With the shift from AI experimentation to the integration of agentic AI in day-to-day business operations, enterprises are faced with an issue they can no longer sidestep: what degree of control should they exercise over the models at the heart of their systems?
Deloitte has its answer. The firm is making a play for open models to ensure agentic AI is as cost-effective and enterprise-ready as it is flexible. To that end, Deloitte has introduced a new global Open Model Engineering practice, with a presence in India and other regions across North America, Europe and Asia-Pacific.
The objective is to give businesses the means to construct and scale agentic applications through a mix of proprietary and open-source offerings.
Why Enterprises Want More Control Over AI Models
For many, model selection is not a matter of raw performance alone. When AI agents are put to work, one must also factor in vendor reliance, regulatory hurdles, data sovereignty and the bottom line. Deloitte’s practice is founded on four tenets: control of IP and data, flexibility, cost predictability and sovereignty.
Open Models Put Cost, Data and Deployment Back in Enterprise Hands
This allows an organisation to pick the architecture that suits a given workload rather than being locked into a single provider. There is more oversight of inference and token economics, and the freedom to run workloads in the cloud, on-premise or in a hybrid setup. Such an approach is well suited to those with sensitive information or applications that have to conform to local languages and rules.
Deloitte’s AI Stack Brings Open Models Into the Enterprise
In practice, Deloitte will be working with clients to assess everything from cloud services and on-premise infrastructure to agentic platforms, all with an eye on improving AI ROI. The initial toolkit will include NVIDIA Nemotron open models and NIM microservices alongside Deloitte’s Zora AI™ platform. The intent is to let companies govern and optimise their AI agents in-house instead of depending on opaque, hosted APIs.
“NVIDIA’s collaboration affords developers the ability to deploy agentic AI securely while retaining command of their enterprise data,” says Kari Ann Briski of NVIDIA.
Deloitte’s Hybrid AI Strategy Goes Beyond Open Models
Yet this is not an open-only proposition. Deloitte sees value in a hybrid strategy that lets workloads be shifted as the business or regulatory landscape dictates.
Sathish Gopalaiah, President of Consulting at Deloitte South Asia, puts it this way: open models let an enterprise mould AI to its own processes and data, as opposed to contorting the business to fit a model.
Building the Talent to Take Agentic AI Into Production
To back this up, the firm will be putting forward-deployed engineers with clients around the world to put these solutions in place, with hiring and certification planned through fiscal 2027. The message for the enterprise is plain. As agentic AI goes to production, the ability to dictate terms on cost, deployment and data will be every bit as vital as the technology itself.
