Bedrock Foundry Copilot and Claude Enterprise AI Fight Is On

In this blog post Bedrock Foundry Copilot and Claude Enterprise AI Fight Is On we will unpack what the major enterprise AI platforms are really fighting over, why it matters to Australian technology leaders, and how to choose a path without locking your business into an expensive mistake.

Most businesses are past the novelty stage with AI. Staff have tried chatbots, developers are using AI coding tools, and executives are asking why the business has not yet seen a clear return. The problem is not usually lack of interest. It is that AI has spread in small pockets, with different tools, unclear ownership, uncertain data controls and no simple way to measure whether it is helping.

At a high level, Amazon Bedrock, Microsoft Azure AI Foundry, Microsoft Copilot and Anthropic Claude are all trying to become the place where enterprise AI work happens. Think of them less as individual chatbots and more as control rooms for AI. They help organisations choose models, connect AI to company data, build agents, apply security rules and manage cost.

The technology behind this is a mix of large language models, agents, connectors and guardrails. A large language model is the AI engine that understands and generates text, code, summaries or analysis. An agent is an AI system that can follow steps, use tools and complete tasks. Connectors let AI read approved business information, such as SharePoint files, emails, CRM records or application data. Guardrails are the rules that stop AI from doing the wrong thing, such as exposing sensitive information or producing unsafe responses.

The fight is not really about the model

It is tempting to ask, “Which model is best?” That is the wrong starting point for most CIOs and CTOs.

The better question is: “Which platform gives us the safest and most cost-effective way to use the right model for each business task?”

That distinction matters. One model may be excellent for legal-style reasoning. Another may be stronger for software development. Another may be cheaper and good enough for summarising meeting notes. The winners will not be the platforms with one impressive model. They will be the platforms that let businesses use multiple models without losing control.

This is why Microsoft, AWS and Anthropic are all moving in the same direction. They want to be the layer that manages the AI work, not just the model that answers the prompt.

Where each platform fits in plain English

Amazon Bedrock

Amazon Bedrock is AWS’s managed AI platform. In plain English, it gives organisations access to multiple AI models through AWS, without having to build all the underlying infrastructure themselves.

Bedrock is attractive for organisations already deep in AWS. It offers model choice, knowledge bases for grounding AI in company information, agents for task automation, and guardrails for safety. For AWS-heavy businesses, the business outcome is speed: teams can build AI services close to the systems and data they already run.

The risk is complexity. If your business is mainly Microsoft 365 and Azure, Bedrock can still work well, but you need a clear architecture. Otherwise, you may create another island of AI governance that security and compliance teams struggle to monitor.

Azure AI Foundry

Azure AI Foundry is Microsoft’s platform for building custom AI apps and agents. It sits closer to the engineering and application development side of AI than Microsoft 365 Copilot.

For Microsoft-focused organisations, Foundry is important because it brings model choice into the Azure environment. That includes OpenAI models and, now, Claude models as well. For many Australian organisations already using Azure, Microsoft Entra ID for identity, Defender for security and Microsoft Purview for data governance, Foundry can reduce friction.

The business outcome is control. Instead of teams experimenting with standalone tools, Foundry gives IT a more governed place to build AI into internal systems, customer portals, reporting workflows or industry-specific applications.

Microsoft Copilot

Microsoft Copilot is the AI layer inside the tools many staff already use: Outlook, Teams, Word, Excel, PowerPoint, SharePoint and the browser. For non-technical employees, this is often the easiest entry point.

Copilot’s value is not that it is the most flexible development platform. Its value is that it sits where work already happens. A manager can summarise meetings, draft documents, search internal files and prepare updates without opening a separate AI system.

The business outcome is productivity, but only if your Microsoft 365 environment is tidy. If permissions are messy, old files are exposed to the wrong people, or SharePoint has become a dumping ground, Copilot can surface information you forgot was accessible. That is not an AI problem. It is an information governance problem exposed by AI.

Claude

Claude is Anthropic’s AI assistant and model family, known for strong reasoning, long-form analysis, coding support and agent-style work. Claude is now showing up in more enterprise settings, including Microsoft environments and developer workflows.

This matters because Claude is no longer just “another chatbot”. It is becoming part of enterprise AI architecture. Organisations can use Claude directly, through development platforms, or through environments such as Microsoft Foundry and Copilot depending on availability, licensing and regional settings.

The business outcome is choice. But choice also creates governance work. If Claude is available in Copilot, Foundry, GitHub and direct subscriptions, someone needs to decide where it is approved, what data it can access, who pays for it and how usage is monitored.

The real problem is AI sprawl

A common pattern we see is this: the board approves Microsoft 365 Copilot, the development team adopts GitHub Copilot, a data team trials Bedrock, a business unit buys Claude subscriptions, and finance sees four different AI cost lines with no clear owner.

Nothing here is malicious. It is normal early adoption behaviour. But left unchecked, it creates cost leakage, duplicated tools, inconsistent security settings and unclear accountability when something goes wrong.

