，这个“仅限美国”的条款，又像是一条小心翼翼画出的线，透露出地缘政治和数据主权在AI落地过程中挥之不去的阴影。

最值得玩味的是“使用量计入现有AWS合同”这一条。这简直是企业IT采购部门最爱听的旋律。它意味着客户无需单独为AI开立新合同、新预算，可以直接动用已有的、可能庞大的AWS年度承诺。这极大地降低了企业尝试或大规模采用AI模型的决策摩擦和财务门槛。AWS用其最擅长的“企业级服务”方式，为OpenAI模型铺好了通往企业财务系统的最后一公里。这不仅仅是技术集成，更是财务流程的集成，后者往往才是决定新技术能否在企业内铺开的关键。

但这片看似共赢的景象下，暗流同样汹涌。价格相同，意味着利润空间完全一致，OpenAI让渡了什么？除了数据和潜在的长期使用承诺，可能还有对其模型在非优化硬件上运行时性能表现的某种默许。AWS的基础设施并非完全为OpenAI架构设计，这里会不会出现性能差异或优化滞后？对于追求极致延迟和吞吐的尖端应用，这可能成为新的比较维度。同时，Codex这种代码模型在AWS的开放，将直接加剧与AWS自家AI编程工具以及GitHub Copilot（背后也有OpenAI技术）的微妙竞争。生态内部的左右互搏将变得更加明显。

最终，这场合作最辛辣的注脚可能是：曾经标榜“开放”与“非营利”精神的OpenAI，其商业路径正变得越来越像一个传统的、精明的、无处不在的平台巨头。它不再仅仅是一个模型提供商，它正通过AWS、Azure这些庞大的云渠道，把自己变成一种“AI基础设施”，像水电一样渗透进每一个企业IT预算里。而AWS，这个曾经的云计算先驱，在AI浪潮中一度显得笨重，现在也终于学会了用“拿来主义”快速补齐短板。巨头之间的合纵连横，最终目的都是巩固自己的围墙。对于开发者和企业而言，这无疑增加了选择和灵活性，但也意味着，你们与顶尖AI能力之间的连接，将更加牢固地建立在这些巨头既合作又竞争的商业协议之上。自由选择的背后，是另一重更庞大、更复杂的商业依附。