In the 1990s, Microsoft bundled Internet Explorer with Windows, made it free, and cut off the oxygen for Netscape Navigator — a product that was, at the time, better and more popular. The Department of Justice sued. The case centered on whether Microsoft was using its operating system monopoly to dominate the browser market by bundling. The district court ordered a breakup. The appeals court upheld the monopolization finding but reversed the breakup, and the Bush administration settled on behavioral remedies. The result: Microsoft was required to share APIs with third parties and was restricted from certain bundling practices. The practical effect: the unbundling created space for competitors — Firefox, then Chrome — to thrive. Netscape was already dead. The remedy didn't save it. It saved the market. In 2024, the same DOJ won a landmark ruling that Google maintained an illegal monopoly in search and search text advertising. The remedies phase focused on similar unbundling questions: exclusive distribution agreements, default placements, and the bundling of Chrome with Google Search. The pattern holds: a dominant platform uses its position in one market to foreclose competition in an adjacent market through bundling and exclusivity.
The current cloud bundling dynamic is structurally similar. Microsoft bundles OpenAI's models with Azure — you can get GPT-4.5 through Azure OpenAI Service, integrated with Azure's enterprise billing, IAM, and compliance framework. You cannot get it through Google Cloud. You can get it through OpenAI directly, but without the Azure integration and enterprise contract terms. The bundling is not exclusive in a contractual sense — OpenAI is an independent company — but it is exclusive in a practical sense for enterprises that want their AI infrastructure, their model access, and their existing workloads on the same cloud. That is most enterprises.
Amazon's Anthropic partnership is a mirror image. AWS invested $4 billion in Anthropic — on top of an initial $4 billion investment, bringing the total to $8 billion — makes Claude available through Amazon Bedrock, and integrates Claude with AWS's data and analytics services. Microsoft has invested a reported $13 billion in OpenAI. Claude is available through Google Cloud as well — Google invested $2 billion in Anthropic — but the AWS integration is deeper, the Bedrock API is native to the AWS ecosystem, and the enterprise procurement path of least resistance is to buy AI from your existing cloud provider.
The FTC and the European Commission are both investigating these arrangements. In January 2024, the FTC launched an inquiry into generative AI investments and partnerships, ordering Alphabet, Amazon, Anthropic, Microsoft, and OpenAI to provide information on their investment structures and competitive dynamics. In 2026, the FTC broadened its Microsoft probe to cloud bundling, Entra ID, and Copilot AI, questioning at least six rivals. The questions they are asking are the same questions the DOJ asked in the browser case. Are cloud providers using their infrastructure dominance to preference their own or their partners' AI services? Are bundling practices foreclosing competition from independent AI model providers? Are enterprise customers being steered toward bundled AI solutions through pricing, integration, and the sheer friction of multi-cloud procurement?
The parallel is not perfect. The browser case was about a horizontally dominant product (Windows) being used to dominate an adjacent market (browsers) through bundling. The cloud case is about vertical integration — the infrastructure layer and the AI layer — with exclusive or preferential partnerships. The legal theory is different. The competitive effect is the same: bundling makes it harder for independent competitors to reach customers.
The remedy that antitrust scholars are discussing is not a breakup. It's unbundling: requiring cloud providers to offer AI models on equivalent terms to competitors, prohibiting exclusive or preferential partnerships, and potentially requiring interoperability standards so that enterprises can use any model on any cloud without being locked into a specific provider's AI ecosystem. This would be the functional equivalent of the browser remedy — behavioral constraints, not structural separation.
If unbundling remedies materialize, the winners are: independent AI model providers who get equal access to cloud distribution, multi-cloud infrastructure platforms that benefit from reduced lock-in, and enterprises that can negotiate AI pricing across providers. The losers: the hyperscalers whose AI revenue expectations are predicated on bundling as a competitive advantage, and the AI startups whose valuations include a premium for their exclusivity relationship with a cloud partner.
The timeline is years, not quarters. Antitrust investigations take 2-4 years as a baseline. Cloud AI bundling complaints are in the early stages — FTC information requests, European Commission consultations, academic papers, think-tank white papers. The investigations haven't resulted in formal proceedings. They will. The precedent is not the browser case. The pattern is. And the pattern says: bundling that forecloses competition eventually gets unbundled. The only question is whether the remedy arrives in time for the market it's meant to protect.