Dive Brief:
- Deloitte is launching its Open Model Engineering practice, a service to assist clients in using open-source models in their portfolios as enterprises build an increasingly mixed tech stack, the company announced Wednesday.
- The practice, part of Deloitte’s AI services portfolio, will hire, train and certify forward deployed engineers through 2027 to grow the practice for companies that are developing enterprise AI applications and products built on open-source frameworks and models. Deloitte will roll out the unit in North America, Europe and the Asia Pacific region.
- Open models are gaining momentum within the enterprise AI landscape and can be complementary to proprietary AI platforms, Nitin Mittal, global AI leader and principal at Deloitte, said in the announcement. “By helping organizations take a mixed-model approach to building, deploying, and scaling multi-agent systems, we can help optimize technical choices while preserving flexibility across their AI stack and maximizing return on investment,” Mittal said.
Dive Insight:
More enterprises are leaning on a mix of proprietary and open source models while building their AI tech stack, and the vendors that serve them are aiming to meet those needs.
Just before Labor Day, chip maker Nvidia confirmed its acquisition of open-source AI platform Hugging Face for nearly $13 billion, underscoring the growing appeal of open models.
Deloitte’s open source practice taps into an industry move to forward deployed engineers — embedded, specialized developers who help speed up enterprise adoption. The company follows Palantir, AWS, Microsoft and others that have introduced this model to deliver AI and condense deployment timelines.
Open models are joining rather than replacing proprietary AI tools in the enterprise AI stack as part of a deliberate hybrid approach, Jim Rowan, U.S. head of AI and principal of Deloitte Consulting, told CIO Dive in an email. A mixed approach gives organizations more control over cost, data and where inference happens, he said.
“We see the future as a portfolio approach — frontier models handling complex reasoning and orchestration, open models handling specialized execution,” Rowan said.
The Open Model Engineering practice exists to help enterprises architect that mix, which can be a complex undertaking, Rowan said.
Deloitte said in its announcement that enterprises have to navigate selecting the right model and architecture for each workload, predicting cost economics as agentic AI usage scales, finding sovereignty over where and how models are deployed and maintaining control over enterprise data, intellectual property and model behavior.
The Open Model Engineering practice will initially focus on developing enterprise AI applications built on Nvidia’s Nemotron models. Through the practice, enterprises will be able to make model and architecture choices between proprietary and open models, agentic platforms, cloud services and on-premises infrastructure.