For a 200-person company, this can quickly become expensive. Imagine 60 staff with Copilot licences, 20 developers using premium coding agents, a few workloads running in Foundry, and a separate team using Claude for research. Individually, each decision may be reasonable. Collectively, the business may be paying for overlapping capability without a clear return.

This is where technology leaders need to shift from “AI tool approval” to “AI platform strategy”.

What most companies get wrong

They start with licences instead of use cases

Buying AI licences for everyone feels decisive, but it can waste money. A better approach is to identify three to five high-value use cases first.

For example: reducing time spent on board reporting, speeding up tender responses, improving service desk knowledge retrieval, automating policy searches, or helping developers clear backlogs faster. Each use case should have a measurable outcome, such as hours saved, faster response time, lower support volume or fewer manual steps.

They forget data readiness

AI is only as useful as the data it is allowed to use. If your files are poorly labelled, permissions are too broad, or sensitive information is stored in random locations, AI will magnify the problem.

Before rolling out Copilot or building agents in Foundry or Bedrock, review who can access what. Microsoft Intune, which manages and secures company devices, Microsoft Defender, which helps protect against threats, and Microsoft Purview, which helps govern data, all become part of the AI readiness conversation.

They treat security as a final checklist

AI security needs to be designed in from the start. Essential 8, the Australian government’s cybersecurity framework that many organisations are now required or expected to follow, is a useful baseline. It helps reduce common risks such as unpatched systems, weak access controls and unmanaged devices.

But AI adds new questions. Can the AI access sensitive customer data? Can it send information to external tools? Can users paste personal information into public AI services? Are prompts and outputs logged? Who reviews failed or risky agent actions?

For cloud security, this is also where tools like Wiz, which helps identify risks across cloud environments, and Microsoft Defender become important. CloudProInc works across both Microsoft security and Wiz because AI systems are only as safe as the cloud and identity foundations underneath them.

A practical decision framework

Here is a simple way to think about the platform choices.

• If the work happens inside Microsoft 365: start with Microsoft Copilot, but fix permissions and data governance first.
• If you are building custom AI applications on Azure: look closely at Azure AI Foundry, especially if you want model choice with enterprise controls.
• If your workloads and data are mainly in AWS: Amazon Bedrock may be the natural platform, provided governance is clearly defined.
• If teams need deep reasoning, coding assistance or long-form analytical work: Claude may be valuable, either directly or through approved enterprise platforms.
• If developers are the main audience: review our related guide on choosing the right AI coding stack for tech teams.

The key is not to pick one vendor forever. The key is to define which platforms are approved for which jobs.

A real-world scenario

Consider a mid-sized professional services firm with 180 staff. The leadership team wants AI to improve productivity, but different departments have already started using different tools.

Marketing uses Claude for first drafts. Consultants use Copilot to summarise meetings. Developers use GitHub Copilot. Operations wants an AI agent to search policies and procedures. Finance is worried about rising software spend, while the CIO is worried about client data leaving approved systems.

The wrong response is to ban everything. The better response is to create an AI platform map.

In this scenario, Microsoft Copilot may be approved for general staff productivity. Foundry may be used for a secure internal policy assistant connected to approved SharePoint libraries. GitHub Copilot may be standardised for developers with repository controls. Claude may be approved only through enterprise-managed channels, not personal accounts.

The result is not just better technology. It is lower risk, cleaner costs and clearer accountability.

What Australian leaders should do next

For Australian organisations, AI governance should sit beside privacy, cybersecurity and operational risk. The Privacy Act still applies when personal information is used in AI workflows. OAIC guidance has also made it clear that businesses need to be careful with personal and sensitive information in publicly available AI tools.

That means your AI plan should include practical controls:

1. Create an approved AI tools list. Make it clear which platforms staff can use and for what purpose.
2. Review Microsoft 365 permissions. Copilot is much safer when SharePoint, Teams and OneDrive access is under control.
3. Classify your data. Decide what can and cannot be used with AI.
4. Track costs by department and use case. AI spend should connect to measurable business outcomes.
5. Test before scaling. Run pilots with success measures, not vague enthusiasm.
6. Build security into the platform. Use identity controls, device management, logging, threat detection and cloud risk scanning from day one.

If you have already read our post on standardising developer AI safely, the same principle applies more broadly: model choice is good, but unmanaged choice becomes risk.

The bottom line

Bedrock, Foundry, Copilot and Claude are not just competing for AI headlines. They are competing to become the operating layer for enterprise AI.

For CIOs and CTOs, the decision is not about chasing the newest model. It is about building a controlled AI environment where staff can move faster, sensitive data stays protected, costs are visible and compliance obligations are not treated as an afterthought.

CloudProInc is a Melbourne-based Microsoft Partner and Wiz Security Integrator with more than 20 years of enterprise IT experience across Azure, Microsoft 365, Intune, Windows 365, OpenAI, Claude, Defender and Wiz. We help organisations across Australia and internationally make practical AI decisions that stand up to security, cost and governance scrutiny.

If you are not sure whether your current AI setup is helping your business or quietly creating cost and risk, we are happy to take a look. No hard sell, no jargon — just a practical review of where you are and what to fix first.