Most of the AI conversations we have with our customers start with the same question: Which model should we use? Giving a single answer to this question is becoming increasingly difficult, because organizations' needs are not one-size-fits-all. One team wants fast access to internal documents, another wants to accelerate software development, and yet another wants to analyze sensitive data without it ever leaving the organization.

That's why we ask the question a little differently: Which job should run on which model, and where? This is exactly what we mean by a model-agnostic approach: enabling an organization to choose the most suitable model and infrastructure for each workload, without tying its AI investment to a single provider. To do this, we bring three pieces together: our Google Cloud, Microsoft and Anthropic partnerships, the GPU infrastructure in our own data center, and our consulting approach that combines them according to each organization's needs.

Why Three Different Partnerships?

At Doğuş Teknoloji, we have long-standing partnerships with Google Cloud and Microsoft. This year, we added Anthropic to our partner ecosystem and became a Certified-level partner in the Claude Partner Network. This level was achieved through our certified teams working with Claude and the projects we have delivered. As a result, we bring organization-specific projects to life for our customers together with our certified teams, whose expertise has been validated by Anthropic.

Each of our partners' models is stronger in a different layer of enterprise AI. Rather than pitching a single product to our customers, we create value by bringing this ecosystem together in the right way for their needs. We also use these models across different workloads in our own operations; we experience the setups we recommend to our customers ourselves first.

Three Ecosystems, Different Strengths

Google Cloud and Gemini

On the Google Cloud side, Gemini is offered both through the Gemini Enterprise app used by employees and through Gemini Enterprise Agent Platform, where developer teams build enterprise agents. What sets the platform apart on the enterprise side is the control structures around the model. Where data is stored and where it is processed can be defined separately, and security layers such as network isolation, encryption key management and detailed audit logs can be enabled.

We especially recommend Gemini for data-intensive scenarios. Chief among these are analytics applications working with enterprise data on BigQuery, search across large document archives, and agents added to applications running on Google Cloud.

Gemini Enterprise also brings together, under one roof, applications such as Gemini Notebook Enterprise, where teams work on their own documents; ready-made agents like Deep Research; and Agent Designer, which enables building agents without writing code.

Microsoft and Microsoft Copilot

Copilot's biggest advantage is that it is built into Outlook, Teams, Word and Excel, the tools employees already use every day. More importantly, the organization's existing access permissions and information security rules apply as they are.

Copilot Studio is used for organizations to build their own agents, while Microsoft Foundry is used for projects that require deeper customization. We see Copilot as an efficient option for accessing internal knowledge, preparing documents, tracking meetings, and self-service processes such as HR or IT support.

Today, Copilot is not tied to a single model either. Alongside OpenAI models, Anthropic's Claude models can also be used in Copilot Chat, in Word, Excel and PowerPoint, and in Copilot Studio. Whichever model is selected, the organization's existing access permissions and security rules continue to apply in the same way.

Anthropic and Claude

Claude can be integrated into an organization's own systems without being tied to a specific office application. Thanks to the Model Context Protocol, it can connect to systems such as CRM, ERP and databases through a single standard, so there is no need to write every integration from scratch.

How Claude is procured also depends on how it will be used. If Claude will be integrated into the organization's own applications via API, it can be purchased through the Google Cloud and Microsoft cloud marketplaces. This way, it can be managed together with the organization's existing cloud investment and billing structure. If features of the Claude app such as Cowork will be used, we recommend obtaining an enterprise license directly from Anthropic.

We position Claude primarily for software development and application modernization, analysis of long and complex documents, and automation of multi-step business processes. With our Certified-level partnership, we provide our customers with stronger integration and consulting support in these areas.

How Do They Work Together?

In practice, most organizations don't move forward with a single model. Employees' day-to-day work can be designed with Copilot inside office applications, data- and analytics-focused applications with Gemini on Google Cloud, and work requiring software development and in-depth analysis with Claude.

In our AI strategy consulting, we work with our customers to identify use cases, assess which use case will deliver better results with which model, and stand alongside their teams throughout the integration process.

When Data Must Not Leave: Local Models on GPU Servers

For some workloads, the cloud option is never on the table. Especially in the finance, healthcare and public sectors, certain data must not leave the organization. In this case, the solution is to run large language models on GPU servers under the organization's control.

The significant progress of open-weight models in recent times has made this possible. Today, for many tasks, a model that runs in-house and never sends data outside can deliver sufficient performance.

The real challenge in this scenario, however, is not software but hardware. GPU procurement, installation, power and cooling are not an easy investment for most organizations to take on. At Doğuş Teknoloji, we made this investment in our own data center. By hosting GPU servers in our own infrastructure, we enable our customers to run local models without purchasing hardware.

This way, sensitive workloads run in this environment while other work can proceed in the cloud with Google Cloud, Microsoft and Anthropic solutions. In our view, a true hybrid architecture means every workload running in the environment best suited to it.

The right choice in AI is not picking the most popular model; it is being able to choose the model and infrastructure that fit the needs of the job. With our Google Cloud, Microsoft and Anthropic partnerships and our own data center infrastructure, we give our customers the flexibility to make that choice. From strategy to use cases, from integration to deployment, we will continue to stand by them at every step.