There is a quiet crisis running through most large organizations right now. People are using AI tools that their companies have never approved. One estimate puts the number at somewhere between 40 and 65 percent of enterprise employees. More than half of those people are feeding sensitive company data into tools that have never been through a security review. Most of them do not think they are doing anything wrong.
This has happened because getting an AI tool formally approved takes months, and the tools keep getting better every few weeks. By the time a company finishes evaluating something, there is already something newer. So people just use what works and pay with a personal card.
Anthropicʼs new arrangement with AWS is a direct response to this problem. By making Claude available through an existing AWS account, with no separate contract or separate vendor relationship, they have collapsed the procurement barrier almost entirely. If your company already uses AWS, and the majority of large organizations do, the path to deploying Claude through an approved, audited, controlled environment is now very short.
The deeper story here is that Amazon has spent over 33 billion dollars investing in Anthropic across several rounds, with the most recent agreement committing Anthropic to spend more than 100 billion dollars on AWS infrastructure over the next decade. That is not a vendor relationship. That is a structural merger of two companies that happen to remain legally separate. Anthropic has also named AWS as its primary cloud and training provider. The Claude Platform on AWS is essentially the commercial product that this deep infrastructure bet was always pointing toward.
For business operators, the practical question is which version of Claude on AWS to actually use. There are now two distinct options, and the difference matters. The existing route, called Amazon Bedrock, keeps your data inside AWSʼs own infrastructure. AWS is the data processor. Your data does not go to Anthropic. This is the option for companies in regulated industries, for European businesses navigating GDPR, and for anyone whose legal or compliance team has drawn a hard line on where data can be processed.
The new Claude Platform on AWS works differently. AWS handles your login credentials, your audit logs, and your invoice. But the actual data, your prompts and responses, is processed by Anthropic outside the AWS boundary. This makes it unsuitable for companies with strict regional data rules, but it gives developers and teams access to Anthropicʼs newest capabilities on the same day they launch, something the Bedrock route does not always guarantee.
The other thing worth watching is the competitive pressure this creates. OpenAI models also became available on AWS Bedrock in late April 2026, within 24 hours of Microsoft removing an exclusivity clause that had previously blocked it. AWS is now the infrastructure layer for both of the most widely used AI platforms. This is less a story about AI and more a story about cloud consolidation. The companies that control the cloud infrastructure are positioning themselves as the mandatory gateway for AI adoption in the enterprise, and they are moving fast to lock in that position.
For procurement managers, operations directors, and anyone dealing with rogue AI use inside their organization, the relevant takeaway is this: the main argument for not formalizing AI access, that it is too complicated or requires a new vendor process, has become significantly weaker. The tools are now inside the infrastructure most companies already pay for.